In year 1, [the Good Judgment Project] beat the official control group by 60%. In year 2, we beat the control group by 78%. GJP also beat its university-affiliated competitors, including the Uniersity of Michigan and MIT, by hefty margins, from 30% to 70%, and even outperformed professional intellgience analysts with access to classified data. After two years, GJP was doing so much better than its academic competitors that IARPA dropped the other teams.
I keep wondering what these other teams were doing. Good Judgment Project sounds like it was doing the simplest, most obvious possible tactic – asking people to predict things and seeing what happened. David Manheim says the other groups tried “more straightforward wisdom of crowds” methods, so maybe GJP’s secret sauce was concentrating on the best people instead of on everyone? Still seems like it should have taken fewer than five universities and a branch of government to think of that.
One result that particularly surprised me was the effect of a tutorial covering some basic concepts that we’ll explore in this book and are summarized in the Ten Commandments appendix. It took only about sixty minutes to read and improved accuracy by roughly 10% through the entire tournament year. Yes, 10% may sound modest, but it was achieved at so little cost.
These Ten Commandments are available online here.
For centuries, [aversion to measuring things and collecting evidence] hobbled progress in medicine. When physicians finally accepted that their experience and perceptions were not reliable means of determining whether a treatment works, they turned to scientific testing – and medicine finally started to make rapid advances.
I see what Tetlock is trying to say here, but as written it’s horribly wrong.
Evidence-based medicine could be fairly described as starting in the 1970s with Cochrane’s first book, and really took off in the 80s and 90s. But this is also the period when rapid medical advances started slowing down! In my own field of psychiatry, the greatest advances were the first antidepressants and antipsychotics in the 50s, the benzodiazepines in the 60s, and then a gradual trickle of slightly upgraded versions of these through the 70s and 80s. The last new drugs that could be called “revolutionary” by any stretch of the imagination were probably the first SSRIs in the early 80s. This is the conventional wisdom of the field and everybody admits this, but I would add the stronger claim that the older medications in many ways work better. I know less about the history of other subfields, but they seem broadly similar – the really amazing discoveries are all pre-EBM, and the new drugs are mostly nicer streamlined versions of the old ones.
There’s an obvious “low-hanging fruit” argument to be made here, but some people (I think Michael Vassar sometimes toys with this idea) go further and say that evidence-based medicine as currently practiced can actually retard progress. In the old days, people tried possible new medications in a very free-form and fluid way that let everyone test their pet ideas quickly and keep the ones that worked; nowadays any potential innovations need $100 million 10-year multi-center trials which will only get funded in certain very specific situations. And in the old days, a drug would only be kept if it showed obvious undeniable improvement in patients, whereas nowadays if a trial shows a p < 0.05, d = 0.10 advantage, that's enough to make it the new standard if it's got a good pharma company behind it. So the old method allowed massive-scale innovation combined with high standards for success; the new method only allows very limited innovation but keeps everything that can show the slightest positive effect whatsoever on an easily-rigged but very expensive test. I'm not sure I believe in the strong version of this argument (the low-hanging fruit angle is probably sufficient), but the idea that medicine only started advancing after the discovery of evidence-based medicine is just wrong. A better way of phrasing it might be that around that time we started getting fewer innovations, but we also became a lot more effective and intelligent at using the innovations we already had.
Consider Galen, the second-century physician to Rome’s emperors…Galen was untroubled by doubt. Each outcome confirmed he was right, no matter how equivocal the evidence might look to someone less wise than the master. “All who drink of this treatment recover in a short time, except those whom it does not help, who all die,” he wrote. “It is obvious, therefore, that it fails only in incurable cases.”
After hearing one too many “everyone thought Columbus would fall off the edge of the flat world” -style stories, I tend to be skeptical of “people in the past were hilariously stupid” anecdotes. I don’t know anything about Galen, but I wonder if this was really the whole story.
When hospitals created cardiac care units to treat patients recovering from heart attacks, Cochrane proposed a randomized trial to determine whether the new units delivered better results than the old treatment, which was to send the patient home for monitoring and bed rest. Physicians balked. It was obvious the cardiac care units were superior, they said, and denying patients the best care would be unethical. But Cochrane was not a man to back down…he got his trial: some patients, randomly selected, were sent to the cardiac care units while others were sent home for monitoring and bed rest. Partway through the trial, Cochrane met with a group of the cardiologists who had tried to stop his experiment. He told them that he had preliminary results. The difference in outcomes between the two treatments was not statistically signficant, he emphasized, but it appeared that patients might do slightly betteri n the cardiac care units. “They were vociferous in their abuse: ‘Archie,’ they said, ‘we always thought you were unethical. You must stop the trial at once.'” But then Cochrane revealed he had played a little trick. He had reversed the results: home care had done slightly better than the cardiac units. “There was dead silence and I felt rather sick because they were, after all, my medical colleagues.”
This story is the key to everything. See also my political spectrum quiz and the graph that inspired it. Almost nobody has consistent meta-level principles. Almost nobody really has opinions like “this study’s methodology is good enough to believe” or “if one group has a survival advantage of size X, that necessitates stopping the study as unethical”. The cardiologists sculpted their meta-level principles around what best supported their object-level opinions – that more cardiology is better – and so generated the meta-level principles “Cochrane’s experiment is accurate” and “if one group has a slight survival advantage, that’s all we need to know before ordering the experiment stopped as unethical.” If Cochrane had (truthfully) told them that the cardiology group was doing worse, they would have generated the meta-level principles “Cochrane’s experiment is flawed” and “if one group has a slight survival advantage that means nothing and it’s just a coincidence”. In some sense this is correct from a Bayesian point of view – I interpret sonar scans of Loch Ness that find no monsters to be probably accurate, but if a sonar scan did find a monster I’d wonder if it was a hoax – but in less obvious situations it can be a disaster. Cochrane understood this and so fed them the wrong data and let them sell him the rope he needed to hang them. I know no better solution to this except (possibly) adversarial collaboration. Also, I suppose this is more proof (as if we needed it) that cardiologists are evil.
In the late 1940s, the Communist government of Yugoslavia broke from the Soviet Union, raising fears that the Soviets would invade. In March 1951 [US intelligence under Sherman Kent reported there was a “serious possibility” of a Soviet attack.] But a few days later, Kent was chatting with a senior State Department official who casually asked, “By the way, what did you people mean by the expression ‘serious possibility’? What kind of odds did you have in mind?” Kent said he was pessimistic. He felt that the odds were about 65 to 35 in favor of an attack. The official was startled. He and his colleagues had taken “serious possibility” to mean much lower odds.
Disturbed, Kent went back to his team. They had all agreed to use “serious possibility” in the [report], so Kent asked each person, in turn, what he thought it meant. One analyst said it meant odds of about 80%. Another thought it meant odds of 20% – exactly the opposite. Other answers were scattered between those extremes. Kent was floored. A phrase that looked informative was so vague as to be almost useless…
In 1961, when the CIA was planning to topple the Castro government by landing a small army of Cuban expatriates at the Bay of Pigs, President John F. Kennedy turned to the military for an unbiased assessment. The Joint Chiefs of Staff concluded that the plan had a “fair chance” of success. The man who wrote the words “fair chance” later said he had in mind odds of 3 to 1 against. But Kennedy was never told precisely what “fair chance” meant and, not unreasonably, he took it to be a much more positive assessment.
Nate Silver, Princeton’s Sam Wang, and other poll aggregators were hailed for correctly predicting all fifty state outcomes, but almost no one noted that a crude, across-the-board prediction of “no change” – if a state went Democratic or Republican in 2008, it will do the same in 2012 – would have scored forty-eight out of fifty, which suggests that the many excited exclamations of “he called all fifty states!” we heard at the time were a tad overwrought.
I didn’t realize this. I think this election I’m going to predict the state-by-state results just so that I can tell people I “predicted 48 of the 50 states” or something and sound really impressive.
The [Expert Political Judgment] data revealed an inverse correlation between fame and accuracy: the more famous an expert was, the less accurate he was. That’s not because editors, producers, and the public go looking for bad forecasters. They go looking for hedgehogs, who just happen to be bad forecasters. Animated by a Big Idea, hedgehogs tell tight, simple, clear stories that grab and hold audiences.
One day aliens are going to discover humanity and be absolutely shocked we made it past the wooden-club stage.
In 2008, the Office of the Director of national Intelligence – which sits atop the entire network of sixteen intelligence agencies -asked the National research Council to form a committee. The task was to synthesize research on good judgment and help the IC put that research to good use. By Washington’s standards, it was a bold (or rash) thing to do. It’s not every day that a bureaucracy pays one of the world’s most respected scientific institutions to produce an objective report that might conclude that the bureaucracy was clueless.
This was a big theme of the book: the US intelligence community deserves celebration for daring to investigate its own competency at all. Interestingly, a lot of its investigations said it was doing things more right than we would think: Tetlock mentions that even independent-to-hostile investigators concluded that it had been correct in using the facts it had to believe Saddam had WMDs. The book didn’t explain exactly how this worked: possibly Saddam was trying to deceive everyone into thinking he had WMDs to prevent attacks, and did a good job? This was part of what got the intelligence community interested in probability: given that they had made a reasonable decision in saying there were WMDs, but it had been a big disaster for the United States, what could they have done differently? Their answer was “continue to make the reasonable decision, but learn to calibrate themselves well enough to admit there’s a big chance they’re wrong.”
[We finished by giving] the forecast a final tweak: “extremizing” it, meaning pushing it closer to 100% or zero. If the forecast is 70% you might bump it up to, say, 85%. If it’s 30%, you might reduce it to 15%…[it] is based on a pretty simple insight: when you combine the judgments of a large group of people to calculate the “wisdom of the crowd” you collect all of the relevant information that is dispersed among all those people. But none of those people has access to all that information…what would happen if every one of those people were given all the information? They would become more confident. If you then calculated the wisdom of the crowd, it too would be more extreme.
Something to remember if you’re doing wisdom-of-crowds with calibration estimates.
The correlation between how well individuals do from one year to the next is about 0.65…Regular forecasters scored higher on intelligence and knowledge tests than about 70% of the population. Superforecasters did better, placing higher than about 80% of the population.
People interested in taking these kinds of tests are generally intelligent; superforecasters are somewhat more, but not vastly more, intelligent than that.
Researchers have found that merely asking people to assume their initial judgment is wrong, to seriously consider why that might be, and then make another judgment, produces a second estimate which, when combined with the first, improves accuracy almost as much as getting a second estimate from another person.
There’s a rationalist tradition – I think it started with Mike and Alicorn – that before you get married, you ask all your friends to imagine that the marriage failed and tell you why. I guess if you just asked people “Will our marriage fail?” everyone would say no, either out of optimism or social desirability bias. If you ask “Assume our marriage failed and tell us why”, you’ll actually hear people’s concerns. I think this is the same principle. On the other hand, I’ve never heard of anyone trying this and deciding not to get married after all, so maybe we’re just going through the motions.
[Superforecaster] Doug Lorch knows that when people read for pleasure they naturally gravitate to the like-minded. So he created a database containing hundreds of information sources – from the New York Times to obscure blogs – that are tagged by their ideological oreintation, subject matter, and geographical origin, then wrote a program that selects what he should read next using criteria that maximize diversity.
Of all humans, only Doug Lorch is virtuous. Well, Doug Lorch and this guy from rationalist Tumblr who tried to get the program but was told it wasn’t really the sort of thing you could just copy and give someone.
[The CIA was advising Obama about whether Osama bin Laden was in Abbotabad, Pakistan; their estimates averaged around 70%]. “Okay, this is a probability thing,” the President said in response, according to Bowden’s account. Bowden editorializes: “Ever since the agency’s erroneous call a decade earlier [on Saddam’s weapons of mass destruction], the CIA had instituted an almost comically elaborate process for weighing certainty…it was like trying to controve a mathematical formula for good judgment.”Bowden was clearly not impressed with the CIA’s use of numbers and probabilities. Neither was Barack Obama, according to Bowden. “What you ended up with, as the president was finding, and as he would later explain to me, was not more certainty but more confusion…in this situation, what you started to get was probabilities that disguised uncertainty, as opposed to actually providing you with useful information…”
After listening to the widely ranging opinions, Obama addressed the rrom. “This is fifty-fifty,” he said. That silenced everyone. “Look guys, this is a flip of the coin. I can’t base this decision on the notion that we have any greater certainty than that…
The information Bowden provides is sketchy but it appears that the media estimate of the CIA officers – the “wisdom of the crowd” – was around 70%. And yet Obama declares the reality to be “fifty-fifty.” What does he mean by that?…Bowden’s account reminded me of an offhanded remark that Amos Tversky made some thirty years ago…In dealing with probabilities, he said, most people only have three settings: “gonna happen,” “not gonna happen,” and “maybe”.
Lest I make it look like Tetlock is being too unfair to Obama, he goes on to say that maybe he was speaking colloquially. But the way we speak colloquially says a lot about us, and there are many other examples of people saying this sort of thing and meaning it. This ties back into an old argument we had here on whether something like a Bayesian concept of probability was meaningful/useful. Some people said that it wasn’t, because everyone basically understands probability and Bayes doesn’t add much to that. I said it was, because people’s intuitive idea of probability is hopelessly confused and people don’t really think in probabilistic terms. I think we have no idea how confused most people’s idea of probability is, and perhaps even Obama, one of our more intellectual presidents, has some issues there.
Barbara Mellers has shown that granularity predicts accuracy: the average forecaster who sticks with the tens – 20%, 30%, 40% – is less accurate than the finer-grained forecaster who uses fives – 20%, 25%, 30% – and still less accurate than the even finer-grained forecaster who uses ones – 20%, 21%, 22%. As a further test, she rounded forecasts to make them less granular, so a forecast at the greatest granularity possible in the tournament, single percentage points, would be rounded to the nearest five, and then the nearest ten. This way, all of the forecasts were made one level less granular. She then recalculated Bier scores and discovered that superforecasters lost accuracy in response to even the smallest-scale rounding, to the nearest 0.05, whereas regular forecasters lost little even from rounding four times as large, to the nearest 0.2.
This was the part nobody on the comments to the last post believed, and I have trouble believing it too.
[There’s a famous Keynes quote: “When the facts change, I change my mind. What do you do, sir?”] It’s cited in countless books, including one written by me and another by my coauthor. Google it and you will find it’s all over the internet. Of all the many famous things Keynes says, it’s probably the most famous. But while researching this book, I tried to track it to its source and failed. Instead I found a post by a Wall Street Journal blogger, which said that no one has ever discovered its provenance and the two leading experts on Keynes think it is apocryphal. In light of these facts, and in the spirit of what Keynes apparently never said, I concluded that I was wrong.
The funny part is that if this fact is true, we’ve known it for fifty years, and people still haven’t changed their mind about whether he said it or not.
“Keynes is always ready to contradict not only his colleagues but also himself whenever circustancse make this seem appropriate,” re[prted a 1945 profile of the “consistently inconsistent” economist. “So far from feeling guilty about such reversals of position, he utilizes them as pretexts for rebukes to those he saw as less nimble-minded. Legend says that while conferring with Roosevelt at Quebec, Churchill sent Keynes a cable reading, ‘Am coming around to your point of view.’ His Lordship replied, ‘Sorry to hear it. Have started to change my mind.'”
I sympathize with this every time people email me to say how much they like the Non-Libertarian FAQ.
Police officers spend a lot of time figuring out who is telling the truth and who is lying, but research has found they aren’t nearly as good at it as they think they are and they tend not to get better with experience…predictably, psychologists who test police officers’ ability to spot lies in a controlled setting find a big gap between their confidence and their skill. And that gap grows as officers become more experienced and they assume, not unreasonably, that their experience has made them better lie detectors.
There’s some similar research on doctors and certain types of diagnostic tasks that don’t give quick feedback.
In 1988, when the Soviet Union was implementing major reforms that had people wondering about its future, I asked experts to estimate how likely it was that the Communist Party would lose its monopoly on power in the Soviet Union in the next five years. In 1991 the world watched in shock as the Soviet Union disintegrated. So in 1992-93 I retunred to the experts, reminded them of the question in 1988, and asked them to recall their estimates. On average, the experts recalled a number 31 percentage points higher than the correct figure. So an expert who thought there was only a 10% chance might remember herself thinking there was a 40% or 50% chance. There was even a case in which an expert who pegged the probability at 20% recalled it as 70%.
As the old saying goes, hindsight is 20/70.
The results were clear-cut each year. Teams of ordinary forecasters beat the wisdom of the crowd by about 10%. Prediction markets beat ordinary teams by about 20%. And superteams beat prediction markets by 15% to 30%. I can already hear the protests from my colleagues in finance that the only reason the superteams beat the prediction markets was that our markets lacked liquidity…they may be right. It is a testable idea, and one worth testing.
The correct way to phrase this is “if there is ever a large and liquid prediction market, Philip Tetlock will gather his superforecasters, beat the market, become a zillionaire, and then the market will be equal to or better than the forecasters.”
Orders in the Wehrmacht were often short and simple – even when history hung in the balance. “Gentlemen, I demand that your divisions completely cross the German borders, completely cross the Belgian borders, and completely cross the River Meuse,” a senior officer told the commanders who would launch the great assault into Belgium and France on May 10, 1940. “I don’t care how you do it, that’s completely up to you.”
This is the opposite of the image most people have of Germany’s World War II military. The Wehrmacht served a Nazi regime that rpeached total obedience to the dictates of the Fuhrer, and everyone emembers the old newsreels of German soldiers marching in goose-stepping unison…but what is often forgotten is that the Nazis did not create the Wehrmacht. They inherited it. And it could not have been more different from the unthinking machine we imagine.
Shortly after WWI, Eisenhower, then a junior officer who had some experience witht he new weapons called tanks, published an article in the US Army’s Infantry Journal making the modest argument that “the clumsy, awkward and snail-like progress of the old tanks must be forgotten, and in their place we must picture this speedy, reliable, and efficient engine of destruction.” Eisenhower was dressed down. “I was told my ideas were not only wrong but dangerous, and that henceforth I was to keep them to myself,” he recalled. “Particularly, I was not to publish anything incompatible with solid infantry doctrine. If I did, I would be hauled before a court martial.”
Tetlock includes a section on what makes good teams and organizations. He concludes that they’re effective when low-level members are given leeway both to pursue their own tasks as best they see fit, and to question and challenge their higher-ups. He contrasts the Wehrmacht, which was very good at this and overperformed its fundamentals in WWII, to the US Army, which was originally very bad at this and underperformed its fundamentals until it figured this out. Later in the chapter, he admits that his choice of examples might raise some eyebrows, but says that he did it on purpose to teach us to think critically and overcome cognitive dissonance between our moral preconceptions and our factual beliefs. I hope he has tenure.
Ultimately the Wehrmacht failed. In part, it was overwhelmed by its enemies’ superior resources. But it also made blunders – often because its commander-in-chief, Adolf Hitler, took direct control of operations in violation of Helmuth von Moltke’s principles, nowhere with more disastrous effect than during the invasion of Normandy. The Allies feared that after their troops landed, German tanks would drive them back to the beaches and into the sea, but Hitler had directed that the reserves could only move on his personal command. Hitler slept late. For hours after the Allies landed on the beaches, the dictator’s aides refused to wake him to ask if he wanted to order the tanks into battle.
Early to bed
And early to stir up
Makes a man healthy
And ruler of Europe
The humility required for good judgment is not self-doubt – the sense that you are untalented, unintelligent, or unworthy. It is intellectual humility. It is a recognition that reality is profoundly complex, that seeing things clearly is a constant struggle, when it can be done at all, and that human judgment must therefore be riddled with mistakes. This is true for fools and geniuses alike. So it’s quite possible to think highly of yourself and be intellectually humble. In fact, this combination can be wonderfully fruitful. Intellectual humility compels the careful reflection necessary for good judgment; confidence in one’s abilities inspires determined action.
Yes! This is a really good explanation of Eliezer Yudkowsky’s Say It Loud.
(and that sentence would also have worked without the apostrophe or anything after it).
I am…optimistic that smart, dedicated people can inoculate themselves to some degree against certain cognitive illusions. That may sound like a tempest in an academic teapot, but it has real-world implications. If I am right, organizations will have more to gain from recruiting and training talented people to resist their biases.
This is probably a good time to mention that CFAR is hiring.