I.
I got to talk to some AI researchers last week, and they emphasized how surprised everyone had been by recent progress in the field. They all agreed on why they were surprised: the “AI winters”, two (or so) past episodes when AI hype got out of control and led to an embarrassing failure to meet expectations. Eventually everyone learned the heuristic “AI progress will never be as fast as people expect”. Then AI progress went faster than expected, and everyone using the old heuristic was caught flat-footed, denying the evidence of their own eyes.
Is this surprising? It’s hard (and possibly meaningless) to segment the history of AI into distinct “eras”, but let’s try it just for fun: suppose that there were two past eras, both of which went worse than expected. If there are equal chances of an era meeting, exceeding, or missing expectations, then there’s a 22% chance that we either get two consecutive booms or two consecutive busts by pure coincidence. If we form a heuristic around this (“it’s always boom” or “it’s always bust”), then we’re interpreting noise and the future is likely to surprise us.
A quick and dirty Bayesian calculation: imagine three models. In Model A, researchers are biased towards optimism: 80% of the time, they will predict greater success than they actually attain, 10% of the time they will get it exactly right, and 10% of the time they will undershoot. In Model B, researchers are biased towards pessimism to the same degree. In Model C, researchers are unbiased and will overshoot, undershoot, and hit expectations with equal probability. Suppose we start with a 50% prior on Model C, and equal 25% probabilities for A and B. After observing one era of inflated expectations, we should have 52% chance A, 6% chance B, and 42% chance C. After observing two such eras, we should think 74% A, 1% B, and 25% C. Adding up the chances of all of the models, there’s a 67% chance that the next era will also be one of inflated expectations, but there’s a 33% chance it won’t be.
This is all completely made up, plus my math is probably wrong. My point is that these kinds of “heuristics” gleaned from n = 2 data points are a lot less interesting than you would think. Getting fooled twice in the same way probably feels pretty convincing, and I can’t blame the people involved for wanting to take a hard line against ever falling for it again. But their confidence that they’re right should be pretty low.
II.
Thinking about this reminded me of an article from The Week, November 2012:
Romney genuinely believed that he would become the nation’s 45th president, and was “shellshocked” by his landslide loss. “I don’t think there was one person who saw this coming,” one senior adviser told Jan Crawford at CBS News. Why was Team Romney so certain of victory? They simply did not believe that younger voters and minorities would turn out the way they did in 2008. “As a result,” says Crawford, “they believed that the public/media polls were skewed” in Obama’s favor, and rejiggered them to show Romney with “turnout levels more favorable to Romney.” In essence, Romney “unskewed” the polls, mirroring widely mocked moves by conservatives to show their candidate with a lead, epitomized by the now-infamous website UnskewedPolls.com. Romney’s defenders say he had plausible reasons to believe Obama’s turnout would be lower; less charitable commentators say Romney and his aides were stuck in a conservative media echo chamber at odds with reality.
Mitt Romney lost in exactly the way all the polls had predicted he would lose, but he wasn’t expecting it because he had cheerfully constructed a story of decreased minority turnout which no real poll supported. This story became a kind of King Canute style warning of the folly of Man – just accept the fricking polls, don’t come up with some private narrative about how decreased turnout will show up on a white horse and save you at the last second.
But we all know what happened in 2016. In retrospect, the fact that decreased minority turnout didn’t happen in one election, with the most popular-among-minorities candidate of all time, shouldn’t have been enough to form a strong heuristic that it would never happen at all.
This is even worse than the story above, because it’s n = 1. I wonder if part of it is the degree to which Romney’s loss formed a useful moral parable – the story of the arrogant fool who said that all the evidence against him was wrong, but got his comeuppance. Well, this last election taught us that arrogant fools don’t get their comeuppance as consistently as we would like.
III.
Speaking of the 2016 election, I feel the same way about this explanation of Hillary’s loss. It spins a narrative where the Hillary campaign management put all of their trust in flashy Big Data and ignored the grizzled campaign specialists who had boots on the ground, as if this was a moral lesson we should all take to heart.
But Moneyball makes the opposite argument. There, managers boldly decided to trust in statistics instead of just listening to the “intuitions” and “conventional wisdom” of professed experts, and they trounced the grizzled people with their ground-boots.
Anyone who learned the obvious lesson from Moneyball (“Hard math can defeat fallible human intuitions) would fail at the 2016 campaign, and anyone who learned the obvious lesson from the 2016 campaign (“Real experience and domain knowledge beat overeducated Big Data hotshots every time”) would fail at the 2003 baseball season.
The solution is: stop treating life as a series of moral parables. Once you get that, it all just becomes evidence – and then you wonder whether a single data point about Presidential campaigns necessarily generalizes to baseball or whatever.
IV.
If I’ve successfully convinced you that you shouldn’t form strong heuristics just by looking at a few salient examples where they seem to hold true, then shame on you.
I still feel this comes back down to the simple concept of: “the future is probabilistic”.
Unfortunately, most people are content constructing narratives based on a sample size of 1 (or a few in your AI case)
Now, the question becomes: can we emulate the “grizzled campaign specialists” with “boots on the ground” using a computer?
If Deep Learning had taught us anything, it’s that human minds are not nearly as complex or as uniquely ineffable as we’d previously thought. So, my vote would be for “yes”.
If.
If?
“After invading southern Greece and receiving the submission of other key city-states, Philip II of Macedon sent a message to Sparta: “If I invade Laconia you will be destroyed, never to rise again.” The Spartan ephors replied with a single word: “If”
Wikipedia
I agree that deep learning gives some insights into how the brain of animals might work.
But brains are much more complex than our best designs, both in terms of architecture and training. We still marvel at the emerging complexity of something like how a convolutional neural networks recognizes images. We don’t know, for instance, how many neuron layers we should use, other than just blindly trying different architectures. This is because we do not yet have a firm grasp of how the emergence comes about.
And these neural networks are a far cry from how the brain work. More faithful neural networks like spiking neural networks still do not perform that well. More importantly, we do not know how to train these more faithful networks! This goes to show we do not yet have a definitive idea of how the brain works.
Oh no, I wasn’t intending to imply that present-day neural networks adequately model the complexity of brains. In fact, the opposite is true.
All of this marvelous biological complexity; our capacity for art, music, poetry; all of these deep mysteries of the human soul… Well, it looks like most of them can be modeled with a relatively small number of matrices running on a moderately powerful computer. Of course, you’d need a decent cluster to run the program quickly. Of, and in case you thought your ineffable soul was better suited to playing the ancient and inscrutable game of Go than creating art… well… the same bad news applies.
@Bugmaster – Truly, the creativity of this AI is breathtaking to behold.
Speaking of Hillary loss, what is you theory there? I liked your “crying wolf” article and I do know that you don’t really have to present alternate hypothesis to ‘white supremacism’, but i am still interested in one. Problem I have with celebrity reaction to Trump is that they might be barking up a wrong tree, thus not really helping except in pissing people off even more by calling them racist. (Assuming Trump’s victory was not due to racism)
thanks.
Will write on this later.
I would love to read if you think that Trump is, as Scott Adams claims, a master persuader.
He doesn’t have to be a master persuader to outperform everyone’s expectations, when everyone is calling him an idiot manchild.
People had even lower expectations for Vermin Supreme’s Presidential candidacy, but this didn’t seem to offer him any kind of advantage despite what you think would be a winning promise of giving every American a free pony.
Sadly, America was not yet ready for an absurdist ticket. Give it four years, and hope SMOD doesn’t overtake him.
To be more explicit, though, I think Adams overbilled Trump and pretty much everyone else underbilled him. If I try to ignore his public persona and just look at results, I get the impression of a moderately bright guy in general and a skilled self-promoter and negotiator (in that order) in particular — and I also think that drawing strong conclusions from a celebrity’s public persona is a profoundly dumb thing to do, since any celebrity that wants to stay one needs to quickly get very very good at image-crafting (or have backers that are). That all probably contributed to his victory, though he had plenty of other advantages too. But I don’t get the impression of any kind of genius.
“Sadly, America was not yet ready for an absurdist ticket.”
I look at the President-elect and beg to differ.
Is there actually anyone except Scott Adams who believes that Trump is a Master persuader? Certainly he can’t be so in the book learning sense, since Trump reads no books and I’m sure he does not even know the names of any of the persuasion techniques that Adams writes about .
I suppose DT could be a Master Persuader in the sense of what he automatically does– tells people what they want to hear and keeps repeating it, like “the Wall.” And of course he’s naturally a bully and a macho person, and he is a billionaire and a successful business man (supposedly, but without tax returns it’s hard to know) and a celebrity reality TV star. All of these are things that many Americans admire to the point of worship.
So you’re abandoning your earlier claim that he’s read Mein Kampf and keeps it by his bedside?
“Is there actually anyone except Scott Adams who believes that Trump is a Master persuader” Two other people at least: Me, and I guy I briefly met who does marketing for non-profits. Adams claims that Trump is buddies with the best living persuader, Tony Robbins.
What does reading books and learning the names of techniques have to do with being a Master Persuader ?
There are 64M people who apparently are convinced.
Trump attended Fordham University in the Bronx for two years, transferred to the Wharton School of the University of Pennsylvania in Philadelphia, graduated from Penn in May 1968 with a Bachelor of Science degree in economics.
Perhaps Ivy league schools required people to read books back then, I’ll have to check.
I believe that Trump was a Dale Carnegie student, who is one of Scott Adams’ main influences.
@Moon
Some people are naturally gifted at certain skills and some are very hard to train for people without natural gifts. For example, I would argue that ‘charisma’ is partly body language and that it is very hard to train yourself to a very high level of charismatic body language.
>So you’re abandoning your earlier claim that he’s read Mein Kampf and keeps it by his bedside?
I’m not Moon, but IIRC Trump claimed this was the case but seemed to have mixed it up with a different book on Hitler, suggesting he keeps it by his bedside but has not read it.
Not a safe inference from the evidence at hand. Voting for Trump does not imply that you believe that he is a master persuader in any way, and not voting for him does not imply that you do not think so.
1. Colleges and even high schools don’t send people to your place of residence to look over your shoulder while you read, so “required reading” should really be interpreted as “expected reading” in this context. I’ve known plenty of people who didn’t do their expected reading either sometimes or all the times, and some of these people were bright enough to get good grades anyway. So even if Wharton expected him to read, that isn’t very good evidence that he did in fact read. If he got really good grades at Wharton, then that would be evidence either that he did the reading, or is actually very bright, or is very good at cheating and not getting caught.
2. Even if we just assume Trump did his required reading in college, that doesn’t imply that he did so out of a sense of intellectual curiosity, or that he’s remained intellectually curious and continued to educate himself since then. In fact, all evidence about this such as the testimony of Trump’s biographer and family and the fact that Trump doesn’t talk with very much sophistication about complicated ideas or even name-drop books or authors he’s read or is reading suggests that Moon is correct: Trump is not much of a reader.
Sometimes it makes sense to pick your battles. It looks a little partisan when you try to defend your guy against accusations that are almost certainly correct and are mostly meaningless anyway.
As long as you hold all Ivy league graduates to that same standard, I’m fine with it. If “picking your battles” and “partisanship” are based on the argument Trump never read a book, then I think you might want to look in the mirror.
It’s my understanding Trump actually authored some books as well.
Our host reviewed one here:
https://slatestarcodex.com/2016/03/19/book-review-the-art-of-the-deal/
“Trump is no psychology expert, but he’s sure done well persuading people in real life. After a few months of attributing his victories to blind luck, most people have accepted Scott Adams’ hypothesis that he’s really a “master persuader””
My view is that signalling like this as a politician would be a big mistake, especially in this past election cycle. My guess is Trump avoids this intentionally. Whether or not he drops this in private is unknown. I am personally not very interested whether a presidential candidate is an expert in comparative literature, for people who do care, Trump isn’t your candidate.
@tscharf:
Like many (most?) celebrity authors, Trump’s books were ghost-written. Here’s a long-form New Yorker article about the ghostwriter’s experience.
I’m not holding anyone to any standards. I’m just stating the fact that, according to the first-hand testimony of the people who know Trump the best, he is not an avid reader or an especially deep thinker.
Stating that Trump is not a reader or a deep thinker is not a criticism of Trump — it is simply an observation about him. To conclude that it’s a criticism is to assume that reading books and thinking deeply about things are activities that everyone should engage in. I don’t believe this, and I’ve never asserted anything like it, so I’m not sure why you would conclude based on this that I am an anti-Trump partisan.*
This is also why Moon’s implicit criticism of Trump fails for anyone who doesn’t already agree with it.
I agree completely.
It’s not unknown. There are many people who know Trump personally. They all seem to agree that he doesn’t read books or think deeply about things. I’ve seen testimony from various people who know Trump to this effect, and none to the contrary. This is not even controversial.
I’m not talking about comparative literature. I’m talking about scholarly writing in general, including in fields such as diplomacy, economics, technology, war, law — you know, the sort of stuff that presidents make decisions about. That doesn’t mean Trump will be a bad president. Arguably (very, very arguably), deep thinkers make terrible presidents (Woodrow Wilson, FDR, Barack Obama) and bold, decisive men of action make great presidents (George Washington, Andrew Jackson, Theodore Roosevelt).
*Actually, I do. It is because you are pro-Trump and all arguments are soldiers. But this is partisan thinking and I think you can do better.
Right, he ghost authored it the same way almost all politicians do for their life stories. Let’s hope he at least read it one time, ha ha.
Nobody is going to confuse Trump for a preeminent scholar, and I doubt he is going to be on the short list for Nobel Prizes anytime soon. Bush wasn’t exactly well spoken, and Reagan/Bill Clinton excelled in plain speak.
The hope is that Trump will surround himself with people who can temper his ADD, have good domain knowledge of their areas, and he takes reasonable care in decision making. There is certainly risk it can all go sideways. If your vote is to reduce risk, Trump is not your man.
Being the smartest guy in the room didn’t fix Syria, ISIS, Libya, etc. so it’s not all about academic prowess, some is instinct, much is luck, some things you can’t fix, and we don’t get alternate universes to compare apples to apples with different leaders.
I liked Bill Clinton (didn’t vote for him), Obama did a reasonable job overall and expect Trump will be a similar randomly mixed bag.
Well I didn’t believe it before (I found him not very persuasive) but as he persuaded enough people to get elected president of the USA I have come round to thinking that, yes, a master persuader is what he seems to be.
Or perhaps Comey and Assange and Putin and Bannon and Hannity and the like are the Master Persuaders.
We have a ridiculous “Great Man” myth in the U.S.– as if he had to have won it all by himself. Not so.
Yes, we’re on the same page now, but mostly because you changed your argument from “Trump might be a thoughtful person and habitual reader — we just don’t know, but he went to UPenn so he probably is” to “he’s probably not a thoughtful person or a reader, but that’s not a bar to being a good president”.
However, I think Bush was actually very well spoken and misspoke intentionally because he knew it would appeal to his base and evoke self-defeating smugness from his critics. Bush’s performances at the debates were masterful. I think his grasp of the (spoken) English language is probably much higher level than pretty much anyone gives him credit for (including his biggest fans).
Consider: “Is our children learning?” Overeducated liberals will recognize this as incorrect grammar, but most people don’t know the rules of grammar and just rely on what sounds right. So most people when they hear this phrase will say under their breath “Are our children learning?” and that sounds even more wrong even though it’s technically correct. So intellectual snobs are all like: “Bush is stupid.” And regular folks are all like: “Liberals are stupid. Bush talk gud.”
I see Obama, Bush (both of them), and Clinton as mostly continuing and elaborating Reagan’s policies, which I think were a wrong turn at the time but basically just something we have to live with now. Trump’s election actually makes me hopeful we’ll start in a different direction even if I don’t think much of him as a person.
But that’s based on a lot of beliefs that don’t play well at SSC, so I’ll leave it at that.
Judging by the result of a quick Google, his ex-wife claimed he had a book of Hitler speeches near his bed. When Trump was asked about it, he said that a friend had given him a copy of Mein Kampf. He didn’t claim to have either read it or kept it by his bed. The friend said he had given him a book of Hitler’s speeches.
Yeah, obviously the “Trump keeps a copy of Mein Kampf by his bedside” was untrue.
It’s just amusing that the same person will claim that Trump is both near illiterate and he’s a avid reader of the literary output of Adolf Hitler. It’s a poke at Moon / Jill’s stupidity that she can’t even keep her claims consistent.
Regarding W. Bush’s intelligence, see this essay. It convincingly argues that his perceived stupidity was a combination of deliberate ploy and media narrative. It seems likely that Trump has a deliberate persona in the same way, but I don’t get the impression there is a brilliant mind behind it.
@rlms
My theory is that George W. Bush is mostly a very lazy, intelligent person, who is poor at off the cuff speaking (and very good at socializing).
This was a great advantage for the debates, because he got misunderestimated in the run up, but then he studied very hard for the debates and thus did very well. Then during his presidency he slacked off and let others do most of the work (Dick Cheney took huge advantage of this during his first term).
On the subject of Trump having a book of Hitler’s speeches …
That would make sense for reasons that have nothing to do with the contents of those speeches. Hitler was a brilliant damagogue, very good at moving people with his speeches. That is a skill that would be worth learning for people whether or not they agreed with anything Hitler said.
Someone posted this video, I’ve forgotten why, in the comments from Scott’s previous blog post. If the McLaughlin character, in the Saturday Night Live skit here, is a Master Persuader, then I suppose Trump is too. But it’s about a certain habitual way of acting, not about a conscious use of particular techniques.
http://www.nbc.com/saturday-night-live/video/mclaughlin-group/n9987?snl=1
Moon, you know John McLaughlin was a real person (he died last year), and the McLaughlin group a real show, right? And the SNL parody wasn’t that different from a real episode.
Definitely not a master persuader. At least, he never convinced Eleanor Clift (also real) of anything that I remember :-).
Yes, I know he was a real person. The guests on the McL show in the skit are like the people Trump abuses– like Hillary, about whom he encouraged people to chant “Lock her up.”. Trump’s voters are like the viewers of the show. I see from wikipedia that the show ran from 1982 until the guy’s death in 2016. So it was popular for decades. I would expect that McL convinced his viewers of a lot of things during that time, with his macho know-it-all act and just making stuff up– just like Trump does.
Scott Adams claims are really dubious. He’s making really huge claims and then accepting really weak evidence as proof of his huge claims. Trump as master persuader lost the popular vote by 2.1% and 2.8 million votes. I would expect a master persuader to do much better than that.
Scott Adams original claim was that Trump would win the election by a landslide, and he made this claim at a 98% probability. It clearly wasn’t a landslide and just because Trump won doesn’t mean 98% was a meaningful prediction.
http://blog.dilbert.com/post/127791494211/nate-silver-gives-trump-2-chance-of-getting
Also consider that Scott Adams endorsed Hillary Clinton when he thought she was going to win. Then he switched his endorsement back to Trump. This guy predicts with 98% confidence that Trump will win by a landslide, then endorses Clinton, then claims he was right all along when Trump wins. Don’t listen to Scott Adams, he’s not interested in facts, just creating entertainment and driving traffic to his blog.
http://blog.dilbert.com/post/145456082991/my-endorsement-for-president-of-the-united-states
http://blog.dilbert.com/post/150919416661/why-i-switched-my-endorsement-from-clinton-to
Anyone who bought the whole “I’m obsessed with Trump but not endorsing him.”, or read “I’m endorsing HC for my personal safety” as sincere is operating from a different place than I am.
Adams was Team Trump from day one. If I’m guessing at his motivations, he was making a gamble, sacrificing speaker fees during the election for larger ones during the administration.
Considering all the obstacles that Trump faced (press and Republican elites hated him, lots of strong primary opponents, being greatly outspent, the Access Hollywood tape, the hack of his tax returns, and having at least one foreign government working against him), he could only have won if he had something amazing going for him.
I always thought his 98% thing was meant as a kind of a joke but that he thought Clinton would lose. He was right.
I don’t think Trump is some kind of secret genius. I do think it was a real shock that he won. And I also think it was more that Hillary lost than that he won.
But there isn’t one big simple reason that she lost, more a mesh of small, discrete ones. Part of which, in spite of what Scott says about disregarding that article, I really do think was the attitude that she was owed the presidency, that it was Buggins’ Turn and all she had to do was go through the tedious but necessary process of running a campaign before the triumphant inauguration of the First Woman President.
I mean, reading things like the recruiter who worked on hiring for her campaign (link courtesy of Multiheaded, formerly of this parish), where (with apparently no insight after the fact) he says:
Which makes me sink my head in my hands and cry “Almost half your campaign headquarters people had never worked on a campaign before? Did you want to lose?” And this is a laudatory article, but I came away from it thinking I wouldn’t ask him to hire a drunk to get pissed in a brewery.
It wasn’t so much that “They used Big Data and it was the wrong decision” as that they copied what worked for Obama because it worked so well, but they copied what worked for Obama and didn’t seem to realise that Hillary was different: she’s an old white establishment candidate, why would minority – especially African-American – voters turn out in the same numbers as they did back in 2008 and 2012 when the choice this time was between “old white rich person and old white rich person”? That’s where the link between running a modern bleeding edge tech campaign that disproves the old chestnuts about knocking on doors is what wins votes, and listening to the campaigners on the ground providing feedback because that’s how you get your data in the first place was missed.
Sounds dumb out of context, but I have no idea what the usual rates are. It could very well be that half is the usual proportion of campaign veterans on staff, or even higher than usual, bearing in mind that a campaign headquarters needs a lot of e.g. lawyers and admin people and number-crunchers for whom previous campaign experience would be no advantage.
Nornagest, I was going on the idea that it’s headquarters. Even granting what you say, having just under half your staff going “Okay, so – what do I do now?” because they have no idea how a campaign works means that the people who do know how it works are going to be doing a lot of hand-holding when they should be working on getting the candidate elected.
Mostly volunteer centre in Nowheresville, where enthusiasm and a strong back means more? No problem. Party headquarters where it’s quite likely you’ll be taking calls and emails from people wanting guidance on strategy or can you send them more “I’m With Her!” stickers or how does the candidate’s position on green energy mesh with the threat this poses to the lesser spotted hornswoggle whose habitat is uniquely threatened by wind turbines? Not so much.
Maybe Scott is right and all these “where did it all go wrong?” articles are all wet, but I’ve been reading about petty squabbling and jealousy and jostling for attention and jockeying for power and back-stabbing amongst the campaign, which I found extremely disheartening. But if half the people hired on were stepping on the old hands’ toes because they didn’t know any better, and were throwing strops because they didn’t understand why – if they were Head of Big Numbers – everyone was asking Joe’s advice instead (it’s because everyone in the know knows that Joe has pull), then this makes sense of it.
I think I might actually expect a campaign HQ to have more people whose day-to-day jobs take no political skill or experience than one of the campaign’s branch offices, because more of the HQ’s job is purely administrative — sure, you need some people figuring out high-level strategy and giving directions to the branch offices and putting out fires, but you’ll need a lot more people managing things like IT and payroll and benefits. There is a ton of that needed for any billion-dollar enterprise, most of it is necessarily going to be centralized, and an Excel spreadsheet looks pretty much the same whether it deals with political donations or revenues from selling widgets.
>I think I might actually expect a campaign HQ to have more people whose day-to-day jobs take no political skill or experience than one of the campaign’s branch offices, because more of the HQ’s job is purely administrative — sure,
Given the unique legal, political, and cultural requirements that a campaign faces, I’d want even my purely administrative types to have experience in campaigns. I want them to know how to set up and tear down a billion dollar operation in a little over a year, to know the legal hoops they’re required to jump through, to know not to cash those campaign checks that come in from NAMBLA. The best way for them to prove that they have that knowledge is to have done it successfully in the past.
Its hard to be a full time campaign worker with major elections every 4 years, there simply aren’t going to be enough to staff a presidential campaign (without stealing all the top players from Congressional runs, which is a poor use of resources), so you have to take people who are willing and able to either take long periods of time off from their jobs, or sacrifice free time/time with their families. This skews heavily young (no families yet and more likely to be passionate about a specific candidate), that 26 year old who worked well during Hillary’s primary 8 years ago? Pregnant with her second kid now. Person who took 3 months off after graduating college to help 8 years ago? Manager who can do weekends… but not every weekend now.
Thanks. I am expecting another quality article.
The key piece of data, to me, is that the Senate Republicans in battleground states almost uniformly outpolled Trump:
WI:
Trump 47.8, Clinton 47.0
Johnson (R) 50.2 (+2.4 wrt Trump), Feingold (D) 46.8 (-0.2 wrt Clinton)
OH:
Trump 52.1, Clinton 43.5
Portman (R) 58.3 (+6.2), Strickland 36.9 (-6.6)
PA:
Trump 48.8, Clinton 47.6
Toomey (R) 48.9 (+0.1), McGinty 47.2 (-0.4)
NC:
Trump 50.5, Clinton 46.7
Burr (R) 51.1 (+0.6), Ross (D) 45.3 (-1.4)
FL:
Trump 49.1, Clinton 47.8
Rubio (R) 52.0 (+2.9), Murphy (D) 44.3 (-3.5)
Notably, none of these 5 Republicans was a Trump supporter; they all range from moderate to traditional conservative.
If Trump was bringing in people who don’t normally vote Republican, he should be outpolling Republican senators in battleground states. The reverse was generally true; more people in purple WI, OH, PA, NC, and FL voted for a moderate Republican senator than for Trump.
That suggests the generic Republican would have done better than Trump, and if conventional wisdom about Hillary being a bad candidate is correct, a generic Republican would have done even better than that. Based on national polling during the primary, even Cruz polled better against Hillary than Trump, and Rubio and Kasich both polled much better against Hillary.
We also know that the fundamentals favored a generic Republican–it’s rare for parties to hold the White House for a third term, the economy wasn’t that great, the right track/wrong track numbers weren’t good, etc.
So to me, the story looks pretty simple. The election favored a generic Republican, and Trump wasn’t bad enough and Clinton wasn’t good enough to overcome that. And if the polling looked wrong until the end, the biggest late October surprise wasn’t Hillary’s emails (which would show up as uniquely hurting her), it was Obamacare premium hikes (which hurt Democrats across the board).
I am sorry, I guess you are unaware, but Tuesday changed the narrative.
An alternative interpretation of those data is that Trump pulled in a lot of voters who normally didn’t vote Republican, drove out some regular Republicans. The regular Republicans still voted for the senatorial candidate, who wasn’t tarred with Trump’s faults, and the voters pulled in voted party line.
Should the default assumption really be that the senatorial candidates get the same proportion of the two-party vote as Trump? All five were incumbents.
Pretty sure the Trump team did use statistics… They just weren’t the rigged ones the entire MSM used.
Because, I’m sorry, if anyone was the ‘grizzled campaign specialist’ it was Hillary.
What evidence do you have that the statistics used by the MSM were “rigged”?
It seems reflexive at this point. “Rigged ____.” “Crooked ____.” Get ready for 4 years of this nonsense.
Actually, I heard that even the Trump campaign didn’t expect Trump to win from their own internal polling.
Trump claims he believed the MSM statistics and was surprised when he won, although I suppose it’s possible he’s lying.
But I think Nate Silver’s explanation is most persuasive.
Of course he was surprised that he won. Because he didn’t do the winning himself. Right Wing “news” media like Breitbart, Drudge, and Fox won for him– and those “news sources” told tons of lies about HRC constantly. Not to mention that even the NYT covered HRC more negatively than Trump. And Comey and Assange, and apparently Putin, helped him too.
He didn’t have any magic, or personal secret, or Master Persuader skills. He had all of what I listed above winning the election for him. He didn’t actively win the election. He just ended up winning when all of the above influences forced HRC into a losing position.
Wasn’t there anyone at all rooting for Hillary?
You find me some news media that did.
Here is a list. Even those that technically endorsed her gave her more negative than positive coverage.
https://pbs.twimg.com/media/C16VBDeXUAA3lNB.jpg:large
And she got more negative coverage than Trump did.
http://www.vox.com/2016/4/15/11410160/hillary-clinton-media-bernie-sanders
Maybe Clinton is actually a worse person than Trump, so there is more negative news to report. You can’t judge their moral worth based on their public personae.
“Maybe Clinton is actually a worse person than Trump, so there is more negative news to report. You can’t judge their moral worth based on their public personae.”
Or maybe the negative news coverage of HRC convinced you, and lots of other people, of that opinion, and that is why HRC lost.
If I were making the case for media bias against Clinton, I would argue it in the other direction: Clinton and Trump got reasonably equivalent negative coverage, but Trump is a significantly worse person/president than Clinton.
I’m not sure why you’d assume that I’m more susceptible to bad information than you are. Isn’t it possible that your sources of information that exculpate the Clintons is at least as flawed or slanted if not moreso than the sources I rely on to conclude that they are horrible human beings?
Also, the Clinton Foundation supports policies that, based on your previous comments on SSC, you should oppose if you were arguing based on principles rather than partisanship. Does the Clinton Foundation’s neoliberal agenda bother you at all? Does conflating charity and “economic development” (building sweatshops) bother you at all?
I should probably make clear: I’m a radical lefty. I consume conservative news media only in the context of contrasting different points of view on an issue. My problems with the Clintons are based entirely on my left-derived values, not Rush Limbaugh two minute hates.
>Here is a list. Even those that technically endorsed her gave her more negative than positive coverage.
Moon, those figures are for the PRIMARY, not the general election. you’re leaving out 5 months of data, and the far more important 5 months, probably because you know they prove the opposite of what you’re claiming.
You’ve posted this assertion before, and have been told the same thing before. Now it’s not just Vox saying the opposite was true in the general, it’s literally the same author who is saying it. Now, for you, I see three possibilities. You can stop reposting those same articles, you can explain how the same author was right in may but wrong in December, or you can ignore these inconvenient facts and continue on as before. I’m betting on the last one.
The polling error for the popular vote in 2016 was actually smaller than the polling error for the popular vote in 2012. On average, final polls put Clinton 3-4 points ahead; she won by 2. The polling error in 2012, in contrast, was a full 2.7 points. The key difference is that the polls overstated support for the underdog in 2012, but overstated support for the leader in 2016.
This is really important. The wrong lesson to take from this is that we shouldn’t trust polls. The right lesson is that people were arrogant in their assertions without evidence to back that attitude up. If you’re going to be arrogant, don’t just look at the raw numbers and go from there. You have to have a deeper understanding of what they mean. That’s why 538 kept warning us before the election that while Clinton was the favorite, she didn’t “have it in the bag”, and people should stop acting like that was so.
To be fair though, while national polls were fairly accurate, the problem came from state polls.
My favorite quote on this comes from a sci-fi webcomic of all places. The character speaking it is an AI to boot. I don’t know if the author got it from somewhere else, though.
–Collective central AI “The Oracle”, S.S.D.D.
The problem was especially infrequently polled Midwestern states that everyone paid no attention to because they assumed Clinton had them locked down.
Well, except PA. That got polled a LOT and still surprised.
Yeah, I think a part of what’s going on is that the “technical data nerd Moneyball” people are still people who are vulnerable to the same biased reasoning and generalizations from n=1 that everyone else is.
Instead of “trust the data not intuition”, I wouldn’t be surprised if a lot of the campaign people’s philosophy was more like “we know what’s going on, we don’t have to listen to how campaigns are ‘traditionally’ run, all these numbers prove us right.”
People took 2012 to mean “the polls are right”, or maybe even “people who think Republicans could win in the face of polls are cranks who will get their comeuppance” when a better lesson might have been “the polls aren’t horribly skewed” (which people were claiming!). As you say, the polls were as right as they normally are.
And relatedly, I see people attacking Nate Silver in the aftermath of the election, when he was the one saying Trump had a 1 in 3 chance of winning, facing lots of criticism from his own side for not putting it lower. He became the face of the “Moneyball” team, when his actual number-crunching led to more pessimism than others on that “team” for whom it’s just a tribal affiliation rather than an understanding of the polling data.
I think the problem with “just following the data” is that you’ll systematically underestimate risks that way. It’s very hard to make a model account for the possibility that it could be wrong.
Experienced people care less about getting the best model, and more about staying on the right side of risks. If you make sure that your gains are unbounded and your losses bounded, this dynamic can work in your favor. If it’s the other way round, you’re playing a sucker’s game.
The numbers in your estimate are still important, but don’t neglect which way the < points.
Nate Silver went even further and explained which states were the hardest to call, and therefore were the reason for the uncertainty in his predictions. At one point he mentioned that if Hillary lost Florida the election was basically impossible to predict. And then Florida was called for Trump, at which point Silver’s remaining predictions summed to no prediction because it was too close.
Really?
Haha willful blindness by Scott. For the love of God, your bias is showing. For someone so hung up on fucking rationality and probabilities and whatever else that is supposedly ‘pure’, you sure do have a lot of bias.
I read a lot of bloggers. The only neutral person has been Robin Hanson. Then Tyler Cowen has been mostly neutral. But you and Ran Prieur literally shat the bed with Trump
Literally?
I don’t think that’s what Trump wants to happen in the bed.
Sigh… rationality is not pure. It never was. Rationality (the epistemic kind) is forming correct beliefs about the world. It doesn’t mean being perfectly emotionless. You still cannot like the world that has been correctly deduced.
I blame star trek.
“Rationality (the epistemic kind) is forming correct beliefs about the world”
Well, there’s 2 ways that can go.
1: Form beliefs that are in accordance w/ reality
2:Form “the right” beliefs.
If you’re going by #2, Scott was correct.
But if you’re going by #1, a lot of arrogant fools were proved wrong by this election.
If biases and preferences are just the same thing, then it is neither possible nor desirable to eliminate biases.
Also, preferences/biases are not the same as predictions. Cowen and Hanson may not have made strong predictions, but that doesn’t mean they didn’t have any strong preferences — I don’t really have any evidence either way. While Ran Prieur might have predicted Clinton would win (I actually didn’t realize he had until he posted to say he was wrong about it), I suspect he did not feel terribly strongly that Clinton was a much better choice than Trump based on his previous output.
Banned, obvious reasons
From where I sit, there were 21 arrogant fools and 20 of them got their comeuppance.
From where I sit, there are 300 million arrogant fools, and they’re all getting their comeuppance, even if some of them need to wait.
Yeah, two arrogant fools won the Primaries and one of them even won the Presidency.
I mean, one of them had to.
Yeah, that was the Right Wing media story, and even the NYT story, as HRC was covered more negatively than DT was. That they both were horrible candidates. However, reality is different. Most people are soon going to be wishing HRC had won,
A lot of people have made a very strong case that Clinton’s policies were pushing us towards war with Russia (which quite plausibly could have become a nuclear war).
If that’s true — and there’s no way to know whether or not it is true now that it has become a historical counterfactual — than any outcome of a Trump presidency short of WWIII is still better than a Clinton presidency.
Doesn’t that make you stop to think for even a heartbeat that your political loyalties might be misplaced?
No, that case seemed strong to Trumpsters, but not to me. To me, the case that Trump is a puppet of Putin seems stronger.
I’m not a “Trumpster” I don’t think, but the case for it seems strong to me as well — based in part of the analyses of various diplomatic, military, and geopolitical experts who made the same case. Why don’t you trust those experts? Are they all “Trumpsters”? Is criticism of Hillary Clinton prima facie evidence that someone is wrong or untrustworthy?
Also, it seems obvious to me that not only do these opinions not contradict each other, but that they actually reinforce each other:
1: Trump is a Putin puppet.
2: Trump will not start a war with Russia.
(Though I don’t think Trump is a Putin puppet, I just think he said nice things about Putin to differentiate himself from the warmongering Democratic establishment.)
Both sides had experts on their sides. And being a Putin puppet does not guarantee no war with Russia. The president could try to appease Russia’s every move, be found to have connections with Russia that make him impeachable, and then we have President Pence, who ends up in a war.
There is no way that DT just “said nice things about Putin to differentiate himself from the warmongering Democratic establishment.” The Dem establishment is not warmongering at all, compared to GWB, and probably not warmongering, compared to what our eventual President Pence will likely end up doing.
If DT was speaking well of Putin just to show that he didn’t want a war with him, he wouldn’t have gone so overboard with it. Trump has nothing but praise for Putin, and keeps favorably comparing Putin to Obama and using the comparison to bash Obama. If a Dem were doing that to a sitting Republican president, then Republicans would be calling that treason in a heartbeat.
“HRC was covered more negatively than DT was”
Wow, just wow.
Just out of curiosity, how many active FBI investigations was Trump under during the primaries / general election?
Just for grins: could you please tell us Hillary’s top three accomplishments as a US Senator? As Sec of State?
Hillary’s emails:
Hillary took the job as SoS so she could sell power and influence in exchange for payoffs to the CGI and CF. She set up a private email server because she knew a lot of her corruption would be carried out via email, and she wanted to keep those emails safe from FOIA requests.
the consequences of doing that included top secret information being placed on insecure systems (felony) and her system being hacked so that multiple foreign governments hostile to US interests (including Russia) had massive amounts of blackmail material to use against President Hillary Clinton.
that fact that this was not a daily source of discussion during the campaign is just one example of how easy the press was on HRC.
Prediction: deep learning will peter out before AGI.
Define “peter out” and “AGI”. I know it means Artificial General Intelligence but what exactly do you mean by that?
A lot more insight will be needed for AGI. Feel free to define AGI in some sensible way.
Seems perfectly reasonable, but don’t forget that deep neural networks aren’t the only cool thing happening in AI research. Reinforcement Learning is pretty important, and so I think RNNs and differentiable neural computers/neural Turing machines are exciting.
Also, it’s not like AI progress now faces the same challenges as it did in the 1970s: there are more researchers + money, but much more importantly more data and compute power, as well as stronger incentives to work on the problem.
Deep learning – AKA deep neural feedforward networks have already “petered out” in a way; lacking the ability to deal with long-term correlations, requiring fixed input sizes. They’re good at perception – extracting a pattern from a single fixed size data input like an image or a short audio clip. We are already trying to move beyond that IMO.
Deepmind uses deep learning + reinforcement learning, so there’s that…
Differentiable neural computers/neural Turing machines are also deep learning, and all the big advances in reinforcement learning in the last few years consisted in applying deep learning to mostly existing RL methods.
I will remember, thank you 🙂
Seconded.
“If I’ve successfully convinced you that you shouldn’t form strong heuristics just by looking at a few salient examples where they seem to hold true, then shame on you.”
Could you unpack this? It seems to me as though you made a reasonable argument, but maybe I’m missing something– possibly that you had too few examples and/or to little logical rigor to have made a sound argument.
“Don’t trust heuristics” is itself a heuristic. The article is self-refuting.
Not trusting heuristics is by definition. You are supposed to trust algorithms cause you have proof they work. Heuristics by definition only work part of the time.
I suppose the next question is what you do when you don’t have an algorithm– which is most of the time.
Also Algorithms as defined work on problems already formally modelled. The first part is often even deciding how to put messy reality in such a model.
Well, you use a heuristic, but don’t trust it, obviously. Put a big comment saying “Here be dragons!” next to it, do all the sanity checks you can think of before and after, and still avoid betting anything important (to you) on the results.
Seems like the moral of the article isn’t “Don’t trust heuristics”, as the final part should make clear, but the more simple adage of “Sample sizes are important”.
That’s something that’s pretty universal I’d say
I gave three examples of cases where people formed a heuristic from a few cases and then it failed.
If you’re concluding that heuristics formed from a few cases often fail, then you’re doing so off n = 3 cases.
That comment made me chuckle, however I went into that article not believing in heuristics. The key words (which I totally missed the first time) are “[i]f I’ve successfully convinced you…”
I would contest BAA’s comment that the article is self-refuting. The article isn’t saying to not trust the information presented, it’s saying to not form a heuristic belief based on the three examples given. There’s a big difference there.
The line “The solution is: stop treating life as a series of moral parables.” is excellent and something I shall be saying to people in the future.
Maybe I already knew lots of cases where small-sample-size heuristics failed.
Then you’re not concluding it from his post, you’re concluding it from his post plus lots of other evidence. So no shame for you.
I think it should be understood in the same way as the fifth Commandment of the Discordians:
“A Discordian is Prohibited of Believing What he Reads.”
It does not claim that everything you read is false, or that you should never believe in anything (or maybe it does, depending on how you define “believe”). But the fact that you read it somewhere should not, by itself, appreciably change your prior of it being true.
Same here. This article is three anecdotes used to make a point. The plural of that isn’t data.
Pay close attention to the punchline here https://xkcd.com/552/
I think the real lesson of the 2016 election was: the First Foundation is not supposed to know about the Second Foundation’s predictions.
The democrats became too dependent on 538 and Nate Silver. They used it to model their campaign, they used it to decide which states to defend and which to ignore and which states were vulnerable to flipping. The Trump campaign -seem to have- used 538 to see which states were vulnerable to flipping too, and so they attacked the states Clinton deemed secure. It seems there was a small statistic error in favor of Trump in those states so the feedback never became apparent.
If Clinton and Trump had campaigned randomly or blindly, they may not have abandoned and attacked respectively Clinton weak states. The question for the pollers and the aggregators is why this effect didn’t show up earlier. Clearly aggregated polling is hiding it somehow.
This is different to the 2012 election where Romney just decided to ignore polling altogether. This time both campaigns acted on the aggregated polls information, they just interpreted it differently.
This kind of dynamic would apply to, say, a county commission race or a board game or something.
By September of the election year, presidential candidates and their campaigns are fully formed, and have no ability to “flip” or “defend” states. Everything they do, or have done, is broadcast to the entire nation, filtered and reinterpreted through thousands of channels, and ultimately overwhelmed by what everybody else has to say.
The presidential race is conducted on a national level. The results in states differ because their voters differ, not because they are exposed to different campaigning. Take any two counties in different parts of the country with closely similar economics and demographics, and their presidential votes will likely be very similar.
When something big happens, like the crash of the stock market in 2008, or Romney’s 47% video in 2012, or the Comey letter in 2016, the impact on voters’ decisionmaking is spread across the entire nation, not bottled up in certain jurisdictions.
In the weeks running up to the election, the electorate is about as fully engaged as it ever is. That means, the democracy, or the polity, such as it is, has seized control of the argument from the supposed principals.
The candidates go on campaigning, because it would look bad if they stopped, but it no longer makes any real difference. You can’t go give a speech in Indianapolis or someplace and scoop up a thousand votes there; most everyone in Indiana who even hears of it has already made up their minds, and if they haven’t, the one speech (repeating the same canned text in front of a different crowd) is unlikely to make a difference.
As I wrote in a previous thread: for a presidential campaign in the final weeks, the cost per marginal vote approaches infinity. That’s not a joke, it’s practical reality.
We can laugh all we want about Hillary’s campaign thinking it was five points ahead in Michigan, but it wouldn’t have helped a bit if they thought Michigan was suddenly teetering on the brink. Anything that was reasonable to do was already being done — because Michigan’s own politicos had their own hides at stake. Anything that wasn’t reasonable to do is ineffective.
I don’t disagree that votes will be hard to flip in the late part of the campaign, but do note that I am not talking about september or even august. The Trump campaign had been eyeing the Blue Firewall states of the Rust Belt as early as March, and maybe earlier
https://www.theatlantic.com/politics/archive/2016/03/trumps-path-through-the-rust-belt/475767/
Now the problem is that, it seems that due to the aggregation, such tendencies won’t show up in systems like 538. The votes slowly and painfully changing through 2016 just didn’t register much on. Giving credit to Nate, he had pointed out there was a small chance something like this happening, while Sam Wang was ridiculously sure of Clinton’s victory. So maybe the lesson is you can use 538 to initiate the campaign, but afterwards you cannot trust it cause you are putting your finger on the balance.
Votes flip all over the place right up to Election Day. It’s just that the presidential campaigns themselves aren’t the ones driving it.
But, yeah, back in March, it’s a whole different situation.
I agree that the cost to change each marginal vote may approach infinity towards the end, but the cost to turn out each additional vote is not similarly astronomical. I think most of the criticism about abandoning / campaigning in a state is about turnout resources – including both professional / volunteer GOTV efforts, as well as campaign stops by the candidates, which then drive interest, engagement, volunteerism, and ultimately turnout in favorable demographics.
We’ve had this conversation before. GOTV is worth doing, sure, but it is hugely overrated as a generator of votes.
https://slatestarcodex.com/2016/11/09/open-thread-62-25/#comment-433411
https://slatestarcodex.com/2016/11/09/open-thread-62-25/#comment-433471
Thank you for the links. Interesting discussion.
That said, if it’s something worth doing, then that would be more useful than trying to blanket the airwaves with yet more tuned-out ads.
Furthermore, I don’t know if the physical side of GOTV encompasses the entirety of the issue, which is surely one of voter perception. You wrote:
The point, as I understand the criticism of Clinton’s campaign strategy, is that a candidate’s behavior can affect the overall perception of the electorate as to the importance of their vote. I live in a deep-blue state, and so I voted mainly out of a sense of civic obligation. But if the candidate projects a sense of urgency by campaigning in a state, that signals to the voters that their vote is more valuable. That is part of GOTV, it seems to me, even if it doesn’t involve knocking on doors or busing folks to the polls. The candidate can’t change minds, but might be able to shift vote/non-vote incentives more broadly.
Now, I concede that it can cut both ways – magnifying the closeness of the race could have hurt Hillary’s perception of inevitability, given hope and raised turnout of Trump supporters, etc. But it’s almost always to the advantage of Democrats to have higher than average turnout, if I’m reading the demographics correctly; fewer people voted in key areas than 2012.
> The democrats became too dependent on 538 and Nate Silver. They used it to model their campaign, they used it to decide which states to defend and which to ignore and which states were vulnerable to flipping.
They really didn’t, because 538 consistently laid out that Clinton’s lead wasn’t that high in “firewall” states like Colorado, Wisconsin, New Hampshire etc. and that if she lost it would be there.
Yeah we have to give credit to Nate Silver for warning about this, but his main election prediction still was 83.5% for Clinton in a state like Wisconsin, and every poll was also showing a slight-to-decent Clinton lead.
Now what Nate Silver would do normally, is add a slight correction to the Rust Belt states that Trump won saying “the polls here skew pro-Democrat but the results are pro-Republican” but I don’t think that is enough. That is just using the same failed methodology and using the same duct tape fix that failed the last time. I hope Nate Silver strives to do better than that.
The fault lies in people seeing e.g. 5 of Nate Silver’s 83% predictions and thinking that all of those states are safe, rather than expecting one of them to flip.
They almost certainly aren’t independent, so this is incorrect.
“As soon as a metric becomes a goal, it becomes a bad metric.”
Also kudos for the Foundation Reference.
If the metric can become a goal without being the goal you want, it was always a bad metric.
Speaking of which, has taking medication which reduces blood pressure or cholesterol been demonstrated to result in lower risk of heart disease?
“If the metric can become a goal without being the goal you want, it was always a bad metric.”
Not entirely. If you can prevent the people making decisions from seeing the metric directly, or if they don’t care about it, Goodhart’s law will not come into effect. This might happen if, say, some lowly academics are studying some phenomenon in far-away countries that don’t give a damn about what Western professors think about them.
Well, no; the point is that the metric is normally coupled to the goal (which is presumably too difficult to directly measure), but that when you try to affect the metric you end up breaking the coupling rather than affecting the goal.
The thing I struggle with regarding not trusting moralising heuristics, is that even if you don’t actually believe them, it’s very difficult to not engage with them and still “get things done” with other people. So if I think “big data” is the way to go for one thing, but not another, it’s very difficult to have those independent views without people lambasting you for being hypocritical, because the only way they view the world is through these heuristics. It’s why I think so many successful people tend to be evangelicals for a certain worldview – they choose one heuristic, then pick problems for which the heuristic works, rather than pick solutions for specific problems. That way you can still “mobilise your troops” by being consistent.
“X works until it doesn’t” or it’s inverse “X doesn’t work until it does” sounds trivially true but it has a deeper meaning when applied to exponential growth. If you understood Moore’s Law in the 1970’s, then you understood that the internet was going to happen but it caught everyone else by surprise. Techies have been talking about the Internet of Things for a few years now. The surprising part was not that it happened but the exact time when it did. Usually these things seem to work slowly until a critical moment when someone does it in a way that is viable, and then suddenly it’s ubiquitous. That’s why I think self-driving cars are going to overtake regular cars by the end of the next decade, solar power will finally overtake coal and (more speculatively) genetic engineering will go from zero to hundred in the blink of an eye.
True that you can always bet on progressives increasing whatever pet project they have for as far as they can but the main limit on solar power is that the sun doesn’t shine at night and there’s still electrical demand then. Saying solar will overtake coal (unless it turns out to be trivially true because new cheap nuclear technology out-competes coal) is saying that battery technology will take a bunch of giant, unforeseen leaps forward – and that will have much more important ramifications than solar panels in a desert.
It’s pretty situational. In hot areas, the biggest draw on the grid is usually air conditioning, which peaks close to the same times that insolation does — it’s a couple hours off, because the hottest part of the day is the mid-afternoon and the most sunny part of the day is noon, but it’s close. In cold areas, though, the pattern is reversed.
Solar might never outcompete coal without technological changes in Boston, in other words. But I think it has a pretty good chance to in LA.
Battery technology is the obvious solution, but there might be non-obvious ones, such as a reasonably efficient way of using electric power to synthesize liquid hydrocarbons out of CO2 and water. Pumped storage for hydropower is another option, but like batteries it’s a pretty old technology so the odds are not that good of radical progress. And there may be others that haven’t occurred to me.
Pumped storage has the issue of requiring a fair bit of space near to where the electricity will be used to avoid significant losses in transfer, so in most places it’s not a great option. One thing that a number of people are chasing at the moment is using molten salt to store heat in a giant hole and then use that heat to generate electricity.
Additionally, grid wide demand management, particularly using internet of things technology, can target some of the worst peak use issues.
In my understanding, pumped storage is virtually impossible to scale up to meaningful levels due to low energy density outside of areas with uncommon topography. The footprint is just too big if you don’t have hundreds of feet of elevation change in a small area. That being said, there’s been some interesting work done near my area on re-purposing abandoned mines for underground pumped storage that tries to minimize the footprint – the underground pumped storage I read about most recently was aiming to build in a populated suburban area.
It occurs to me that if you are trying to reduce fossil fuel use rather than to eliminate it, one possibility would be fossil fuel generators that were easily turned on and off. Use them to fill in the gaps in the renewable power sources.
This is something that is already done to an extent with peaking plants, which often use gas. They aren’t as cheap to run as your base load generators using coal and the like, but the turbines spin up quickly and they can be used when energy prices go over a certain amount to make them economical.
I think they will still serve a purpose in a renewable future to round off demand peaks, but I think the biggest challenge at the moment is how to replace the full base load of energy while your solar/wind/other renewable isn’t producing anything.
Stem CTO: Lithium-Ion Battery Prices Fell 70% in the Last 18 Months
Battery prices may or may not continue to drop in price but I certainly wouldn’t say it needs to be unprecedented. We’ll see what happens in the next 10 years.
Could the explanation for the drop in prices be related more to the cost of materials and energy than any actual improvement in the technology?
http://www.nasdaq.com/markets/crude-oil.aspx?timeframe=3y
(Note that the article was posted about 6 months ago, almost exactly 18 months from the start of the drop in crude oil prices.)
Falling prices also don’t say good things about demand in many cases.
The article explicitly states:
> According to the National Renewable Energy Laboratory, there was a total of 53 gigawatt-hours of lithium-ion cell production capacity in 2015 — but only 40 percent of that was utilized.
So the price drops because of over supply.
It’s a power transmission problem. That’s a materials strength/weight problem.
Once we have transmission power lines on the space elevators, solar will have overtaken coal.
If Solar Energy out-competes Coal, it will be most likely due to changes in the coal industry, and not the other way round.
I’m guessing you weren’t around in the 1970s. And I think you need a refresher on Moore’s law, because there is nothing about raw transistor count that makes the internet inevitable.
Moore’s law, in the 1970s, was compatible with the internet. It was also compatible with the science-fiction future where everybody has a walking, talking C-3POish robot servant that communicates with us and with other computers by voice and written text and maybe exchanging magnetic tape reels because nobody ever really developed the high-end networking technologies. Or because it was understood that connecting your robot to any sort of network pretty much guarantees that it would become a deep-cover zombie slave of someone who means you ill, and that never ends well.
The internet was not inevitable, it certainly wasn’t inevitable just because bignum transistors, and if you are going to say that it was obvious in the 1970s, then you kind of need to have said so in the 1970s.
I agree. Moreover, while “techies” have been talking about the Internet of Things for years now, a lot of the content of that talking has been “oh god, what a horrific security nightmare”.
Yeah, the techies I know have mostly been talking about the internet of things that should not be on the internet. But I know a lot of security and ops people; maybe the hype’s more positive among the founder class.
The (complete lack of) security of IoT devices is really more of a legal liability problem than a technical one. Neither the manufacturer nor the customer currently bears any liability when their device is compromised and used as part of an attack, so neither has an incentive to demand security.
Making manufacturers of compromised devices liable for their devices’ participation in botnets would cause an immediate shift in incentives and priorities. It would also probably be the end of the IoT, since securing Internet-connected devices is an extremely costly proposition that isn’t really economically feasible for mass-market consumer products.
Is that a worthwhile tradeoff? My gut says that yes, this is a case where an externality needs to be internalized, because the already-realized danger of teenagers (to say nothing of criminal organizations and rogue governments) who can exercise an effective veto over anyone’s participation on the internet is just too high, and worth whatever cost in innovation and consumer convenience we’re giving up. But it’s definitely a debatable point.
Cypren, institute that rule (making manufacturers of compromised devices liable for their devices’ participation in botnets) for computers means the end of the Internet. It seems to me this is a very bad rule. In general “make the manufacturer of X responsible for deliberate criminal misuse of X” is little different from “ban X”.
It’s more a question of negligence than direct liability for use in a criminal act. As an example, companies which are licensed to possess explosives are not held liable for a criminal act committed if their explosives are stolen. But we likely would hold them liable for gross negligence if they left those explosives in a marked, unlocked crate sitting in an unguarded parking lot.
There’s a fundamental recognition that companies which are licensed to handle explosives, firearms and similar devices are working with dangerous tools that can be put to bad ends, and they’re held to a certain standard for securing and controlling those tools as a result. We have no such standard for internet-connected hardware manufacturers, and I’m suggesting that perhaps we should.
“Your device was used in a DDoS attack” is probably too harsh of a standard. “Your device shipped with open administrative ports to the internet, a default password and open-source libraries with known vulnerabilities 3 years out of date and was used in a DDoS attack” is probably not so harsh of a liability standard that it would destroy all innovation.
Edited to add: I’m also far more in favor of stricter liability standards for special-purpose devices than general-purpose user-programmable ones. Asking a manufacturer to bear liability for a device that an end-user can install custom software onto is completely unreasonable. But asking them to bear liability for an unmodified device running only their own software strikes me as perfectly reasonable, at least to a certain point. There’s no such thing as perfect security with a networked device, but most manufacturers aren’t even trying to do the bare minimum right now, and won’t until the economic consequences of their negligence are forced back upon them. As I said, this is a classic case of a market externality; those typically can only be solved by outside intervention.
It’s also not just DDoS. Here’s a scary Twitter thread about pacemakers.
The problem with coming up with some heuristic to apply to the election is that the election was very close, and the winner of the popular vote didn’t win the electoral college. Therefore, there were a large number of circumstances, where if even 1 of them had been different, the election would likely have gone to HRC.
Assange leaks, Comey’s statement right before the election, voter suppression, humongous amounts of false news that many voters believed e.g. the story about the pedophile ring in the basement of a pizza parlor that doesn’t have a basement, e.g. the story about pay to play at the Clinton Foundation– just to name a very few of these circumstances.
Heuristics of course are useful in some cases. But in a case where the outcome was extremely close to being the opposite of what it actually turned out to be, you end up splitting hairs, and I doubt that they have real relevance.
The main use that heuristics are usually put to in such a situation, are that of seeming to support one’s pet theory about the reason for the outcome.
The election wasn’t close by any means; of course there is the issue of the popular vote, but in the end what matters is the electoral college votes, and there Trump won handily. In that aspect, the popular vote is just a curiosity, a foot note.
So in this vein, calling it “a close vote” or “a 10% chance of something bad happening will happen 1 in 10 times” is disingenuous. I would believe it if Trump had won by a few electoral votes. “Close” was either Bush Jr victories. It was not a landslide either, but Trump still could have lost a couple of states and still have won handily.
I think the margin by which he won in some states was what I’d call close, but having a lot of small chances to win, and not needing them all for you to win, is not normally what I call close.
Most people who actually put numbers on the election gave Trump something like a 10% chance of winning in the days leading up to the election (538 gave him a much higher chance). So, it’s not really clear that any of these forecasters got anything wrong. Things that have a 10% chance of happening do occasionally happen (roughly 10% of the time!). Of course, it’s also not clear that they got things right—maybe there was some way to give Trump a much higher percent chance of winning but they were all deluded. (Obviously, one can do a more careful analysis using state-by-state data, but that’s sort of orthogonal to my point.)
The people who definitely are wrong are those who conclude from the fact that Trump won that the forecasters were all wrong.
As Scott pointed out in a post before the election, it is also wrong to make big demographic conclusions based on the results. We knew going into the election that roughly half of the voters would vote for Hillary and roughly half would vote for Trump, and that’s what happened. If we learned anything on election day about the electorate, it was only at the margins. (We did, however, learn that we’re in a for a bumpy ride…)
Don’t confuse calibration for accuracy. People who said Trump was 10% to win were 90% wrong. Things don’t “have a 10% chance of happening”, they will either happen or they won’t. 10% describes your level of *belief* in whether or not those things will happen.
It’s customary these days to distinguish between subjective probabilities or credences on the one hand and objective chances on the other. The objective chance that the coin I just flipped would land heads is 50%; this remains true even though I know it landed tails. We need objective chances rather than credences in a number of places in natural science, notably, for the Born Rule in quantum mechanics and to calculate fitness values in evolutionary biology.
OK but almost everything is subjective probabilities, and certainly the election was. Trump did not win because of the blessings of the God of Quantum, he won for deterministic reasons.
Coin probabilities are subjective too, you’re just not good enough at predicting force and spin and air currents to be very accurate. You’d have a more accurate view of the world if you put the idea of “objective chances” out of your head.
We need objective chances in the special sciences as well. Like I mentioned above, fitness values are chances– a paleontologist might want to compare the fitness of two species of dinosaur at a time just before the Chicxulub impact, and it would be unhelpful to say that both had zero fitness because all of the creatures perished in the inferno. Chances also show up in statistical mechanics and some other places, enough that they’re pretty much indispensable if we want to have any picture of the world at all.
Should have evolved some meteor resistance!
Yes, there are cases where you can abstract away the idiosyncrasies of each individual case and pretend you’re running a repeated experiment. And in those cases, the frequentist statistical view, while still false on a fundamental level, works well enough as an abstraction. But a presidential election is not one of those sorts of things. You can’t rerun the election 100 times controlling for various factors and determine that Trump only wins 10 of them due to variance in whatever factors you didn’t control for.
Do you think that I could rerun the Chicxulub impact if I wanted to?
The image of the world handed to us by the sciences teems with dubious entities, from centers of gravity and average incomes to frictionless vacuums and the number eight. Chances are really the least of our worries. I know it might be painful, but your only real choices are to accept that your world is overrun by spooky whatsits or grow comfortable talking about fictions.
That has nothing to do with how appropriate it is to excuse people by saying “Things that have a 10% chance of happening do occasionally happen (roughly 10% of the time!).”
Sure it does. It tells us that if there is a problem with that claim, it’s not with the fact that it invokes chances– Trump did indeed have a 30% chance of winning. You’re uncomfortable with chances because you have no idea what they’re supposed to be. But this gives us no more reason to object to talk of chances than it does to object to talk of centers of gravity.
The actual outcome was determined by a few tens of thousands of votes in a few locations. Studies have shown that “in a few locations”, voter turnout can be affected by even unusual weather.
I actually still go with Nate Silver after this: he won because 1/3 chances come up one time out of three, and he had that high a chance (even though all kinds of polls show the public mostly don’t like him very much, even among his own voters) because Hillary Clinton was a completely godawful, massively arrogant, self-destructive candidate… and also because of all the other reasons.
Lots of real events are multicausal and noisy.
Somebody wrote a model that spit out the number 30. Other people wrote a model that spit out the number 2. A perfect model (let’s call that model voting machines) would spit out the number 100.
The statement implies the model was correct and only probabilities ruled from there. The fundamental issue is how wrong the model was, not how probabilistic voters and data collection is. To the extent that this statement is about an estimate of error in the model or data I suppose it is OK.
I think the issue is whether the modelers were over confident in their results because they had too much faith in their models. A lot of seemingly pseudo independent models were all wrong with 30 being the “least wrong” of them all.
Combining the fact that all models strongly pointed to incorrect results leads one to conclude the modelers have problems, not the data they used. Another argument could be made that these models are not independent at all so it would be wise to have more diversity in the models.
If one wants to make an argument that this election was some sort of black swan event in models, then the black swan needs to be identified (for example people being shamed into not saying they would vote for Trump to a poller). Otherwise the proper action is too revise the models and increase the error estimates next round.
This is becoming a metaphysical question about the nature of uncertainty and randomness and their interactions with models designed by human beings. The perspective of tscharf seems like one extreme of this to me — that everything is ultimately knowable and that uncertainty reflects merely a lack of knowledge. There’s another extreme, that of Pyrrhic skepticism, that nothing at all is knowable and apparent knowledge is essentially a lucky coincidence of beliefs and contingent states of reality. And then there’s a whole spectrum of perspectives in between.
It might make sense at this point to claim I’m taking the middle path, but actually I try to stick to the Pyrrhic skepticism side. This is because human beings and their institutions tend to be biased towards certainty and away from skepticism, and so skepticism needs more and stronger advocates even if they’re surely wrong in some absolute or cosmic sense.
Fitnesses might be objective within a well specified fitness model and calculation method, but fitnesses are very much the result of ignoring a lot of information and trying to give a relevant summary for certain purposes. I wouldn’t say in general that they differ in kind of epistemic content from Nate Silver saying given this model with these assumptions and yada yada, the probability of Donald trump winning is X.
The born rules are a much, much, much more well validated model and calculation method and they appear to summarize the world better, but there’s still a philosophical question of whether they differ in kind from claims made from Nate Silver’s models or only differ in degree.
Expert opinion and mathematical probabilities are different things. Voters don’t randomly vote for people based on quantum uncertainty. All the pollsters had models that they tuned based on expert opinion which is why they all had different answers. It would appear their was systemic error in expert opinion of how to tune the models in the battleground states, namely estimating turnout (reading tea leaves).
They can go back and find where these errors are, but the next election cycle is different candidates, different voters, and different issues.
This weeks postmortem at 538 says more HRC voters stayed home. These postmortems seem to be changing constantly though, they are just like a-holes, everyone has one.
https://fivethirtyeight.com/features/registered-voters-who-stayed-home-probably-cost-clinton-the-election/
There were three really close states any two of which would have swung the election. There are a million different things that legitimately cost Clinton the election, because she didn’t need much more to break her way.
I agree with you.
I’ll just ‘save’ the original post by saying: a perfectly calibrated predictor at 90% is still wrong 10% of the times.
That doesn’t really save the claim that “it’s not really clear that any of these forecasters got anything wrong.”
Making lots of properly calibrated 90% claims isn’t doing anything wrong.
It’s getting things 10% wrong, and occasionally 90% wrong. There is more to life than calibration!
I think that the election is clearly a sufficiently chaotic system that it makes sense to view it probabilistically in practice.
For example, as Scott pointed out in a previous post, the weather likely has a large effect on turnout, and I think everyone more-or-less agrees that the weather is a chaotic system that, though it might be purely deterministic, is impossible to predict with certainty in practice. The election is also affected by other chaotic systems, such as traffic, which people happen to be sick that day, the economy (including both long-term economic trends and extremely short term things like who’s working on that day), whatever news stories come out that day, etc.
So, just like I don’t say that the weatherman is wrong if he said there was a 10% chance of rain and then it rains—even though perhaps in theory some godly weatherman could have known that it was going to rain with certainty—I don’t say that election predictions were wrong because they said that Donald Trump had a 10% chance of winning and then he won.
Unfortunately, in the case of election forecasting, this leaves us with very little to go on to judge experts. Calibration is difficult in general, and with a small sample size, it’s harder still. Obviously in theory we should have some priors regarding the accuracy of various forecast models, and we can update those priors based on election results. This might be roughly approximated by saying that a model is “90% wrong” when something happens that it assigned a 10% probability to. But, since elections and election models are quite complicated, in practice our priors should be quite weak, and since we get so few data points, we don’t get to update our priors very much. So, I don’t really think we’re in a position to judge these models very harshly.
Edit: For what it’s worth, I bet a good amount of money on Trump winning at odds between 8:1 and 10:1 because I thought that the forecasters (other than 538) were likely underestimating his chance of winning. I still don’t think that it’s reasonable to conclude that they got it wrong based on this one sample.
Yeah this is the thing.
538 and Nate Silver said that the night before the election, by aggregating polls, they predicted Trump had a ~20% of winning. But the _REAL_ probabilities of Trump winning the night before the election were 99.99999%.
It is just by limiting themselves to averaging polls that they concluded that Trump was gonna lose except by a random stroke of luck. It may be the best we can do, but the fact that they got the election wrong indicates we can do better.
Agreggating polls is no longer enough, particularly not when the campaign themselves use them to decide their strategies.
Presidential polls tend to be right until they’re not. Every national poll in 1996 except Zogby badly overestimated how much Clinton would beat Dole by.
That made Zogby a lot of money for awhile, but he didn’t prove to have a magic touch after that.
Presumably, other pollsters figured out what they had done wrong in 1996 and took steps to fix it. Presumably, they’ll try to fix whatever they did wrong in 2016 as well.
If you aren’t in the polling, betting, or forecasting businesses, it’s probably not worth your time to develop an expert opinion on the current strengths and weaknesses of polling, because the professional pollsters will keep changing how they do it in response to events.
Polls are, by their nature, inaccurate. Pollsters practice “herding” (they’ll cluster their results near each other because they don’t want to be the outlier) and it’s actually quite hard to get a truly unbiased sample, particularly as response rates decline for traditional polling methods.
Whoa, whoa, whoa. You’re misreading the Moneyball story. In the beginning, they followed the models exactly and it didn’t help. Instead, Billy Beane combined human judgement with the models and then they succeeded.
And the A’s happened in 2002 to have a shortstop, who is barely mentioned in the book, who took a lot of PEDs and drove in 131 runs and was voted the league MVP.
They also consistently lost close games in the post-season. The book/movie also ignores the contribution of key players that were obtained based on traditional recruiting.
https://www.theatlantic.com/entertainment/archive/2011/09/the-many-problems-with-moneyball/245769/
I argued back in 2000 that it made more sense (due to the Electoral College) for the GOP to pursue northern blue collar whites than to try to win Hispanics with more immigration:
http://www.vdare.com/articles/gop-future-depends-on-winning-larger-share-of-the-white-vote
I’d say that analysis looks pretty good 16+ years later. But I’d still say it made a lot of sense even if Hillary had eked out a win.
The older I get, the more broadly I generalize this anti-heuristic-heuristic. How often do unsupported chains of reasoning ever result in a correct answer? If we’re talking about anything other than pure math, I mean. Is it much better than chance? You can aim this principle at business models, political theories, scientific theories, economics. We’re pretty crummy at reasoning, and as a society we tend to collectively lurch along through intermittent accidental success. Even the damn betting markets got Trump wrong.
Wish I had much more useful to say than “yeah, I agree,” but: yeah, I agree.
Especially in the current media environment, there are a lot of people who build these vast cathedrals of logic balanced on top of a single shaky premise, or more commonly even just wishful thinking. Then they somehow reason their way all the way back down to ground level to find some “hidden truth” which must exist to justify it all. Then when that “hidden truth” turns out not to be true, they go to pieces along with their logic.
I wonder if there is some way to make this more explicit and more embarrassing. There is currently no shorthand way of describing this phenomenon that doesn’t carry the wrong connotations.
For example, if you call somebody on “ivory tower reasoning”, you inevitably come off looking anti-intellectual. In fact, what you’re trying to communicate is that you’re pro-rigor, pro-evidence, pro-caution, anti-wild extrapolation.
There are certain domains where you can mention the Conjunction Fallacy but it will often not be clear how that applies. There’s still a big inferential gap if you say “Ayn Rand and Karl Marx are equally guilty of the Conjunction Fallacy.” People aren’t used to thinking in terms of the idea that each consecutive statement that a political philosopher makes is a prediction that can individually turn out to be right or wrong, and making a long chain of them is exactly susceptible to conjunctive errors. (You could say the same thing about most of the “big name” philosophers all the way back to Aristotle. If each statement is a prediction with some probability of being wrong, then long chains of logic immediately look terrible, and usually are terrible.)
The biggest problem of all is that this is pretty much the native mode of thinking for humans. We love building these highly compressed, fatally lossy models of reality. The same instinct that gave us Zeus as a model explaining lightning now gives us White Supremacy as an explanation for Trump.
We need a snappy name…
IfTIT (for If This Is True) Fallacy.
Or the Inverse Pyramid.
Basically, a pyramid is extremely strong because it has so much foundation. Inverting it creates a structure that extremely unstable because so much weight rests on so little foundation.
I don’t think it’s entirely fair to say that the AI progress model is based on just 2 data points. Instead, it’s a weighted average of the following:
1). AI winters
2). Every other prediction ever made by starry-eyed journalists since the advent of scientific method, based on some scientific discovery. Notable examples include “soon we will cure cancer”, “flying cars”, “fusion power”, and “soon we will cure cancer, for real this time”.
That’s a lot more than just 2 data points.
“There will be a world market for maybe 5 computers”
“Next up, polio.”
Apparently there’s some doubt as to whether Watson actually said it. That said though… if he did say it, he probably would’ve been right !
Well, sort of. There wasn’t much of a market for IBM-style massive mainframes. There still isn’t. Instead, there’s a massive market for lots of little computers, whose emergence Watson (either the real one or the apocryphal one) entirely failed to predict.
The market for IBM-style massive mainframes is larger than you think. There are lots of big financial institutions that still depend heavily on COBOL code written decades ago, running on IBM mainframes that are backwards compatible with the original System/360 from the 60s.
“Romney… and was “shellshocked” by his landslide loss.”
Not your words, but… landslide? Since when is 51.1% to 47.2% a landslide? Nixon versus McGovern in 1972 — 60.7% to 37.2% — was a landslide.
More reasonable than referring to Trump’s win that way, but yeah. He still got over 200 electoral votes, it was a decisive loss but far from a landslide.
OTOH, do we really expect “The Week” to bullshit us any less than Scott Adams?
Romney lost almost all the close states. Did the Democrats outsmart the Republicans? Was it just luck?
The interesting thing about 2012 was that Romney was getting pretty decent at campaigning toward the end and was drawing big, enthusiastic crowds. That’s what gave him hope. It’s not unreasonable to feel that if you appear to be building momentum in terms of crowd size and excitement and if you are targeting the close states, you could pull this thing off. That’s exactly what Trump did in 2016. But for Romney he wound up losing every battleground state except North Carolina.
The criticism of Hillary that she stopped campaigning in a couple of states she lost brings up the problem that there’s not much evidence that Hillary campaigning more in a state would boost her performance. Her rallies tended to be big downers for all concerned.
The one thing I get from post-mortems of Hillary’s campaign was that she was going for The Big Win and lost sight of the more important goal of Winning At All.
Part of it has to do with how we use stories, it reminds me of Andrew Gelman’s piece on stories (http://www.stat.columbia.edu/~gelman/research/published/storytelling.pdf).
Human brains are algorithms designed to process information like a causal story. The election of a leader, the night before the election, the polls, it’s all a story. The information lost is probably too weird for humans to understand. It’s different in a non-storied way. In the past I would have thought that’s just boilerplate nihilism, but after seeing the strange way neural-networks/autoencoders take thousands, or tens of thousands, of junk features we viewed as meaningless, and squeeze out substantial classifying power, I’ve become convinced it’s actually a salient feature of reality.
A worthy question: which algorithm is prone to interpret data as stories? Which algorithm feels from the inside the way we feel from the inside?
Hierarchical, nonparametric probabilistic program learning, according to many papers out these days.
Can you point me to some? I would be very interested.
Simpson’s Paradox is my favorite example of this. We are so eager to categorize and compress into simpler models that it continually blindsides us when this instinct leads us to simplify everything useful out of a model. A neural network would never fall for Simpson’s Paradox, because it’s not going to be eager to cut observations into bins prematurely.
Do you have a citation for the claim that NN are immune to Simpson’s paradox? I’m thinking that it would still depend on the existence of unmeasured variables data. Since you don’t necessarily get a causal diagram from a NN, you can’t condition out unless you have some prior information about the structure of the latent effects (for example).
Maybe the cleanest way to state it is that NN won’t succumb to Simpson’s Paradox any more than Linear Regression will succumb to Simpson’s Paradox — both methods are vulnerable if the human feeding data into the algorithm is biased, but by default both methods just do a model fit, and neither of them is going to commit the particular mistake that humans make.
But that’s because Simpson’s Paradox happens in the gathering of data, not the evaluation of data. Since NNs (and linear regression) deal with the latter, they obviously can’t cause the effect. But they can’t prevent it, either. Garbage in, garbage out holds for every method of analysis.
The way to avoid it is to always gather data about everything. But this has nothing to do with NNs, and everything with the raw computing power and memory you have at your disposal. In this vein, linear regression would be less vulnerable than NNs, since it probably uses less resources to crunch large amounts of numbers.
Worth noting that the heuristic being off regarding big-data is essentially what happened with Trump’s campaign vs Hillary’s. Her’s were full of focus groups and targeted attack ads, full of non reactive messaging to a presumed core demographic. Whereas trump’s campaign was literally A/B testing most of his content using direct feedback from his communities. They would even repost blogs retitled the next day with whatever meme’d title was most resonating at the time.
And by avoiding most of the normal ways of running campaigns, their scrappy team ended up being agile enough hone in on real value. As apposed to fighting against shitty platforms that were used more out of tradition than anything. As a result, at least from my perspective, Trump’s campaign ended up treating it like the popularity contest it really was instead of the noble endeavor our institutions pretend it to be.
There’s two good articles on the phenomena. One with an interview with Jared Kushner describing how they used startup “growth hacking” to disrupt, and another one that broke it down from an expert outsider’s point of view (note that it was published on 2016-11-09, before the interview, and still nailed the tactics).
It’s a classic “disruption” gambit. It was horrifying to watch unfold because I saw everyone vying against him saying how he’ll never win with the campaign team he had. They didn’t take him as a serious threat because the didn’t understand it. SMH
The thing is, it was a losing strategy. He won more or less by chance.
Had the election been held two weeks earlier, he would have lost. In fact, the only points at which he would have won (and I use won in the electoral college sense) were a few points in the campaign. Through most of it, he was quite a bit behind, but his support fluctuated.
Think about sports like soccer and football. Does the better team always win?
The answer is no. This surprises a lot of people, but it is so. A single game doesn’t really prove which team is better; a lot of factors can combine to cause a team to win or lose a game.
Margin of victory is slightly more informative – if you win by a large margin, you were much more likely to be better than if you only won by a small one.
Trump actually lost the popular vote by a wide margin, and only won the electoral college due to about 100,000 votes.
This, combined with the polling data, suggests that there were a lot of situations where Trump would have lost the election.
To put it bluntly: his win was basically luck.
Is this the bit where everyone points out that the popular vote is meaningless in an American presidential election, and so you can’t draw this conclusion based on it? In a popular vote-based Presidential contest, the sides would have campaigned differently. (Well, Clinton might not have…)
“Candidate X only won the electoral college due to [relatively tiny number of votes in close states]” is also an evergreen statement even in races that are not considered particularly close. A difference in something like 300,000 votes in FL, OH, and PA would have flipped the 2012 election to Mitt Romney.
As for the conclusion that “his win was basically luck,” well… you have to be in a position to get lucky. 100,000 votes that might have swung his way through “luck” wouldn’t have helped Trump if he had been more than 100,000 votes behind.
I question your equivalency here. Obama had a margin of over 500,000 votes across the three states that Romney would have needed to flip, while Trump had a margin of less than 100,000.
Ah, we’re talking about the same thing but slightly differently. Romney had a margin of 500,000 he needed to overcome, which could have been accomplished by flipping 250,000. And along the same lines Clinton’s margin of 100,000 could have been accomplished by flipping 50,000.
That being said, we’re still not talking about a fundamentally large number of people here. 250,000 is about the population of Lubbock, Texas.
On arguments of the form of “if Trump had just gotten X fewer votes in N states he would have lost, so the outcome was largely a matter of chance” …
If you imagine rerunning the election with random changes, you shouldn’t assume that the only changes are in states Trump won. It might turn out that Trump got a few fewer votes in one of the states he won and lost it, a few more in one of the states he lost and won it, still giving him a majority of electoral votes.
The other reason I don’t really buy the view that looks at the votes as “random” is that the swing states seem to swing as a group, instead of randomly distributing themselves between candidates.
Here’s a heuristic with a good track record: whoever wins Ohio wins the presidency. Indeed, Ohio was regarded as safely in the Trump column for a while before the election; the polls got that one right. The NY Times responded by writing a long article about how that heuristic no longer applies. It makes for interesting reading in retrospect.
That statement confuses correlation with causality/entailment. Winning the popular vote does not cause you, under the rules, to win the Electoral College. However, the two victories correlate strongly: 8/9 Presidents have won both, 8/9 presidential losers have lost both. Campaigning deliberately to win the Electoral College on a popular-vote loss has traditionally been an extraordinarily bad idea, and has only begun to really work now because population concentration (coupled to nobody rejiggering the number of Representatives to each state since 1911) has made it possible to win the Electoral College with decreasingly small portions of the popular vote.
If we look at the data we have, we saw a fluctation from approximately the event we saw in the end to a clear HRC victory in the polls. This happened repeatedly over the cycle. This suggests that the natural state was a HRC victory, but Trump had a chance of winning. The election happened at a low ebb that put Trump barely over the edge; a 100,000 vote difference in three states is quite small, and losing the popular vote means you very nearly lost the election.
The reality is that the margin of victory does indeed matter; if you see a game of football, and one team beats the other by two points, that doesn’t give you much information on which team was better. If one team beats the other team by 20 points, that gives you a lot more information.
If you looked at the polling, it was clear that this was a possible outcome, but it was “lucky” for Trump – it was pretty much the best he could reasonably hope for.
The reason that silver had it at 70/30 was the polling data. The polling average is off by +-2, and it had fluctuated by about +-4 over the course of a number of months. He had to have both his support at a high ebb and a polling error in his favor (i.e. underestimating his support) to win it. That was what happened.
But that doesn’t suggest his win was likely. In fact, all the data suggests otherwise. Had the election been held a couple weeks earlier, he would have lost – the margin of error on the polling data was simply insufficient to give him a victory. A lot of things had to break in his favor for him to win it.
We can never know for certain the true probability of him winning, but Silver’s prediction was entirely reasonable and was probably close to the true probability.
What is the sense in which you are using the term “true probability”? Which facts of the real world are you taking as given, and which are you allowing to vary, and what is the reference class you’re using for statistics on how the latter will vary?
@thetitaniumdragon
How can you know that there was no systematic polling error all along?
Because pretty much all of the recent elections where ‘populists’ did well had the polls predicting that they would do worse. This suggests a big error in the methodology.
@Aapje/thetitaniumdragon:
Has anyone come up with an explanation that goes beyond waving in the direction of “shy Tories” and mumbling something about pollsters being too concentrated in circles where right-wing populism is weird and alien?
I was adjusting Trump’s “chances” (leaving aside the whole argument of polling aggregates and whether or not they can be taken as a chance of winning) up by ten or so percent, due to seeing right-wing populists do better than expected in Brexit, German provincial elections, Austria, etc, and my gut feeling.
If I’d listened more to my gut, I might have called him to win. The one pundit who I saw make a “Trump wins” prediction that didn’t involve a landslide which didn’t happen, Michael Moore, was going largely off of his gut feeling, not data. I suppose, though, there’s an ego-protection element – if you’re wrong following the data along with everyone else, you can say “well, I was wrong, but so was everyone else!”
@dndnrsn
Moore was predicting by anecdote, as he comes the exact poor blue collar background that has been abandoning the Democrats more and more. That is not a proper methodology though and he was just right by chance.
My theory is that a major reason for the prediction gap is that anti-globalists are way more likely to lie to pollsters, due to a combination of shame, hatred of the establishment, general willingness to create chaos to create change and because many of the non-liars refuse to participate in polls (there are also liars among Hillary supporters and bremainers, but they are probably a smaller percentage of the poll respondents).
AFAIK, pollsters merely correct the data for underrepresented groups, by weighing the responses from certain demographics more, but don’t correct for differences in ‘lyingness.’
The systematic error is that polls use an assumption of each demographic’s base-chance to actually show up in an election (So for example, fewer young Democrats are considered in a midterm election poll) which, with regards to anti-globalist Republican voters, uses 20 years of ahistorically suppressed turnout to build the baseline.
I assume a Clinton win would have been just luck also?
A Clinton win by 100,000 would have been luck, because that is easily within the margin of variability caused by weather effects in different regions.
Of course, being in the game to only lose by that margin also requires stable non-luck effects; it is not like there is a dice roll that could have mad Jill Stein president.
You might be missing an ‘e’ there, but then again, you might not 🙂
If it was a 100,000 vote margin of victory? Yes.
Right, I’m not disagreeing. There was another post earlier on this blog that detailed exactly what you’re talking about regarding luck. More pointing out how the idea that it even came down to luck to the first place was due to using bad heuristics on how “successful” looks in modern campaigns.
I don’t think we can say this. Are you basing it on polls? But those polls also said he would have lost on the actual day of the election.
Someone here called the polls “epistemic garbage.” I think we need to grapple more with that possibility, instead of constantly turning back to the polls because they put numbers on things and we like numbers.
On the day of the election, the result was within the margin of error. Two weeks before the election, Clinton’s lead was larger than the margin of error. The polls were fine; it was the interpretation of the polls that failed.
The most likely answer is that the polls were off by a relatively constant amount. And that this constant, uhm, didn’t change much over the course of the race. So we can probably pick various times during the campaign when either candidate wins.
Do not forget: There were also times when Trump was leading Clinton in the national popular polls. (And then he would go pick a fight with the family of a dead veteran for some reason.) It wasn’t just November 8th when Trump wins, it’s a lot of other weeks, too.
In FL the HRC attack ads got really monotonous. The week before the election there were 5 political ads in a row during one sporting event. 1 for Trump, and 4 for HRC. 3 of HRC’s were the same exact attack ad repeated. Bizarre. I wonder how effective these things really are.
Political ads are kind of like internet ads to me, I don’t really even see them. My brain has a political ad blocker. The ads now are so obviously scripted and focus group tested to the hilt that they lose their authenticity.
There is some evidence that if you show an attack ad forty or so times the week before an election, about half the people who saw the ad will misremember it’s content as having been the headline story on the nightly news.
Possibly redundant in a campaign where the nightly news was showing the same attacks forty times a week.
I think you took the wrong lesson from the election.
The polls suggested it was fairly close. The election ended up fairly close. That’s why Nate Silver put it at only 70/30 for Clinton to win, and Clinton won the popular vote by a wide margin.
People who claim that Trump had some brilliant strategy is problematic – the reality is that the polls consistently showed him behind, but when he was close, he was within the margin of error of winning. The combination of Comey and the Russians’ efforts likely did far more for Trump than anything else, but he still lost the popular vote badly.
Had the election been held two weeks earlier, Trump would have lost pretty badly. Instead, he won just barely because the election was held close to his peak point of support.
Basically, his victory was luck. It could have very easily gone the other way, and indeed, the very small margin of victory illustrates that fact.
I agree with this – see post I linked. The point is that expecting to win despite polls b/c decreased minority turnout is not necessarily dumb.
Trump’s strategy is only brilliant in the sense that he ran an unconventional campaign that everyone thought would never work, and it did. Trump broke numerous accepted norms and won anyway. The DC establishment both red and blue, academia, Hollywood, and the media were all vehemently against him, and he won anyway. Trump has numerous “disqualifying” traits, had an embarrassing video come out, shot himself in the foot every other day, and still won. People were literally laughing at him when he announced, and most of the way they continued to do so, and he still won. He clearly understood something all these experts did not. He broke the mold and that is newsworthy beyond a narrow margin of victory.
The RNC, Fox News, and the propaganda arm of the Kremlin were all firmly on Trump’s side of the court, though.
The RNC wasn’t exactly enthusiastically for Trump, but was enthusiastically anti-Clinton for the most part. Putin apparently hated Clinton and preferred Trump but I’m not sure how much this really mattered. For narrow victories almost everything “mattered” so everyone gets to project their favorite bias upon the deciding factor.
It might be a more useful discussion to determine what would have been the easiest thing to change that would have swayed the election?
Clinton – no basket of deplorables, no private email server.
I think she could have survived the private email server thing (though Anthony Weiner seems to have been the gift that keeps on giving) because there was enough dull technical parsing of ‘what is the real difference between confidential and secret but not top secret etc.’ to bore the general public and make them ignore it in the end, but the “basket of deplorables” remark really hurt her. I think it switched just enough “eh, can’t stand either of them, probably will stay home” voters to “well, to hell with them anyway, I’m gonna vote!” to make a small but measurable where it counted difference. It was certainly a godsend to Trump’s campaign: even if she didn’t really call half the nation a bunch of don’t even deserve to be called Americans, it sure could be made to sound like that. (“Now, some of those folks — they are irredeemable, but thankfully they are not America” isn’t really a politically savvy thing to say about anyone; how would the LGBT for Hillary Gala have reacted to a Republican candidate breezily dismissing LGBT activists as “irredeemable”?)
I realise you have to suck up to your rich donors, but forgetting that this isn’t a private after-dinner speech circuit affair anymore, it’s in the middle of a major political campaign and people are going to want to know what your speech was about (including your supporters) so the media are going to report on it was a bad stumble by somebody running on “I’ve got the years of experience for the job”.
The smallest change that could have swayed the election is very likely Comey’s decision to release his letter. Here’s a summary of the evidence.
I find it difficult to blame Clinton for the “basket of deplorables” comment. If you look at her remarks in context, she is saying the opposite of what they are generally taken to mean. I keep seeing “basket of deplorables” deployed in the comments section here as evidence that Clinton didn’t care about the white working class — but the very next paragraph of her speech boils down to: “But there are also lots of Trump supporters who aren’t evil racist sexist homophobes. They’re just suffering and desperate, and are supporting Trump because they know something has to change, and he offers them hope. We need to reach out to them and empathize with them.” (Seriously, folks, go read it.) Clinton’s only real mistake was using the word “half” in a fuzzy rhetorical sense, which left her politically vulnerable.
If Trump had given that speech about black people, SSC commenters would be falling out of their chairs to deny that it was a demonstration of racism. (Even more so if we keep the comparison fair and postulate that he was just talking about Clinton voters broadly.) Why, then, do we keep throwing “basket of deplorables” out like it actually proves something?
And you would be accepting that denial, right? Not insisting that what Trump said actually was a demonstration of racism? After all, he did only refer to “half” of blacks as lazy, shiftless watermelon-stealers; the rest are merely pitiful wretches who need to be patronized by us.
No, Clinton’s mistake was far more than saying half (although saying half was in itself a poor choice of words). The entire speech you link wouldn’t change a typical Trump supporter’s vote if you broke it down, eliminated the “half” remark, and discussed it with them over tea.
The problem Clinton has/had with the white working class is that she has no idea (or demonstrated no idea which is different especially in a campaign” how they view themselves. Take the context that you quote
How many Trump supporters define themselves as “suffering and desperate”? Most have jobs and homes, the UE rate is under 5% and even with lower LFPR you can’t make the argument that most or even a significant amount of Trump voters are good and decent, but suffering/desperate people, and even if you could they don’t see themselves that way. Most Trump supporters (if you read their posts online at least) viewed themselves as under attack by various groups, as struggling against an enemy, not beset by natural trails. There are a wide number of culprits blamed, but the language used is combative, not conciliatory.
The right don’t, in general, view themselves as a group that needs saving, they need more weapons to fight with and want a figure that they believe fights with them, not a figure that is protective/compassionate/empathetic.
The term “deplorables” was clearly what got this particular speech noticed, and flung around, but it is the general tone and approach that locked Clinton out of the WWC, not that individual word.
@Iain:
You are right. Taken in context, she was not crapping all over poor and lower-middle class whites. Unfortunately, it was phrased in a clumsy way. It was less “bad opinion” than “bad politics”.
If she’d just said “most Trump supporters are good people, but they’re hurting, and they are being taken for a ride by someone who says he will help them, but can’t, and won’t” it would have come off a lot better.
Deeper for Clinton was the problem of presenting herself as someone who could fix that hurting, or really any hurting – it is hard for someone who is part of the establishment, the system, to say “I will fix the establishment, I will fix the system” because the obvious rejoinder is, “oh, you’re going to do that now?”
One thing that made it easier to spin what she was saying as being a shot at poor and lower-middle class whites is that some of her supporters definitely did and do that. Some people really have contempt for them, and are not afraid to express it.
EDIT: Another way to put it is that part of the “hurting” is less about material issues (although the spike in Oxy overdoses and the like is surely material) and more about feeling insulted. Nobody likes feeling insulted. Even if Clinton wasn’t insulting them … I know a lot of people (most of them white themselves) whose attitude towards any white person without a degree is at best one of condescension, and as I said, they don’t even try to hide it. It is rare to find someone so stupid, ignorant, or both, that they can’t tell they’re being insulted.
@Cerebral Paul Z: Accusing me of hypocrisy in a counterfactual hypothetical world proves very little. You may well be right that I would be a tribal hypocrite in that context; that doesn’t mean you should sink to my level.
@baconbacon: To be clear, the bit that you quoted was my quick paraphrase, not Clinton’s actual wording. Here’s the actual wording:
I don’t buy your claim that this is counterproductive material. I’m sure it wouldn’t change the typical Trump supporter’s vote, but that’s completely irrelevant. The typical Trump voter was never going to vote for Clinton no matter what she said. The important question is whether it would sway the marginal Trump voter in the Rust Belt — and it seems quite plausible to me that it could.
@dndnrsn: I absolutely understand why these remarks are easy to spin as demonstrating contempt for the white working class. But I’d like to think that SSC commenters are willing to hold themselves to a slightly higher intellectual standard.
“If Trump had given that speech about black people, SSC commenters would be falling out of their chairs to deny that it was a demonstration of racism.”
——————————–
“And you would be accepting that denial, right? Not insisting that what Trump said actually was a demonstration of racism? After all, he did only refer to “half” of blacks as lazy, shiftless watermelon-stealers; the rest are merely pitiful wretches who need to be patronized by us.”
——–
A liberal would not have accepted that denial initially. But they would be dog piled onto, and nitpicked to death, and would finally give up conversing in that thread, and would go look for another thread that is not yet dog piling onto them.
Bingo. My point of course is that your accusing Trump supporters of hypocrisy in an equally counterfactual hypothetical world proves precisely as much. The Argument From Hypothecrisy is inherently lame.
To go further into the “deplorables” incident you have to consider not the words, or the context of the words, but the context of the speech. Check out Iain’s link above, the first line is
She wasn’t addressing the WWC, she wasn’t working for votes among the undecided, she was standing in front of people who already supported her (in all likelihood to the tune of $1000 a plate just to get to listen to a speech in person and cut another large check in person afterwards) and saying what she thought would make them open their checkbooks.
The mistake was bringing up Trump supporters at all, once you decide to do that there is no parsing of language, no subtlety that will achieve the goal of getting more donations/energizing your base that won’t alienate those outside of it to some degree. It automatically turns the campaign from “I am Hillary Clinton and I will be a better president than Trump could ever dream of being” to “People who vote for me are better than people that vote for Trump.”
The deplorables comment probably didn’t hurt her any more than all of Trump’s “Lock her up” chants at his rallies and all of his lies. The Right Wing grabbed control of the media and the public narrative, with the help of Comey, Assange, Putin, and even the NYT. Hatred is not a losing proposition. It’s a winning proposition. Hating is one factor that helped DT to win. If hating kept you from winning, DT wouldn’t have won. HRC was never going to get Trump voters to change and vote for her anyway, no matter what she called them.
@Iain:
Oh, definitely. I don’t think Clinton is guilty of hating the white working class. She’s guilty of phrasing things badly, or one of her speechwriters was.
@Ian,
47% of people really didn’t pay (federal income) taxes either. It’s not very controversial to say that income tax credits or tax breaks for those people wouldn’t be very useful in getting their support.
I’m not arguing the proper nuances of interpretation here, only whether it had a material affect on the election and that it was easily avoidable. I read somewhere that she walked off the stage and told one of her aides that she knew she “had stepped in it”.
I am very, very, weary of the gotcha game the media has been playing for the past few decades in elections. Some illegal Mexican immigrants are actually rapists. There is lots of nuance one could put into that statement involving actual numbers and relative risk. I’ve heard a 1000 times now that Trump essentially said all Mexicans are rapists which is an irrational interpretation of this statement.
I very much doubt HRC believed what she said. It was corrected as “not half” later. (This was problematic when she should have just said nobody is deplorable or irredeemable). Some in the media tried to defend the half statement which was a bit ugly.
People say dumb things, give them a chance to correct it, and let that be that. I don’t care what Trump said 20 years ago or what Ellison said about the Nation of Islam if they don’t hold those views today. They deserve some points off for bad judgment but media driven automatic disqualification rules and lifelong purity tests have gotten out of hand.
Semantics, but the easiest way to change that was to never have set the server up in the first place.
“Semantics, but the easiest way to change that was to never have set the server up in the first place.”
Yes, HRC should not have made small mistakes, and she did, and people claim that that is why she lost. But DT was given a free pass for doing tons of corrupt and fraudulent things e.g. for Trump university, for failing to pay lots of contractors, for rape and bragging about it etc. etc.
@ Iain
Generically politicians have to balance 3 things. First is getting their base energized, they need people that already support them to donate money and time and to actually vote and encourage others to vote. Second pull in more undecided voters than you push away, and Third don’t excessively energize the opposition’s base.
Can you construct a speech about the WWC that makes Hillary more appealing to them? Sure, but you have to trade off one of the other pillars to do so. The deplorables comment that is a “mistake” in retrospect was actually liked by a large portion of her base, that is what they think about Trump supporters and what they want to hear, and the us vs them campaign helped raise huge amounts of funds for Clinton.
The trouble is that you can’t elevate the opposition above your side, so you can’t write a speech that is earnestly complimentary, it has to contain language that sets you and your supporters above them or you are sapping your own energy while trying to grab those votes.
@Cerebral Paul Z: The difference is that I could pull up any number of comments on SSC that parse Trump’s comments with that level of charity. (Start with “You Are Still Crying Wolf”.) I have not seen nearly the same response to equivalently uncharitable readings of Clinton’s remarks. I am therefore providing that response myself, in the hopes that it may trigger some introspection.
@baconbacon: An equivalent standard, applied to Trump, would have prevented him from ever talking about Mexicans or inner-city black people. More importantly, you’ve already conceded the only point I care about in this debate: regardless of the strategic wisdom of Clinton’s remarks, they are not evidence that she has contempt for the white working class, so SSC commenters should stop treating them that way.
@tscharf: I think the controversy about Romney’s comments was less about his argument that tax credits won’t sway people who don’t pay federal taxes, and more about the idea that people who don’t pay federal income tax don’t “take personal responsibility and care for their lives”. My intuition is that Romney’s gaffe seems worse, because it is easier to think “Well, I’m not a racist, so I must be in the other basket” than it is to convince yourself that you really do pay federal income tax. I agree, though, that they were both politically unwise comments. My main point, as I said above, is simply that Clinton’s comments in particular should not be accepted as evidence of actual antipathy towards the white working class.
Semantics, but one of the two actions was an unprecedented breach of convention, and it wasn’t the lax email security. I don’t really want to get back into that debate, though, so I’ll grant you your point and we can move on.
If you look at her remarks in context, she is saying the opposite of what they are generally taken to mean.
Iain, that’s part of her problem – she has no talent for rhetoric. She wanted a snazzy soundbite and unfortunately for her, she got what she wanted.
It’s no good having the “but some of the poor morons are actually bearable because if we train them right they may be useful” part after your zinger, because people will remember and quote the zinger and not the rambling qualification after it.
She (or her speechwriter) set up the “you know what I mean?” punchline for the nice rich gays to laugh at first, and it worked. They all chortled at the notion of the deplorable rednecks. I mean, yes, those kind of people, the very notion!
Unluckily, the deplorable rednecks saw the clip all over the media, and all Trump’s campaign had to do was hammer away at the “out of touch coastal elites laughing at you”. It was handed to them on a plate.
And the people in Hillary’s second basket didn’t hear, or didn’t believe the condescending pat on the head afterthought; they felt that they, too, were being laughed at – because they were working-class, because they lived in rural or semi-rural areas, because they’d never have a few thousand to drop on a dinner to donate to a politician’s campaign.
And just maybe, if they weren’t going to vote in the first place, they decided that to hell with it, if all they were good was to be fodder for their betters to laugh at, they’d go vote for the guy who wasn’t using them as punchlines in his begging speeches.
Perhaps not, but she does have too much of the air of Lady Bountiful descending on the cottages of the tenantry with a basket of hand-me-down clothing and improving pamphlets about her. She and her team like to make mention of her Methodist background (“You know, my family and my faith taught me a simple credo — do all the good you can, in all the ways you can, for all the people you can,” she said) when it’s deemed appropriate, and she does have the air of the Nonconformist Conscience about her.
That’s not necessarily a compliment.
Bill, on the other hand, can successfully get away with it because he does come from those roots, even if he’s risen above them. He still has the instincts about how to talk to the rednecks. Hillary is more like Margaret Thatcher – the grocer’s daughter who strove to advance herself but who was always painfully aware that she was lower middle-class, thank you very much, and not working-class. These minute differences mean a lot when you’re trying to lift yourself up by your bootstraps into a higher social sphere, as Hyacinth Bucket could tell you.
Iain, a charitable interpretation of Trump’s remarks is still nasty towards Mexican illegal immigrants (despite his “and some of them, I assume, are good people”). The uncharitable interpretations extend this to all immigrants from Mexico or all Mexicans period. A charitable interpretation of Clinton’s remarks is still nasty to Trump supporters. There’s no double standard here; Clinton just stepped in it.
@Deiseach: I don’t know how many times I have to repeat myself. I am not arguing that Clinton didn’t screw up and open herself up to being misrepresented and pilloried by Trump. I am arguing that this corner of the internet should be smart enough not to take that misrepresentation as gospel truth.
@The Nybbler: Uncharitable to some of Trump’s voters, absolutely. Uncharitable to the white working class? Only if you want to argue that the white working class is racist, sexist, homophobic, xenophobic, and Islamophobic — at which point maybe you should reconsider who is demonstrating charity to whom.
For the umpteenth time: I am making a very specific point about the way that “basket of deplorables” is misused on this website.
Just for an example this is a quote from the mexicans are rapists by Trump
The equivalent quote would be something like
Trump didn’t attack Democratic voters, the worst you can infer is that he is saying specific things about a specific subset of democratic voters, and there is a world of difference between the two (even just quantifiably, there is enough space between “border jumping rapists” and “all Hillary supporters” that you aren’t going to rile that many people up.
The issue comes back around to this: How can Hillary bring up Trump supporters in a positive light while also saying nasty things about Trump within a few lines of the speech either way, while also generating enthusiasm from her supporters?
I conceded no such thing, I don’t particularly care, but I the fact is that she decided in a pre planned fundraising speech to attack Trump’s supporters as a group. At best it was a mistake, I certainly don’t see how a Trump supporter should see this as anything other than contempt.
There is no question Trump was held to different standards than Clinton. If the Clinton standards were the law, Trump would have gotten crushed like a grape. Part of the election was a revolt against said standards and who gets to make them (identity politics, political correctness, blah blah) . I think the media effectively held Trump to Clinton standards and reported every violation. In the end the voter decides and many of them did give him a pass for obnoxious behavior.
She wasn’t addressing the WWC, she wasn’t working for votes among the undecided, she was standing in front of people who already supported her (in all likelihood to the tune of $1000 a plate just to get to listen to a speech in person and cut another large check in person afterwards) and saying what she thought would make them open their checkbooks.
Which was precisely the problem. She was preaching to the choir. She likes talking to her natural constituency, and for all her “I stand with you” chatter, that’s not the dispossessed of the earth.
Of course she trotted out lines that would appeal to her donors. That’s the very point: she used a section of the population as a punchline for the rich coastal elite (or those who could be painted as such) to laugh at. Not laugh at a joke, laugh at them.
Which is fine if it’s the usual private speaking gig she was accustomed to, but it wasn’t. It was a fundraiser during a presidential campaign, which does mean media coverage and publicity, and video of her waiting at the appropriate moments after her zinger to let the audience laugh. And the media, even those sympathetic to her (do you think the Guardian is Trump-friendly?) of course only quoted the soundbite, not the whole “Then again, the rest of his supporters go into a second basket” part because the media like attention-grabbing headlines.
Even the deplorables, non-deplorables but adjacent in the second basket, and possibly not deplorable at all but still not living in San Francisco as tech entrepreneurs could see this, and how do you think they liked it? Not as much as the guffawing rich white guys and gals with thousands to throw into the donation basket. That was the mistake: not making the remark (though I think it was a bad remark to make because never alienate anyone with a vote), but forgetting that it wasn’t going to remain within the little circle of gala guests to chuckle about.
You guys are twisting yourself into pretzels trying to argue that you’re not more charitable to Trump than you are to Clinton (or that if you are, it’s justified).
Just admit it and let’s move on already.
I don’t think that is how most people respond to the deplorables statement, it is closer to “I’m not a racist, but she is calling me one and that is offensive”. Because racist is so loosely defined now it is impossible to determine who a person is including in generalized statements like this, it is different for different speakers. I have no idea who HRC would include in that basket or how big her “not half” basket really is. It would be useful if that definition was tightened up.
What posters are included in this generalization?
The moral is that you’re going to confuse people if you say “I find it difficult to blame Clinton” when you actually mean “If I were to extend to Clinton a level of interpretive charity which other people are willing to extend to Trump but I am not, I would find it difficult to blame Clinton”.
ETA: I suspect that among WWC voters who heard the entire quote, the most common reaction was “I’m not a pathetic loser, so she must be calling me deplorable.”
A cabinet secretary deliberately concealing her official communications from Congressional investigators is not a “small mistake.” It is in fact far worse than some random carelessness with classified information, as it demonstrates deliberate contempt for the separation of powers, congressional oversight, and one’s ultimate responsibility to the people of the nation one theoretically works for.
Now, you might assert that Congress is full of idiots. Doesn’t matter. Those are the duly elected idiots, and they get to see whatever communications they want to see right away. Period, the end.
I know I shouldn’t feed trolls, but… seriously. Anyone who has ever held a security clearance will pretty much be lit on fire by the attempts to characterize Hillary’s email server as “a technical mistake” or something “everyone does in Washington”. Any rank and file government employee who used private email for classified public business would be at a minimum facing permanent revocation of their clearance, termination of employment and probation or actual jail time for mishandling classified information. More likely, they would face espionage charges just to be on the safe side.
Of all of the lies told during the election, this was the one that got under my skin the most. It’s possible to make an honest argument that Trump’s volatile personality and rank dishonesty were so terrible that it was preferable to have Clinton in charge, even if she should have been in jail. (And I have friends in DC who did make this argument, and I respect that.) But claiming that she did nothing wrong and received no special treatment is impossible without being either staggeringly ignorant of federal security procedures or just manifestly dishonest and willing to say anything to further the interests of the Party.
@Cypren: How do you distinguish between Clinton’s server and Colin Powell’s use of a personal AOL account?
Last time we had this argument, I looked into the report from the FBI investigation. Clinton’s server was not being used to discuss classified material on a regular basis. Classified discussions took place on a different system. There were a handful of cases where the investigators retroactively decided that certain emails should have been considered classified. Nothing in that list seemed (to me) like a smoking gun. For example, several of the email chains took place around Christmas, at a point when many State Department officials didn’t have access to their more secure devices. Confronted with an urgent issue that required immediate discussion, they attempted to “talk around” the classified material. The investigators came along later and decided that their efforts were insufficiently successful.
Moreover, of all the email chains that were determined to contain classified information, none of them originated with Clinton. Clinton didn’t personally send classified information via email; at worst, other officials with security clearances sent classified information to her personal email account, and she replied to or forwarded those messages. (Discussing classified information using a state.gov account — rather than more secure channels — is just as prohibited as using a personal account, so the private server is a bit of a red herring here.)
Certainly, Clinton should not have used the private server. Hiring a firm to delete her “non-work-related” emails was also a clear mistake. I think it’s unlikely that they deleted anything particularly noteworthy, because the FBI would still have been able to see the other end of the conversation on State Department servers, but it definitely didn’t look good. I’ve also seen claims of her (or her underlings) lying about various things during the investigation. I haven’t looked into those claims in any depth, so I’m prepared to accept them for the sake of argument. So yes, Clinton fucked up pretty badly with her emails — but the accusations that she sent classified information appear to be the weakest part of the case against her.
(If you want to see the previous conversation, Cypren, I can likely dig it up for you. I’m not particularly interested in having the whole argument again, though, so if this subthread explodes then I am likely to peace out.)
@ian
>Cypren: How do you distinguish between Clinton’s server and Colin Powell’s use of a personal AOL account?
Not Cypren, but in brief, the rules were different back then, Powell did it with the approval of the State Department it/security people, and no one as alleged he used it for classified material.
>There were a handful of cases where the investigators retroactively decided that certain emails should have been considered classified.
this is directly contradicted by comey’s statements. to quote:
It’s certainly true that most of the material was later upclassified, but 2000 emails is decidedly more than a handful, and more than 100 were not. And, of course, that is only the material that clinton gave to the FBI, not the tens of thousands of emails that she deleted while under subpoena.
>Moreover, of all the email chains that were determined to contain classified information, none of them originated with Clinton. Clinton didn’t personally send classified information via email; at worst, other officials with security clearances sent classified information to her personal email account, and she replied to or forwarded those messages.
She set up the server and told her underlings to email her there. They should know whether or not the system they are using is secure, but that they screwed up doesn’t absolve her of responsibility. She also has a positive duty to report information that is not in a secure location.
>(Discussing classified information using a state.gov account — rather than more secure channels — is just as prohibited as using a personal account, so the private server is a bit of a red herring here.)
the private server was even less secure that state.gov, and was used to violate federal record keeping laws.
>Certainly, Clinton should not have used the private server. Hiring a firm to delete her “non-work-related” emails was also a clear mistake.
That mistake is obstruction of justice if you or I do it.
>I think it’s unlikely that they deleted anything particularly noteworthy, because the FBI would still have been able to see the other end of the conversation on State Department servers, but it definitely didn’t look good.
Not if the conversations were between the many (dozens, I would think? I’m actually not sure, I should know that.) of people who had email accounts on that server.
>I’ve also seen claims of her (or her underlings) lying about various things during the investigation. I haven’t looked into those claims in any depth, so I’m prepared to accept them for the sake of argument.
Clinton has claimed, at various times, that the state department had calculated that 90% of her email was preserved in their system, that she set up the server so server so she only had to carry one phone, that she willingly turned over tens of thousands of emails to the state department without being asked, that there was zero classified email on her server, that there was classified material but it was all up classified material, and that there was classified material that wasn’t up classified, but none was marked as such, and that while she might classified material she certainly never sent any one any.
All of these claims are false.
>So yes, Clinton fucked up pretty badly with her emails — but the accusations that she sent classified information appear to be the weakest part of the case against her.
It’s not the sending of the material that’s an issue, it’s the whole thing. Classification rules are very strict, and by the standards of the US government, strictly enforced. Everyone knows to expect different spanks for different ranks, that unavoidable but, as Cypren puts it, this is of a different order altogether. Petraus was a 4 star general who showed classified material to someone with a clearance that wasn’t sufficiently high, Sandy Berger was a former national security advisor who took a few documents out of the national archives. That’s almost as high as you can get, but they both had to take their lumps, plead guilty, and give up their clearances.
Now, is that more than a slap on the wrist? no, but at least they got that slap. They at least had to admit publicly that they fucked up and deserved to be punished for it, to have their careers terminated. Clinton broke the law, then broke more laws trying to cover up how she broke the law. Her steadfast insistence that this is all just some witch hunt is infuriating.
@Iain: Cassander already made most of the points I would have discussed. The only point I would really add is that saying “there’s no smoking gun” is itself a red herring in this case because she destroyed most of the evidence. (Or to be precisely accurate, one of her contracted employees claimed to have spontaneously decided to destroy the evidence immediately after a subpoena was issued for it but before he was notified of such and there wasn’t a clearly provable written trail showing that she ordered it, because she is presumably not a complete drooling moron.) This is not an argument most people would accept as proof of innocence in a criminal case, and I sincerely doubt it’s one that people making it would accept if it were used to defend one of their political opponents.
That’s why the relevant standards for handling classified information don’t consider intent or quantity of information: intent is extremely hard to prove and it’s rare that someone faced with federal charges isn’t going to make some attempt to conceal or destroy evidence.
My point isn’t that what Hillary did was some irredeemably evil cackling super-villain plot. What was publicly acknowledged was pretty innocuous, and there’s no way to know what was in the destroyed emails. Critics of Clinton will assume that it was damning evidence of [insert conspiracy theory here] and supporters will assume it was nothing but sweet grandmotherly discussions of tea and cookies. My point is that the treatment she got — handing out immunity to all of her subordinates who provably broke the law, not forcing her to testify under oath, having the entire DoJ side of the investigation run by a longtime crony of her chief of staff… all of this just emphasizes that Washington has a different set of rules for the powerful, and Clinton supporters are okay with this when it benefits their woman.
That’s what makes me so mad.
If it was sweet grandmotherly discussion of tea and cookies… why was it destroyed? As a public figure, you don’t destroy evidence during a criminal investigation unless it will damage your image more than the destruction of evidence during a criminal investigation would.
It is, I think, plausible that Hillary’s semi-paranoia (they really are out to get her, but her response is diametrically opposed to reason) and demand for control (no mere civil servant will look through my private affairs!) led her to destroy files that contained no evidence of criminal wrongdoing or severe impropriety, merely an ordinary level of dirty laundry that nobody would want outsiders rooting through.
If so, that was one of the more impressive own goals in US political history.
Hasn’t it been a recurring pattern with the Clintons that their wrongdoing is usually fairly minor, and it’s their attempts to cover it up that get them in trouble?
That is the charitable way to look at it, the uncharitable way is that they are great at covering up their tracks and we only hear about the handful of minor incidents because they get rid of all the damning stuff.
@baconbacon
>That is the charitable way to look at it, the uncharitable way is that they are great at covering up their tracks and we only hear about the handful of minor incidents because they get rid of all the damning stuff.
I’d argue that the uncharitable version is closer to Al Capone going to jail for tax evasion. Al Capone murdered a lot of people, but it was easier to prove that he didn’t pay his taxes, so that’s what they got him on. The clintons, so the story goes, do a lot of shady shit, but are great at wriggling out of/excusing/obfuscating their wrongdoing, so they only get really nailed on the few cases where there was incontrovertible proof of wrongdoing, like a hard drive full of emails or a stained dress.
“He clearly understood something all these experts did not.”
No.
If you do the right thing for the wrong reasons, are you brilliant or lucky?
How clear is this? It seems to me there are plenty of explanations for why his campaign could work in the face of negative expert opinion that do not rely on him actually knowing anything that the experts didn’t know.
Certainly could be a lucky guess. He may have picked up valuable insights on reality television in the sense he knew better than a DC politician what resonated with the public. Reality TV is full of manufactured conflict that people apparently like, and there is no doubt he plays that game well. The real question is whether winning election skills carry over to governing. This is not obvious at all.
Well yeah, that’s the question but it’s not a question that’s unique to Trump.
In fact, it applies much more strongly to the progressive candidates who are all using the “import as many genetically low IQ government dependents who will vote purely on the issue of keeping the payments going (and who will have kids who do the same)” strategy which has the long term effect of utterly destroying the country. Basically this means that the long term of “winning elections” is antithetical to good governance.
Progressives are making no attempt to import more Republicans from West Virginia into this country. Where would they even get them from?
Luckily for the Democrats they have Africa, Mexico, the Horn of Africa and the Middle East all of which have genetically lower IQ, less cooperative, more violent people than those in West Virginia (which still average out to about 100 IQ or so) to import.
The problem was that he ran an unconventional campaign but the Dems did not counter and crush him. When someone in Chess does a weird, bad opening, you don’t let them continue to see what happens. You just punish their error immediately, win material advantage and stop their advance. The Clinton campaign just let him run his strategy unopposed.
Politics is not like chess.
I misread that at first as “Politics is not like cheese” 🙂
This seems related to Tetlock’s fox/hedgehog dichotomy.
Minor errata: The 2012 polls were off by about the same amount as those in 2016. The main difference is that in 2012 the polls were off in Romney’s favour (and also that the individual state errors were closer).
I don’t think N is really 2 in this case.
I was also somewhat surprised that go was solved when it was, that and self driving cars were both “solved” very quickly following the discovery of a few new tricks but they’re still looking like they’re following the same pattern as always.
typical pattern:
Step one: new trick is discovered solving some problem X which couldn’t be handled before.
Step two: people try to apply it to everything that the old styles didn’t work on like problem Y which is sort of in the same problem class. At this stage overly enthusiastic people may over-promise. “I’m sure it will work amazingly on Y”
Step three: “Bah! These CS types never deliver, Y will always be better done by humans.”
Step four: Interest and funding flees as the news stops paying attention, a few people keep chipping away at the problem and eventually slightly outperform humans on Y and try to get it to work on Z.
Step five: Someone proves mathematically that it can never solve major set of problems in Z.
Step six: Someone comes up a new trick… GOTO 1
This pattern I’m pretty sure has an N of far more than 2 and is pretty consistent in almost all areas of algorithms and CS.
AI development is like an adventure game? Every time we discover some new technique it gets applied to every problem and every other technique to see if it does something interesting.
I don’t think this changes the conclusion. At some iteration of your cycle, the “new trick” will actually be the magical formula underlying true general intelligence, and your prediction that it’s just more hype will be wrong. In fact, your confidence that each subsequent cycle will be the last cycle should increase over time, as technology moves closer to that endpoint. But in your model, each subsequent cycle convinces you further that the last cycle will never happen.
I can’t find the link, but this is reminiscent of Yudkowsky’s line about the fallacy of reasoning that you can pull out a few more neutron inhibitor rods because pulling out the last few didn’t cause a meltdown.
There’s no real reason that general intelligence should be a cheaper, quicker way to solve any given problem than a well-chosen narrow trick. In fact, the nasty part of general intelligence is just how expensive it is to construct a system that can learn many well-chosen narrow tricks and apply them when relevant all on its own.
This assumes that every innovation is driven purely by a targeted attempt to find something cheaper and quicker. Researchers within academia and corporations often pursue “blue sky” goals with minimal short-term return.
And also, the “final trick” may just be stapling together a handful of narrow paradigms in such a way that the gestalt becomes generally intelligent, in much the same way that each individual area of the human brain is relatively incompetent without the whole architecture.
Not if there isn’t a magical formula underlying true general intelligence. If in fact ‘general intelligence’ is not one clear discrete thing but a huge number of small abilities aggregated together, we might well continue to see lots of steadily more capable intermediates rather than a bright line between ‘narrow AI’ and ‘general AI’.
I actually think it’s very unlikely that we will recognize any bright line between narrow and general AI, even in hindsight after we’ve obtained general AI. The “last trick” may be something as mundane as glomming together supposedly “narrow” algorithms in such a way that the overall system suddenly becomes more powerful. This is roughly analogous to many of the recent breakthroughs in deep learning where simply adding some particular type of regularizing layer, or a random dropout parameter, leads to a discontinuous leap in the power of a deep neural network.
The current state of the art in machine learning is actually pretty well described by slapping together “narrow AI tricks” and getting progressively more mileage out of novel combinations.
Or there might not be a last trick that produces a huge increase in power. Especially if we are considering the “intelligence is just a big pile of small tricks” scenario, I can’t see why you would expect there will be such a sudden final jump at all.
If there is a day where we don’t have AGI and that day is followed by a day where we do have AGI, then something happened. That something may have been a “big innovation” or a “small tweak”. It may be a small tweak that doesn’t cause a “huge increase in power” but rather nudges a certain capability over a certain threshold that enables self-modification (for example).
In any case, none of this impinges on my initial statement that each subsequent example of a new innovation not immediately leading to AGI should not cause you to conclude that no new innovation will ever lead to AGI.
A fetus does not possess AGI (or GI, I suppose, in this context). An adult human being presumably does. Can you point to the precise day where your GI appeared?
It is not necessarily the case that there is a sharp boundary, beyond which everything is clearly AGI.
I agree. I don’t really care about these distinctions. My only point was that “we don’t have AGI yet” is not evidence that we never will.
There’s also uncertainty in the number of cycles.
This seems like it might be one of those cases where the number of future cycles you should expect = the number of cycles that have happened so far.
It could be that there’s some single trick…. but I think it’s illuminating to look at the ways that mundane human intelligence goes wrong. Look at severely autistic children, idiot savants and kids with various mental disabilities. In terms of the “general intelligence” toolset they have pretty much all of it but what would probably count as fairly small deviations from the norm that can utterly cripple them as general intelligence’s.
Which sort of points to there being a great many things which have to be just right rather than one neat trick or algorithm around which everything else will default to coming neatly into line.
Perhaps the heuristic should have been “minority turnout will be high when you run a black guy, because obviously” but no one, Romney apparently included, was comfortable admitting to that conclusion?
Even that heuristic is debatable. Will “when you run a black guy” always have the same turnout power as “when you run the first black guy in history who has a shot at winning and thereby breaking through the color barrier of the position”? Is Larry Doby as famous and inspiring as Jackie Robinson?
Although, both of these heuristics are probably just overfitting. A priori I would have put my money on “when you run the first of any socially-defined underdog class in history who has a shot at winning and thereby breaking through the corresponding barrier of the position”, but in fact neither the gender gap nor the woman voter turnout in 2016 seemed remarkable.
Regarding your second paragraph – perhaps a fresher female candidate would have gotten the bump.
I heard a recent (pre-election) Lewis Black bit to the effect of “Hillary’s problem is that she never went away.”
Barack Obama, even outside of his position as “first black guy with a real shot at winning the presidency” was fresh, new, exciting. None of those are words you can apply to Hillary, who’s been at the top of American politics for the better part of 3 decades at this point. And turnout does have a lot to do with excitement.
Maybe “exciting candidates will boost turnout among the groups they excite” is a good heuristic (although it’s also a bit of a tautology). In that sense, Romney underestimated how exciting “first black president” would still be after four years, and Hillary overestimated how exciting “first old white woman president” would be among minorities (and/or underestimated how exciting Trump was, but he doesn’t seem to have boosted turnout). Neither of those invalidate the heuristic, they just mean it was misapplied.
HRC’s problem is that Right Wing media, and Right Wing columnists and pundits within Left Wing and Center media, had been bashing the Clinton family for 40 years. Part of Obama’s secret to winning was that he hadn’t been a very public figure for long beforehand, so media had not had much time to bash him repeatedly.
I think the average woman identifies less strongly with women-as-a-group than people of minority races identify with that race-as-a-group. I mean, it’s always there in your life and colors a lot of things a bit one way or the other, but there’s not the same level of solidarity.
If you’re black, most likely your whole family is black, or at least most of them, and especially in neighborhoods that are de facto segregated, most of your neighbors and associates are. Racial issues affect almost everything and are often a serious factor in one’s life. But if you’re a woman, well, statistically speaking about half your family members are male, and half your neighbors, and a lot of your co-workers and customers etc. too although that percentage varies dramatically depending on what your job is. You’re not surrounded by all/mostly women and all experiencing the effects together; you also see a lot of men, and the differences beyond the physical between them and you aren’t extreme. Life outcomes are similar, especially if they’re in your family. It turns out that gender doesn’t matter all that much… it does have some effect, but if you’re being held back by something in a really noticeable way, that’s rarely it. Poverty, race, disability, etc. make a much bigger difference.
And most women understand this on some level, and get less pumped up about the possibility of a woman President/etc. than someone of a minority might about the possibility of a President of their race. I mean, I voted for Hillary, but I wasn’t excited about it. It was more like “ugh, they both stink, but at least Clinton acts like an adult and I trust her not to screw everything up too badly for the 2020 winner to fix.” Trump was too much of a wild card for my tastes and more of what he said rubbed me the wrong way, even if Clinton’s apparent disdain for the working class chafed too. The fact that she and I share a gender didn’t matter much; at most that would be a tiebreaker factor. We still have little in common – different social class and background, different education, different generation, even sharing a gender matters little when it was shaped so differently. I could see this being a different case for people who share a minority race and really DO share at least some of the same issues and concerns.
Seems like people are applying a similar heuristic to the AI Winter example when it comes to predictions of job losses via automation. There is a common refrain that often in the past people have predicted new machines and computers to lead to job losses and they haven’t, so you shouldn’t expect the next 10/20 years to be any different. But if you look at the number of major disruptions to work styles, there really seems to be a very small number on which to base such a heuristic. The Luddites, the steam engine, Ford’s production lines, early computers seem to be the biggest, which is too small a sample to make such assertive claims for the future.
Of course it does depend to an extent on how you measure this. You don’t have to look at major events and could instead just go by years, saying ‘since the invention of the steam engine/computer, there have been x years of job growth greater than losses due to automation’. Such an argument then runs into the problem that you cannot just extrapolate forward, but at least has a large enough sample.
“But if you look at the number of major disruptions to work styles, there really seems to be a very small number on which to base such a heuristic. ”
Literally every labor-saving device ever invented could theoretically be counted in this though. If you define the n specifically enough, it could very well be into the thousands.
I think if you used all of human history then you would in fact run into a lot of cases where labour saving devices did in fact cause absolute job losses, as when a leader decides they no longer need so many slaves so they simply slaughter them all so as not to have to feed them.
Well now you’ve shifted the entire premise of the debate away from “sure X never led to Y in the past but the N is small so who cares?” to “actually there are plenty of times where X did lead to Y” which is a different argument entirely.
I’ll just point out that I have yet to hear any serious critics of automation make the argument you are suggesting – that actually plenty of labor saving devices have, in the past, led to massive job losses as evidenced by the wholesale slaughter of slaves following the invention of the plow or whatever.
I think even in the modern age we’re not looking at job loss correctly. We have the phenomenon of vast numbers of office workers who spend hours a day on Facebook and Slate Star Codex. Obviously the productivity-per-actual-hour-worked is extremely high and a lot of human energy is being wasted for social reasons. The employment numbers might look different if our culture hadn’t morphed to accommodate the reduced demands on our actual labor.
Fire, stone tools, bronze tools, domestication of animals, domestication of plants (you cold probably put major individual animals and plants as their own categories), dozens of different advances in irrigation, fertilization, transport.
Life was crazily different from the past, and it isn’t in a “things were like this, then the printing press came along and then they were like that”.
I would argue that AI hype is in fact vastly out of proportion with the public’s perception of what AI can currently do, and so that heuristic is still in fine shape. Most people I talk to believe that we are currently in possession of, or will be in the next several years, a human-like general AI, and that’s not remotely true.
[emphasis mine]
Who are these people you talk to?!
Seconding this question. I am really curious, because this is not true in my social circles.
Isn’t this what MIRI/CFAR generally believe ?
MIRI thinks that AGI research is an extremely risky pursuit, but I don’t get the impression that it thinks we have AGI now or will in the very near future. The last time I saw estimates from MIRI staff on when we’re likely to have it, they were spread out between ~15 and ~200 years from now with the average being around 50. That was a couple years ago, though.
aw hell yeah, keep commenting here, thanks in advance
I’m not sure the comparison to moneyball is a great one. Back in those days, there were plenty of people in baseball who would loudly and publicly mock the use of statistics, openly declare that they ignored such things, and that their intuition was totally superior.
I didn’t follow the election that closely, but I suspect nobody on the Trump campaign ever did this. As far as I know, they claim to have used data as well, just maybe different data and maybe different assumptions about the data. I don’t think they came out and said “who needs your fancy data nonsense? our candidate is a master wizard!”
There may be various reasons to believe that the Hillary campaign was more “data-driven” than the Trump campaign, but I think the Trump campaign would probably dispute that. I’d also suggest one possible reason this assumption exists is to further a partisan narrative wherein Republicans are ignorant manipulators who only win by spreading fear, propaganda, and non-genuine news; whereas Democrats appeal to logic and reason and would clearly never lose an election if only the ignorant masses would act rationally.
I would suggest that the primary reason to believe it is that the vast majority of people with concrete experience in “big data” analytics are in the college-educated coastal elite demographic which forms the strongest core of the Democratic base.
Unlike a lot of professions, data science isn’t a field where the value of experience at the #1 player in the world (Google) and the #50 are equivalent (or even within several orders of magnitude) due to the power of aggregation effects. The Democrats’ enormous advantage in support in Silicon Valley — to the point at which employees of Facebook thought it was a perfectly reasonable question to publicly ask at a company meeting if the company had a moral duty to tilt media exposure to help Hillary win — is going to give them a nearly insurmountable lead here.
How many of those guys quit/took a sabbatical to work full time for Hillary?
I’m not sure that’s the correct metric to use when you’re talking about intellectual work products. The real question is “who was the single best person who took time to share their models and experience with the Clinton campaign?” Full-time labor isn’t necessarily required to share a model or algorithm that is a game-changer.
That said, I know (personally) at least three Google and two Facebook employees who took time off work to volunteer with the Clinton campaign, specifically to help with their analytics capabilities. So the answer has to be “more than a few”.
Its not like previous elections were the dark ages where chicken blood was splattered over an electoral map to determine where and how money was spent. To develop a game changing approach to electoral data probably does require some deep level of commitment.
Man, you had me laughing so hard imagining a bunch of political consultants in suits with a dead chicken and a dish of blood hovering over an electoral map. Maybe I just have a really weird sense of humor.
Sometimes I wonder if they wouldn’t get better results that way.
Speaking from my own experience with data analytics, you tend to need two things: a cohort of moderately-skilled analysts who do the “grunt work” of coding datasets and producing reports, and one or two very smart data scientists who develop the models that make sense of all of the information. The latter job doesn’t necessarily need to be full time, and it especially doesn’t need to be full time if the person you’re employing already does full time data science for a living and is just dropping into share their insights with you.
Profiling the average American consumer for marketing purposes, after all, is similar-verging-on-identical to profiling them for political purposes, not least because political predilections are a huge predictor of consumer spending habits. As a result, the two leading companies in advertising profiling (Google and Facebook) already have virtually all of the information and models that a political campaign would want.
Federal law governing in-kind contributions prohibits them from directly sharing this data and code with a campaign without remuneration, but Eric Schmidt (chairman of Alphabet, Google’s parent company) created a second company called Timshel in this last election cycle to sell consulting services to Democratic campaigns (primarily the Clinton campaign). Timshel’s staffing and such are confidential, but it’s something of an open secret in the Valley that some prominent Googlers have gone to work for it and subsequently returned post-election; it doesn’t seem like an enormous leap of imagination to speculate that the company is probably extending special consideration to those employees who wish to make a temporary transition to politics without giving up their lucrative jobs at Google.
Now, as I have no doubt that the principals involved are intelligent enough not to put any of this in writing where it could be used to prove obvious collusion, this is nothing more than a speculative conspiracy theory on my part. And certainly, no one I know personally has admitted to being part of a shadowy illegal scheme to circumvent campaign finance laws orchestrated by Google management. But it all aligns a bit too conveniently for me to think this is purely an innocent coincidence of an arrangement, either. And I suspect if the Republicans had a similar arrangement in place with the world’s largest data-collection corporation, Democrats and the media would be a lot more interested in knowing the details.
This is a reasonable assumption – but it’s still just that, an assumption.
And once again, my issue is with the moneyball comparison, where there were clearly people saying “this data stuff is nonsense and will never work!” which is NOT the position that the Trump campaign has taken…
>The solution is: stop treating life as a series of moral parables
How do you distinguish moral parables from explanations?
As an example, let’s say that I argue that the trouble with Hillary’s campaign wasn’t data or grizzle, but that she was trying to knock out a landslide rather than focusing on winning core areas. Is that a moral parable or simple explanation? It’s certainly not hard to turn it into a parable about hubris, overreaching ambition, failure of strategic thinking, or something else, but does it have to be? Or, perhaps a more important question, is it possible for it not to become such a parable? Because I certainly have no doubt that my predilection to think that Hillary suffer from hubris, excessive ambition, and poor strategic thinking inclines to to accept such an answer, and to want to think of it as unvarnished truth rather than a just so story that appeals to my pre-existing prejudices.
What is the point of the explanation? If it doesn’t have predictive value (or informational value for a future event) the it is a moral parable “do x if you want y to happen” rather than a robust view of the world.
To jump into the election debate with a differing point of view, consider a world where consensus polling is always wrong and can’t be right. Voters preferences are altered by the polling data, if Clinton has a large lead over trump and is 80%+ to win the election then voters flip for Trump and come out in droves, if the polls are close then voters stick with Clinton and Trump supporters stay home and Clinton wins in a landslide.
Or a more realistic version- what happens to Hillary’s turnout if Obama barely squeaks out a win in 2008 and or 2012?
The point isn’t to come up with an answer that is “correct” or even believable, but to figure out how much we actually know about causes of complex outcomes. It is easy in turn based games, like say basketball, to define a close win and a blowout, it is harder in different games. Take a two player game, two numbers are written on the board that both players can see, one player picks a number the other player gets the remaining number, whoever has the bigger number is the winner. It doesn’t matter how “close” the final numbers are to each other, they could be 1,000,000,000 and 1 or 101 and 100.99, the game isn’t close.
This is a good point. More of the “Eff you” vote may have come out to the polls, because Trump was perceived as the Underdog, because people thought HRC was going to win, due to the polls, so they wanted HRC not to win by very much.
Also, there is the fact that fraudulent software can easily be written for voting machines, and voting machine makers are often big donors to the GOP. And the software can be changed over time. E.g. if you have written the secret (by law; it’s a “trade secret”) software to take every hundredth vote for the Dem Congressional or presidential candidate and assign it to the Republican, in certain counties of Swing States, you just shift it as the polls change. If HRC has a bigger lead according to the polls, maybe you then adjust it to instead take every 50th vote for the Dem candidate and assign it to the Republican, in certain counties of Swing States.
So,yes, a lot of stuff can be adjusted, depending on how polls come out.
No, its not a good point, which is the whole point that you missed. “Maybe X, Maybe Y, Maybe X+Y” only sound good, what you have to know is what you don’t know before you can judge those points.
I am not even going to dignify that conspiracy theory with a response, so here it isn’t:
>Voters preferences are altered by the polling data
Are they? this seems like a very strong assumption. that the likelihood of someone voting goes up or down according to the polls is clear, though probably not in a way that is easily predictable or linear, but I can’t imagine that there was anyone out there was switching between candidates based on polls.
The first AI will be directed to publicly predict the following election. IT will then explode in a puff of smoke from the paradox.
That, or it will return an Error: Halting Problem Problem (Problem Problem Problem…).
Or a reference to zebra crossings.
Depends on how much attention to detail its programmers (or neural network architects, etc.) had.
@ cassander I’m posing a hypothetical
OF course there was, for third parties at least. I voted third party because the polls showed HRC to be a clear winner in my state, so to me there was no point in choosing between the big two. It turned out to be closer than expected, although HRC did win my state. It is possible that some others may have done the same in states that actually did flip unexpectedly, like say those who believed that Wisconsin was a foregone conclusion for Hillary. I don’t have a lot of confidence that this was a major factor, but it is possible.
Plus of course there could be people staying home if they thought the election was already determined in their state. Probably not significant in 2016, but possible.
A fair point, but I was thinking of cross-aisle switching, not third party switching.
Nassim Taleb talks about heuristics form the point of view of advice from your grandmother. The point isn’t that your grandmother, having lived for decades, has accumulated the knowledge, experience and wisdom to figure out the world and give great advice. Instead it is the observation that a lot of her thought processes came down through her mother/grandmother and their base came from generations before them, so anything your grandmother says has been vetted over numerous generations with superfluous information stripped away leaving only a core of “truth”.
Beware! Maybe superfluous information is stripped away leaving only a core of unfalsifiability.
Geochemist Randy Korotev has a site on id’ing meteorites. He often gets emails from people claiming to possess a meteorite handed down through generations. Often there’s a dramatic story about how their great grandma saw it falling from heaven in a ball of fire.
Of course, these rocks are NEVER actually meteorites when he examines them.
So these claims are falsifiable then.
What is a type of heuristic that would be generally applicable but not falsifiable?
Not if they stay on your mantelpiece and are never shown to a geologist.
Are there ways to know when your heuristic is about to stop working?
A heuristic shouldn’t give you an answer, it should eliminate false answers, and it should be generalizable. For example
“Don’t put all your eggs in one basket” can be seen as a heuristic. It doesn’t say “divide your eggs by 3, put in equal baskets, put any remaining eggs in the basket your dominant hand carries”- that would be both specific and non generalizable, but the attitude of the original is vague but helpful.
How many colleges should I apply to? More than 1.
How many different stocks should I aim to own? More than 1.
How many jobs should I apply for when I am looking? More than 1.
This is a really simple heuristic, but it gives good general results. There are times when it isn’t practical to use
I have one egg, how can I put it in at least 2 baskets? The heuristic “fails”, but you don’t need help to decide how many baskets are needed to carry 1 egg.
Sometimes you need to reclassify:
How many houses should I buy? The Heuristic implies more than 1, but I can only afford 1. Q: Why do you want to buy a house? A: Its a better investment than renting. Ok, reclassify buying a house as “investing” and then follow the heuristic- don’t buy a house that is so expensive it precludes other investments.
If your heuristic is good (i.e. follows those two guidelines) then it basically won’t “fail”, if your heuristic is bad then it is easy to know when it will stop working. The answer is “soon”.
Yogi Berra’s “heuristic” — “Prediction is hard, especially about the future.” — remains intact. I expected Trump to win this time but I can’t say I had much confidence in that belief and it certainly wasn’t based on data.
FiveThirtyEight gave Trump a 29% probability of winning the election, and it turned out that he did win by a small margin in the key states, losing the overall popular vote. For the 2012 election, FiveThirtyEight correctly predicted Obama’s landslide, giving him a 91% probability of winning.
Seems pretty good to me. Just because statistics don’t always correctly predict discrete outcomes, it doesn’t mean that they are useless.
There were people predicting 98% in favor of Clinton or 98% in favor of Trump, but these people were trading in propaganda and/or getting clicks by making a spectacle, not in making accurate predictions.
Yeah, I agree. But click to see the replies on ANY @NateSilver538 Twitter post and they’re full of “LOL YOU GOT THE ELECTION WRONG WHY SHOULD I LISTEN TO YOU.”
Twitter’s the biggest collection of retards ever assembled. Some dream of world peace. Some dream of Jeannie. I dream of Twitter running out of money, closing down, and its hard drives being ground into a fine powder.
Except that Nate Silver’s only authority comes from “he got the election right, you losers” so “LOL YOU GOT THE ELECTION WRONG WHY SHOULD I LISTEN TO YOU” is a perfect rebuttal.
If you want to engage with why his predictions and probabilities are bad read Taleb’s breakdown. Silver doesn’t publish probabilities, he publishes nonsense dressed up in math-ish language. Actual election probability predictions would graph like a binary option.
I think this is the wrong conclusion to draw from Moneyball and it’s certainly not a heuristic employed by any MLB team. All of the teams use advanced statistics (to varying degrees) and they all use old-school scouting where old guys drive around the dusty towns of America, (and around the world) to watch kids play and evaluate them.
Scouts can see things that PhDs can’t. It’s not uncommon for good players to have subpar stats when they’re 19 in A-ball, especially for pitchers who might spend 3 whole months working on one pitch. The team knows this and doesn’t care if their stats suck during that time. Another example: hitters who crush AAA pitching but suck in the majors because they can’t hit advanced breaking balls. If you only used math in these cases, you would lower your probability of picking winners and degrade your competitiveness. Scouts can see the potential for brilliance in 16 year olds, long before the skills have actualized and can translate to positive game outcomes.
I think the Moneyball lesson is that conditional on nobody else using hard math, using hard math plus traditional scouting will give you an unfair advantage.
I’m conjecturing that in a competitive market like the MLB (or winning elections?) these types of heuristics tend to disappear quickly. They’re basically ways of exploiting an inefficiency in the market. Once the competition knows about the inefficiency, the desire for power/money ensures that the inefficiencies disappear. Baseball’s “old school culture” was so entrenched that it took all of 7 years before basically everyone accepted that both scouting and advanced math were required to be competitive.
Billy Beane was a victim of Moneyball (the book, not the concept) in a way, because the owners of other teams starting asking hard questions like “why aren’t we doing that?” and then pretty soon they were all doing that and then people had to dig deeper for untapped inefficiencies.
Pedantry follows: King Canute walked into the waves to demonstrate the limits of even a king’s earthly powers, demonstrating humility rather than vainglory.
Sort of like the Ugly American’s origin as a eponymous character who might have been physically unattractive but who wanted to actually to understand local needs and values in Southeast Asians, unlike most of his more arrogant brethren.
Was that the origin of the phrase “ugly American”, though, or an ironic play on an existing term?
Further pedantry:Canute didn’t walk into the waves, he sat still and the rising tide came to him.
And presumably to his flattering courtiers as well, the point of the tale being to demonstrate that said courtiers’ Canute>>God theory of relative divinity was as wet as their regal finery.
Why would anyone want to use a heuristic to predict AI progress in the first place??
Heuristic-based-thinking is good for rapid decision-making under uncertainty. A car just cut into your lane: do you slam on the breaks, swerve, honk your horn…? You don’t have time to think it through. Use a heuristic. Great.
In any situation where you DO have time, heuristics are a terrible way of understanding the world! If it actually matters to you whether you are wrong or right, a heuristic is never going to be reliable enough, because the world is too complicated for any set of simple rules to reliably yield good outcomes.
AI research is complicated. Part of what it means to do AI research is to work on the definition of what AI is, because no one really knows yet. The model discussed here, where experts have some stable under-confidence / overconfidence factor that can be used to multiply their time predictions, might be a semi-okay model to use in a well-defined field, where the paradigm is stable, and there are lots of measurable, incremental improvements. AI is not such a field; there are probably at least a few paradigm revolutions in between where we are today, and the kind of AI that people speculating about AI progress care about.
The only way to actually make real predictions is to actually dig into why a given expert has the opinion they do, and evaluate the arguments for themselves. Aggregating across expert predictions is just going to lead to noise.
This is a lot of work. You actually have to read the opinions of a diverse set of experts, and understand the subject matter well enough to evaluate what parts of those opinions you agree and disagree with. You probably will have to learn a bunch of math and computer science.
So, are heuristics are shortcut if you don’t want to do all that work? Nope! If your goal is to sound intelligent about AI at cocktail parties, sure. But if you are actually making decisions where getting the right answer matters, using heuristics and saying things like “Well, I think there’s a 60% probability it’ll happen in the next 10 years…” is just fooling yourself with a false sense of precision.
A better way of looking at it is: “I don’t know. I have no idea. I can: either invest the time to actually know, or: accept that I don’t know, and make decisions based on that total uncertainty”. That’s an uncomfortable statement, because neither is very appealing: the first is a lot of work, and the second is scary. But unfortunately, that’s the game, people… it’s not easy to predict the future!
Scott, this is really embarrassing. You’re better than this. You can’t think of the one glaringly obvious difference between baseball and politics?
I can think of a couple glaringly obvious differences between baseball and politics, but it’s not clear to me which one you think matters.
Baseball is a relatively static system and we get hundreds of games every year and we have for decades. Essentially the data we have for baseball is sufficient to make very specific predictions at a very high level of accuracy and confidence. The famous OBP stat from Moneyball is a good example.
Meanwhile we’ve had a whopping 45 presidents, and 60 presidential elections. Furthermore the background to each campaign has dozens and dozens of outside factors.
There’s simply far less room for your starting assumptions to taint your results in baseball vs politics.