"Talks a good game about freedom when out of power, but once he’s in – bam! Everyone's enslaved in the human-flourishing mines."

SSC Journal Club: Serotonin Receptors

Pop science likes to dub dopamine “the reward chemical” and serotonin “the happiness chemical”. God only knows what norepinephrine is, but I’m sure it’s cutesy.

In real life, all of this is much more complicated. Dopamine might be “the surprisal in a hierarchical predictive model chemical”, but even that can’t be more than a gross oversimplification. As for serotonin, people have studied it for seventy years and the best they can come up with is “uh, something to do with stress”.

Serotonin and brain function: a tale of two receptors by Robin Carhart-Harris and David Nutt tries to cut through the mystery. Both authors are suitably important to attempt such an undertaking. Carhart-Harris is a neuropsychopharmacologist and one of the top psychedelic researchers in the world. Nutt was previously the British drug czar but missed the memo saying drug czars were actually supposed to be against drugs; after using his position to tell everyone drugs were pretty great, he was summarily fired. Now he’s another neuropsychopharmacology professor, though with cool side projects like inventing magical side-effect-free alcohol. These are good people.

And they have a good theory. One stumbling block in past attempts to understand serotonin was the brain’s dozen or so different types serotonin receptors, all of which seem to do kind of different things. Carhart-Harris and Nutt (subsequently: CH&N) focus on two of these which show up again and again in psychiatry: 5-HT1A and 5-HT2A. Past studies had always shown these two receptors having kind of opposite effects, which confused things pretty thoroughly: why would you want a chemical that does two opposite things?

5-HT1A is the most common serotonin receptor in the brain. When SSRI antidepressants like Prozac, Zoloft and Celexa increase serotonin, this is the receptor most of that serotonin goes to. Some other antidepressants and antianxiety medications like BuSpar, Viibryd and Trintellix just stimulate this receptor directly. So it looks like this receptor does something like “reduce depression and anxiety”. But this falls afoul of a version of Algernon’s Law: there shouldn’t be any switch in the brain which is 100% good or 100% bad. Why have a receptor for treating depression and anxiety, rather than just always keep the receptor at maximum so you’re never depressed or anxious?

5-HT2A is another pretty common receptor. Most new antipsychotics like Seroquel and Abilify block this receptor. And most psychedelic drugs like LSD and magic mushrooms stimulate it really hard. Since psychedelics make you kind of crazy, and antipsychotics make you stop being crazy, 5-HT2A must have something to do with psychosis. Of course, this is another Algernon’s Law violation: why is there a receptor just to make you psychotic?

1A and 2A seem to “fight” each other. The more you activate 1A, the quieter 2A becomes – this is why people on SSRIs get less effect from psychedelics. And all the drugs that block 2A are also decent antidepressants – this is why people recommend Seroquel for depression even though it’s an antipsychotic – and this seems to work because blocking 2A increases 1A.

On the other hand, there also seems to be some deeper unity. 1A makes you less depressed. 2A – well, we keep hearing all these studies, some of them from Dr. Carhart-Harris himself, showing that magic mushrooms treat depression really well. Not just as a once daily medication, but in the sense that one trip on mushrooms can make you long-term – maybe permanently – less depressed. This is pretty weird. Blocking 2A makes you less depressed? But stimulating 2A also makes you less depressed, in a different and more permanent way? What’s going on?

CH&N argue: both 1A and 2A promote coping with stress. 1A promotes “passive coping”. 2A promotes “active coping”.

Passive coping is basically being stoic, having a stiff upper lip, and waiting it out. Imagine you’re at some kind of terrible job and your boss is bullying you all the time and you can’t stand it and you get depressed and anxious. Your psychiatrist gives you an SSRI (or BuSpar, or Viibryd, or some other 1A stimulator) and now, you can stand it. Your boss is still just as mean. Your life is still just as bad. But you sort of shrug, think “what can I do?” and get back to work. This isn’t the most inspiring story, but it’s better than alternatives like “being a wreck” or “snapping and attacking your boss”. Did I mention that 1A is known to decrease impulsivity and aggression? Makes sense.

Active coping is…uh…sort of unclear from the paper. It sounds like it should mean working to solve the problem – quitting your job, finding a way to stand up for yourself. Heck, even snapping and attacking your boss would tie in with the psychosis angle. This is…not exactly where CH&N go, as far as I can tell. Active coping is like…an LSD trip? It’s some kind of grabbing the brain and shaking it, in the hopes that maybe when it settles it will be in a state that’s better able to deal with whatever’s going on. This sort of makes sense, insofar as big steps like quitting your job might require a lot of mental shake-up to consider. It seems to have something to do with a process of increased plasticity, becoming bolder to avoid getting trapped at local minima, and increasing the information-theoretic entropy of brain states. This definitely sounds like the sort of thing that can cause psychosis, and maybe it sounds like the sort of thing that might help?

MDMA, a strong 2A agonist, is currently in Phase III trials as a treatment for post-traumatic stress disorder. It looks really promising. Under CH&N’s theory, this makes a lot of sense. If you have trauma, your thoughts get stuck in some pattern which is useful for dealing with or avoiding the traumatic situation – for example, an abused child learns to be suspicious and afraid of everybody. People do therapy for years trying to cast off these thought patterns; they know they’re no longer adaptive, but they just can’t get rid of them. On MDMA – and especially in MDMA-assisted therapy – people find it easy; the usual metaphor is calcified thought patterns suddenly become fluid and re-writable. Is this the sort of “increased plasticity” that CH&N describe?

This theory gives an explanation of how 1A and 2A can have such a complex – and sometimes antagonistic – relationship. When a person undergoes adversity, their brain releases serotonin, which starts by hitting the 1A receptors. They bear it stoically and hopefully soldier through. But if the adversity gets really bad and the serotonin release passes some threshold, it starts hitting the 2A receptors instead. Now their brain realizes things are pretty bad, it’s got to try high-variance strategies, and so it increases its randomness in the hope of stumbling across a way-out-there solution to the problem.

(not super-clear what problem John Lilly thought he was solving by accusing space aliens of orchestrating a massive conspiracy to manipuate the world’s coincidences, but it’s a pretty safe bet the 2A receptor was involved somehow.)

I find the whole thing pretty plausible. But as written, it doesn’t entirely answer the Algernon’s Law questions. Why doesn’t everyone just have 1A and 2A functions set to max all the time? What’s the tradeoff?

There are some obvious possibilities. Too much 2A stimulation makes you psychotic. This puts the efficacy of atypical antipsychotics like Seroquel in a new light: they’re saying something like “keep your thoughts very careful and low-risk, this isn’t a good time to be deviating from normal patterns”. And so maybe someone who otherwise would have believed the space aliens were putting a transmitter in his teeth will decide not to think that. Is there a shade of Bayesian brain theory here? Is the phrase we’re looking for “strength of priors”? I don’t know.

Likewise obvious: if 1A promotes stoic coping, then too much of it prevents you from actively making your life better. One can imagine how this would be more relevant in the environment of evolutionary adaptedness than today. Back then stressors could have been some specific person whose skull you could bash in with a rock. Nowadays they tend to be things like corporations, national governments, and groups of people with terrible politics on Twitter; attempted skull-bashing, as satisfying as it might feel, is highly disrecommended.

I don’t know if these stories are true. They don’t really explain why 1A and 2A function seem inversely related. Is this just a wiring issue? Or is there some fundamental reason why ability to passively cope can’t coexist with creative outside-the-box problem-solving? Maybe the coping involves some sort of mental resolution not to let all the stress change the brain at all, and the problem-solving involves the brain becoming superplastic and really easily influenced by external events. But it’s not really clear why either of those things should be necessary.

Also, we should remember that although CH&N’s theory explains a lot, we’re reading the case they’re presenting, and there’s a lot they leave out. Some might complain that calling 2A the “active coping receptor” is as reductionistic as the whole “dopamine is the reward chemical” thing – 2A is also involved in obesity, sexual dysfunction, some forms of insomnia, possibly chronic fatigue syndrome, platelet clumping, et cetera. All of these psychedelics do opposite things acutely and chronically – something CH&N acknowledge – so you have to be really careful with time course in order to figure out whether your acid trip is treating depression due to acutely increased 2A stimulation or chronically decreased number of 2A receptors. Both Carhart-Harris and Nutt have spent big parts of their careers advocating more use of psychedelics, so them coming up with a theory of why psychedelics are really good is both reasonable and suspicious.

Still, this is as good a theory of serotonin function as anything else I’ve seen, and it will be exciting to see if it suggests any avenues for experimental research to confirm or refute it.

This entry was posted in Uncategorized and tagged , . Bookmark the permalink.

Leave a Reply

93 Responses to SSC Journal Club: Serotonin Receptors

  1. Winter Shaker says:

    To be pedantic, David Nutt wasn’t fired as the UK government’s chief drugs adviser for saying that drugs in general are pretty great, rather for claiming that the government’s assessment of the relative risks of different drugs makes no sense in terms of their actual known risk profiles (for instance, that it makes no sense that MDMA is in the ‘most dangerous currently-illegal drug’ category while alcohol isn’t even in the ‘least dangerous currently-illegal drug category’), and also for pointing out that the risk profiles of some currently illegal drugs compares favourably with some fun-but-risky non-drug recreational activities. The tipping point that got him sacked appears to have been his claims about the risks of horse riding being in the same ballpark as those of taking MDMA – my attempt to link the article tripped the spam filter, but you can look up ‘Equasy – A Harmful Addiction’. The government didn’t even try to show that he was mistaken in those claims, just that, in his position, he wasn’t supposed to appear to legitimise MDMA use (or, presumably, delgitimise horse riding) by saying so.

    He’s actually pretty concerned about the dangers of alcohol, hence his efforts to come up with a low-risk substitute.

    • mindspillage says:

      After reading Scott’s description I went to look at Nutt’s Wikipedia article (citation needed, etc.) and he sounds like exactly the kind of guy I would want as drug czar.

      • Murphy says:

        Yep, a fairly rational person, unfortunately he was what his bosses didn’t want: a thinking person, they wanted someone who’s say what he’s supposed to say and basically be Mr Mackey.

    • Speaker To Animals says:

      Horse riding on MDMA is very risky indeed.

      Stick to ketamine.

    • theory says:

      David Nutt also went too far in some of his claims. You cannot be a drug czar and claim that alcohol is more dangerous than heroin and crack. That’s insanely irresponsible to suggest.

      His basis for that is his study titled “Drug harms in the UK: a multicriteria decision analysis”, and it is incredibly shoddy science, the type usually deconstructed on SSC as garbage science. It consisted of measuring the absolute rates of harm for each drug (shockingly, more people suffer from alcohol than from heroin and crack when you don’t adjust for rates of use), with the pathetic caveat that “harms of drugs are affected by their availability and legal status”. Yes, publish a study purporting that alcohol is more dangerous than heroin, and then quietly insert in the caveats that you just measured overall harm and didn’t adjust for the fact that one is far more readily consumed, knowing that none of the headlines will report that caveat. And it is insanely irresponsible to do that from a government platform, because there really are young people out there that will take him at his word for it, or use it to rationalize their behavior.

      Actually, I take that back – he didn’t even actually measure the harms, but rather it was a survey of 15 people (including himself!) on their views of the absolute rates of harm for various drugs. The group was specifically chosen as “a new organization of drug experts independent of government interference” and taken to be an objective measure of harm. I’m going to go out on a limb and say that since Mr. Nutt was at liberty to choose who he wanted to choose, it shouldn’t be surprising that his study proved what he wanted to prove.

      Drug policy is often wrongheaded. That doesn’t mean we have to dignify everyone on the other side as “good people”, particularly when they engage in such garbage science.

      • Protagoras says:

        Alcohol is involved in a lot of violence, and alcohol withdrawal has an unusually high risk of death. People don’t OD on alcohol as much as they do on heroin, but while that’s partly the result of the fact that people usually throw up and/or pass out before consuming enough alcohol to kill them unless they’re chugging overproof vodka or something, it’s also partly because of the unreliable purity of black market heroin. And of course it’s easy to get the misleading impression that while there are lots of alcohol users who don’t seem to have problems, there aren’t a lot of “hard” drug users who don’t have problems. But that’s an illusion caused by the fact that users of “hard” drugs who don’t have problems are also the ones together enough to keep their drug use secret.

        Of course the absolute numbers are worse for alcohol, which is already enough reason to say “alcohol is more dangerous than heroin or crack” is true in some sense (quite possibly the most relevant sense for someone directing public policy). The rate of problems relative to the population of users may make alcohol look less bad on that measure (though note that it’s a lot harder to be sure what the true size of the population of heroin or crack users is), but that may not be the right measure for someone directing public policy to employ. Nor is it obvious that, say, decriminalization of heroin and crack would change the absolute numbers enough to knock alcohol out of the lead, given the ways in which prohibition makes drugs more dangerous and the differing cultural attitudes toward the various drugs. So I’m not seeing what’s “insanely irresponsible” about his claims, or how you’re so confident they’re based on “garbage science.”

        • theory says:

          Of course the absolute numbers are worse for alcohol, which is already enough reason to say “alcohol is more dangerous than heroin or crack” is true in some sense (quite possibly the most relevant sense for someone directing public policy). The rate of problems relative to the population of users may make alcohol look less bad on that measure (though note that it’s a lot harder to be sure what the true size of the population of heroin or crack users is), but that may not be the right measure for someone directing public policy to employ.

          I strongly disagree. Absolute numbers are way less important than relative numbers for public policy reasons. Imagine if the FDA regulated by absolute numbers instead of relative numbers. You would have to absurdly claim that aspirin is more dangerous than, say, a fourth-line chemotherapy drug.

          Nor is it obvious that, say, decriminalization of heroin and crack would change the absolute numbers enough to knock alcohol out of the lead, given the ways in which prohibition makes drugs more dangerous and the differing cultural attitudes toward the various drugs. So I’m not seeing what’s “insanely irresponsible” about his claims, or

          It’s irresponsible because people want to believe certain things, and if you’re a government policy-maker, it’s your job to tell them the truth, and not just what they want to hear. The truth is that heroin is far worse for a person than alcohol, but plenty of people want to believe that heroin isn’t that bad compared to alcohol. Misleading them into thinking that alcohol is more harmful, because you commissioned a study of your friends that said alcohol was worse, as a government spokesperson, is an abuse of your position.

          how you’re so confident they’re based on “garbage science.”

          Did you read the part about his methodology? Imagine that Trump decided to outlaw marijuana because he commissioned a study where he invited a group of his friends to rank how harmful marijuana was. I am sure you would consider that garbage science. You shouldn’t see it any differently just because it reaches a different substantive conclusion.

          • Protagoras says:

            I kind of do think aspirin is more dangerous than your fourth-line chemotherapy drug, though widespread use also means it produces much more benefit, which the FDA should also take into account. But the much more widespread use, and resulting greater number of deaths, is absolutely a reason the FDA should be especially vigilant when it comes to OTC painkillers. So while I think comparing therapeutic drugs to recreational drugs is kind of apples to oranges, I also don’t think it ends up supporting your point very well even if they were comparable.

        • Douglas Knight says:

          it’s a lot harder to be sure what the true size of the population of heroin or crack users is

          It’s pretty easy to measure how much of each drug is consumed by measuring metabolites in sewage, or even rivers. There are lots of easy things people don’t do.

          • Protagoras says:

            I don’t know to what degree sewage is centralized, but unless it is extremely centralized there will be some difficulty finding representative samples for large communities as a whole.

          • Nornagest says:

            It’s pretty centralized in urban areas — usually it all goes to a small number of treatment plants. Often just one for small to mid-sized cities.

            In more rural areas, you start seeing distributed solutions like leach fields and septic tanks, but those by definition aren’t large communities.

      • Douglas Knight says:

        claim that alcohol is more dangerous than heroin and crack

        Which he didn’t claim.

        It is striking how every time anyone defends the firing of anyone for saying anything, they lie about what was said. Why don’t people stick to the original lies by the firing agent? Why do they make up new lies?

        • Nancy Lebovitz says:

          I doubt it’s a conscious process. I’m betting people just get dominated by their preconceptions.

        • theory says:

          Uhh, what? Why are you calling me a liar when he literally says this, over and over again in describing his study?

          Professor David Nutt ‘Alcohol is More Dangerous Than Heroin’ – A Royal Hangover: Interview Clip – https://www.youtube.com/watch?v=qU886JT8734

          Alcohol ‘more harmful than heroin or crack’ – https://www.theguardian.com/society/2010/nov/01/alcohol-more-harmful-than-heroin-crack , opening statement “Alcohol is the most dangerous drug in the UK by a considerable margin, beating heroin and crack cocaine into second and third place, according to an authoritative study published today which will reopen calls for the drugs classification system to be scrapped and a concerted campaign launched against drink.”

          David Nutt: Alcohol More Dangerous Than Crack – https://www.cbsnews.com/news/david-nutt-alcohol-more-dangerous-than-crack/

          How scientists rank drugs from most to least dangerous — and why the rankings are flawed – https://www.vox.com/2015/2/24/8094759/alcohol-marijuana “Still, Nutt is confident that alcohol would be ranked most dangerous in the US. “I don’t see how it couldn’t be, really,” he said.”

          I can go on and on and on. So yeah, let’s ask:

          It is striking how every time anyone defends the firing of anyone for saying anything, they lie about what was said. Why don’t people stick to the original lies by the firing agent? Why do they make up new lies?

          I might ask this of you. Why are you lying, when it is so easily disproved with Google?

          Your absolute best case scenario, giving you every benefit of the doubt, is that you think that David Nutt only said alcohol was more harmful, not more dangerous. That isn’t true, because he uses the word “dangerous” in interviews all the time as a motte-and-bailey, but it is true for his study. But that is an awfully nuanced and subtle distinction that, 99% of the time (as shown by all the press around his article and his own words) is interpreted to mean that alcohol is safer than cocaine.

          • Douglas Knight says:

            Thanks for the links!
            I was wrong both on the facts, and to call you a liar.

            But I’m pretty sure that you are also wrong when you say that he said this as drug czar and with the implication that it lead to his firing. All of your sources are afterwards. Since he actually did say it, eventually, it’s a reasonable error to make, but it is an error. Since the study was started by the advisory council, maybe his boss knew it was coming, but if you want to claim that this was the secret reason he was fired, that’s a different claim, requiring very different evidence. (If it’s so egregious, why would the minister hide his real reason?)

            Finally, the article does not consider absolute harm, but harm per use.

      • Loris says:

        (This is a re-post of something which didn’t seem to get published. I don’t know why it didn’t go through, and since I saved a copy I thought I’d try again.)

        I think you’re doing him and his research a disservice. The claims made arn’t even what you state: “alcohol is more dangerous than heroin”. (edit – note that this was written before you posted links to other people’s articles making that claim.)

        In fact it states “heroin, crack cocaine, and metamfetamine were the most harmful drugs to individuals, whereas alcohol, heroin, and crack cocaine were the most harmful to others”

        I think you’re wrong about the normalisation issue (i.e. disregarding that more people use alcohol than heroin) as well. The Lancet article describes the method as, for each criterion: “Drugs were scored with points out of 100, with 100 assigned to the most harmful drug on a specific criterion. Zero indicated no harm.” And looking at figure 4, which shows the contribution to the score of each drug of each criterion, we can see that drug-specific mortality of alcohol is (very roughly) about half that of heroin. I found this data with a little searching:

        Deaths involving heroin and/or morphine doubled between 2012 and 2015 from 579 to 1,201, and these 2 substances remain the most commonly mentioned opiates in drug-related deaths.

        In 2015, there were 8,758 alcohol-related deaths in the UK, an age-standardised rate of 14.2 deaths per 100,000 population.

        (my emphasis, to show values used below)

        We don’t need to try to figure out what the numbers of alcohol and heroin users in the UK are to see that your claim that they haven’t normalised their data is false, since 1/2 (alcohol_mortality/heroin_mortality) is nowhere near an approximation of 8758/1202.
        Unless, that is, you think that they just made stuff up – which you do seem to imply, by calling it a “survey of 15 people”. My interpretation of the article is, however, that they used data for their scoring as far as possible. (The article does not, however, include any input data.)

        The caveat you mention: “Many of the harms of drugs are affected by their availability and legal status, which varies across countries, so our results are not necessarily applicable to countries with very different legal and cultural attitudes to drugs.” I think has a very different meaning to the one you’ve ascribed to it. It’s effectively saying that the risks associated with a drug will be higher if you have to buy it illegally – and this means you can’t generalise the results to other countries.

        Now you could argue that combining lots of values into a single metric will conflate issues. I think that is a valid concern – but it’s a different discussion.

        One last point. You say he did this “from a government platform”. Actually he didn’t. He did it in his capacity of Professor of Neuropsychopharmacology, Imperial College London. This was reported to be in line with governmental guidelines.

        • theory says:

          So, first, he definitely does say alcohol is more harmful overall than drugs. The study itself says so, in the very next sentence after your quote:

          Overall, alcohol was the most harmful drug (overall harm score 72), with heroin (55) and crack cocaine (54) in second and third places.

          And he promotes it in his interviews to media outlets, as described above.

          Second, his data is not normalized, again because the study says as much. But if you don’t believe me, I suggest you read his interchange with Caulkins et al. Caulkins criticism is here: http://faculty.publicpolicy.umd.edu/sites/default/files/reuter/files/basing_drug_scheduling_decisions_on_scientific_ranking_of_harmfulness.pdf Nutt’s response is here:

          Caulkins raises the same point I do. Nutt agrees with it but says let’s not let the perfect be the enemy of the good.

          Third, I think you completely misunderstood the methodology. They don’t use data as an input into their harm results. Their harm results are generated by a survey of a group of people selected by David Nutt (including, and I cannot stress this enough, David Nutt himself). The survey consisted of asking them “how much harm do you think X causes” and “how much weight do you want to assign to X”, and explicitly asks them on an absolute and not relative level. (See, e.g., “(at the population level, not the individual level)”)

          Like, seriously, just imagine President Trump creating a study like this to justify outlawing marijuana. Where he appoints himself and Jeff Sessions and a group of friends to be surveyed on how harmful marijuana is, compared to inhaling coal fumes, and concludes that marijuana is far worse because way more people in NYC smoke marijuana than inhale coal fumes. There is absolutely no way that’s acceptable, even if it produces an outcome that we like.

          Finally, I reference “government platform” because the post was describing why he was correctly fired. Not because of his position on drugs. I happen to sympathise with his position. But because of his misleading garbage science for his own advocacy purposes.

          • Loris says:

            So, first, he definitely does say alcohol is more harmful overall than drugs.

            Rather than quibble about this (ie. that environmental damage is included as a harm, etc), I think it’s probably better to concede it for the sake of argument, because I think this is your real objection.
            Have you considered that alcohol might actually be more dangerous than heroin?

            I’ll come back to that later, because I think it’s worth going over the normalisation issue.
            I looked at the “Caulkins criticism” paper you linked to.
            If you’ll forgive a brief interlude for snark:

            Being hit on the head by a meteorite is highly harmful to the individual, but meteorite strikes are a negligible source of population-level mortality.

            Given that asteroids (which are big meteorites waiting to happen) are known to have caused mass extinctions in the past, and resources are being spent against this risk, this seems a rash statement.
            The principal objection seems to be the aggregation of various forms of harm, in particular for the individual and aggregated over the population. They state essentially that this single metric is not appropriate for deciding policy.
            Nutt’s responce to this is to point out that they did present a two-dimensional matrix (true – it’s fig.3 of the paper), and that yes it’s ‘lossy’, but (I wildly paraphrase) essentially however you create evidence based policy, you’re going to be making decisions based on some sort of function which generates a single result.
            They also only consider prohibition as a legislative option, when Nutt’s favoured option (for alcohol, at least), from what I’ve seen, is to increase costs through tax.
            They also complain that other countries would be different, apparently disregarding (or perhaps misinterpreting) the qualifier I covered in my previous comment.

            Third, I think you completely misunderstood the methodology. They don’t use data as an input into their harm results. Their harm results are generated by a survey of a group of people …

            I still don’t see this as a survey. The article is at pains to emphasise the discussion which entailed and the iterative nature of the process. And my impression is that this involved interpretation of data as far as was available. The issue being that there sometimes isn’t one universal, reliable dataset, and sources may conflict with each other.
            Later in the paper, there’s a big section going over the correlation of their results with prior work, and where discrepancies are found they try to explain them. For example:

            We also investigated drug-specific mortality estimates in studies of human beings. These estimates show a strong correlation with the group input scores: the mean fatality statistics from 2003 to 2007 for five substances (heroin, cocaine, amfetamines, MDMA/ecstasy, and cannabis) show correlations with the ISCD lethality scores of 0·98 and 0·99, for which the substances recorded on the death certificates were among other mentions or sole mentions, respectively.

            Like, seriously, just imagine President Trump creating a study like this to justify outlawing marijuana. Where he appoints himself and Jeff Sessions and a group of friends to be surveyed on how harmful marijuana is, compared to inhaling coal fumes, and concludes that marijuana is far worse because way more people in NYC smoke marijuana than inhale coal fumes. There is absolutely no way that’s acceptable, even if it produces an outcome that we like.

            Trump is… not my problem. I imagine, however, that such a study would fit in 140 characters.
            If Trump and collegues went off and looked at the data on inhaling coal fumes, I imagine that they’d soon find that inhaling coal-gas had quite a high mortality rate, and advise people against. But since noone in America was particularly doing that any more, no additional legislation was required to that effect.

            On the other hand, would Trump need to look at the evidence to ‘justify’ outlawing marijuana? Let’s suppose he did, and he published graphs showing how dangerous marijuana was. The data is either relevant, or not. It’s also either true or not. In the event it was relevant and true, I’d hope people would commend him for his work. If not one or the other, I’d hope people would point that out. But then, that’s true for everyone.

            So, going back to this idea of alcohol being more dangerous than heroin. Firstly it is actually pretty dangerous to the individual. It kills people quickly, of alcohol poisoning. It kills people slowly, of cirrosis and heart disease and other chronic diseases. It’s also pretty dangerous to others, because of violence and accidents (car crashes etc). Now, I think heroin is also pretty bad for both, in somewhat different ways (danger to others would mainly be burglaries and muggings I imagine), but it’s not clear to me that it’s worse; I just don’t have the data.
            But as I see it, that’s what Nutt’s work was to try to do, and I think that’s worthwhile. It would be nice if the raw data was accessible – where they got estimates from. If I cared enough I could write and ask, I suppose.

        • Nornagest says:

          Being hit on the head by a meteorite is highly harmful to the individual, but meteorite strikes are a negligible source of population-level mortality.

          I read a paper on this yesterday, actually! That was on a different system and I don’t feel like digging up the cite, but the gist of it was that on an annualized basis the expected damage from meteorite or comet strikes is 10-100 deaths and a few million dollars in property damage, largely because of impact-induced tsunami effects.

          All the damage happens way out on the tail of the distribution, but we can deal with that.

    • beoShaffer says:

      The tipping point that got him sacked appears to have been his claims about the risks of horse riding being in the same ballpark as those of taking MDMA

      So I guess he should have kept the yokes?

  2. Goldragon979 says:

    One of the most compelling theories on serotonin and dopamine I have come across is that of Y-Lan Boureau and Peter Dayan (http://www.nature.com/npp/journal/v36/n1/full/npp2010151a.html). They argue that DA codes apetite/desire and invigoration, whereas serotonin codes for aversion and inhibition. They acknowledge that it is simplistic and offer no hypothesis for the different roles of serotonin other than it is evocative of a ‘hierarchical regulatory system’. In any case, I feel it is a good complement to the hypothesis here presented.

    • hypnosifl says:

      This sounds like it could be connected to the notion of distinct approach vs. avoidance subsystems in the brain, with studies indicating that ‘approach’ involves more left-hemisphere activity and ‘avoidance’ involves more right-hemisphere activity. And just googling a little, this paper claims some evidence that seratonin preferentially activates the right hemisphere…but I don’t think they made a distinct investigation of 5-HT1A vs. 5-HT2A, would be interesting to know if there was any difference in the lateralization of activity associated with each.

  3. Peter says:

    Norepinephrine – I did a google search for ‘norepinephrine “the * chemical”‘ and various of the hits highlighted things saying that dopamine was the reward chemical and so on but nothing for that compound itself.

    I say “that compound” because I speak British English and we say “noradrenaline”. We don’t seem to have that odd thing where informal American English says “adrenaline” and technical American English says “epinephrine”, it’s adrenaline all the way over here. Etymologically it’s all the same. Anyway, “noradrenaline” makes it more obvious that it might be kinda like adrenaline, and structurally and physiologically you’d be right, and adrenaline, well, it’s “the adrenaline chemical”.

  4. Yashabird says:

    Regarding John Lilly’s ketamine-induced synchronicity fetishism/paranoid delusion, I tried googling ketamine’s effects at the 5-HT2A receptor, but I couldn’t find anything conclusive. If anyone can find a Magnetic Resonance Spectroscopy study on the downstream effects of ketamine on serotonin distribution in the brain, this might prove useful.

    Otherwise, I think it’s just easier to theorize that ketamine produces a higher “information-theoretic entropic” brain state simply by agonizing at AMPA receptors in the cortex. After all, visual hallucinations look pretty different between dissociatives and classic psychedelics, and there are probably lots of different routes to fiddle with the entropy of brain states.

  5. Chebky says:


    So, G64 Paxilon HCl “Pax” is a very intense 1A agonist, causing people to get locked it one thought pattern and “stoically” stand in the face of hunger and, uh, death, while a small fraction of the population have their 2A receptors susceptible, and thus – Reavers. (Although Reavers seem to still have moderately predictable social behaviors, they’re just very – though cooperatively – violent. Maybe the effect was a a kind of group psychosis branding all other humans as enemies?)

  6. Nancy Lebovitz says:

    Algernon’s Law: Why is it difficult/impossible to increase intelligence, but fairly easy to increase strength?

    • baconbacon says:

      It is fairly easy to increase knowledge, which is probably the analog to strength.

    • gwern says:

      It’s easy to increase strength because you are far from maximal on muscle mass or efficiency; for example, there’s a simple myostatin mutation which will drastically increase your musculature, and exists spontaneously in humans (you might’ve see news articles a few years back about a ‘super baby’ or ‘muscle baby’) and can be edited into pigs and dogs with CRISPR. Why isn’t it universal? Because muscle is metabolically expensive in calories and proteins – your body is constantly trying to shed as much muscle as possible. This is why bodybuilders and weight lifters are constantly fighting their bodies and homeostatic limits; stop lifting or roiding, and your body immediately begins liquidating all the excess muscle. (Some relevant anecdotes in “The Power and the Gory”, which is also a great read: http://www.bronxbanterblog.com/2013/10/01/the-power-and-the-gory/ ) Or the tradeoff in primates between muscle efficiency and apparently brains. Similar issues for astronauts in space or aging women: there’s no inherent reason for bone or muscle mass to be liquidated, but it’s more efficient and (used to be) more adaptive.

      • baconbacon says:

        Also related, very simply if you took the strongest* species and attempted to improve strength you would probably have a hard time increasing it in them. It is not unsurprising that if you take the smartest species that increasing intelligence is harder than increasing something that they have sacrificed for intelligence.

        *however you are defining that

        • gwern says:

          You might have a harder time breeding horses than cows or something, but I wouldn’t say it’s that hard in any absolute sense. Wild horses are some of the fastest animals around, yet breeding wild horses into thoroughbreds yields quite a bit of speed. Similarly, there’s a ton of genetic variants affecting intelligence in humans and so you know you could increase human intelligence straightforwardly by simply breeding for it.

          The point I/Yudkowsky/Bostrom make is that these interventions obey the Algernon’s law: the effects are small, the intervention may be simple but the mechanism is far from simple (increasing/decreasing slightly thousands upon thousands of genetic variants), and they probably come at considerable fitness in the wild.

          (Thoroughbreds are very fast indeed, but you don’t need to watch one being euthanized on the racetrack after falling to know that they would not do very well in the wild… Human intelligence is a little different, as I increasingly think the problem there is not tradeoffs changing but actually mutation load – ie evolution being too slow to optimize intelligence, the advantages of intelligence being too small to cope with the constant introduction and random spread of bad mutations.)

      • Inside a semicircle of displays says:

        It’s easy to increase strength because you are far from maximal on muscle mass or efficiency

        In the same vein, probably the only people putting nearly as much effort into increasing their physical fitness as most people do regarding intellectual ability (education alone) are top athletes, and you can see some sort of natural ability (and use of illicit substances) making the difference there – seems fairly similar to intellectual pursuits to me.

      • nhnifong says:

        Could we we be smarter if our brains were in soft bags like cephalopods, and there wasn’t a barrier (other than metabolic load) to increasing brain mass?

        Likewise, if our muscles were encased in solid tubes of bone like crabs’ we might find it harder to quickly grow stronger.

      • Nancy Lebovitz says:

        Do smarter people’s brains use more energy than stupider people’s brains? Maybe we can stop wasting time with IQ tests!

        • bbartlog says:

          As I recall they actually studied this and found that, to a first approximation, the answer is no. Smarter people mostly have more efficient brains. There is no doubt some residual difference; after all, even with constant power density, you still have bigger brains in smarter people and so they’ll be using more energy in total. But for the most part, being smarter is not some kind of overclocking phenomenon.

          • Nancy Lebovitz says:

            So it probably isn’t reasonable to say that people can’t afford to be smarter.

          • LadyJane says:

            “So it probably isn’t reasonable to say that people can’t afford to be smarter.”

            It would be possible to optimize the human brain so that, for instance, everyone had an IQ of 160. But it might not be possible to optimize people to be any smarter than the smartest people today, at least not without severe trade-offs.

      • vV_Vv says:

        Why isn’t it universal? Because muscle is metabolically expensive in calories and proteins – your body is constantly trying to shed as much muscle as possible. This is why bodybuilders and weight lifters are constantly fighting their bodies and homeostatic limits; stop lifting or roiding, and your body immediately begins liquidating all the excess muscle.

        But why doesn’t the same apply to brains? If human intelligence was limited by some simple metabolical tradeoff, then why isn’t there a mechanism that increases intelligence with use and decreases it with lack of use, as there is for strength?

        The mutation load hypothesis seems more plausible, but this implies that it should be possible to obtain significant gains (up to the smartest naturally occurring healthy human) by interventions. Certainly by breeding, and possibly to some extent with drugs.

    • willachandler says:

      Algernon’s Corollary Conundrum  Given that IQ and BMI alike are epidemiologically stable traits, how can it be scientifically rational to regard obesity as effectively treatable, but not stupidity?

      • Eponymous says:

        There’s at least one highly effective treatment for obesity: bariatric surgery. I don’t know of one for low IQ.

        (But yes, when I learned that twin studies find that BMI is about as heritable as IQ, that required some adjustment of my worldview.)

      • bbartlog says:

        ‘epidemiologically stable’ is a bit of a dodge here though, isn’t it? Yes, there are no easy ways to treat obesity in a population of people who are mostly running around loose in the first world – but if you change the setting to one where you’ve got lab animal levels of control, suddenly obesity is fairly treatable while you still have no practical way to treat stupidity.

        • Nornagest says:

          Don’t feed the troll.

        • willachandler says:

          bbartlog asserts [debatably]  “If you change the setting to one where you’ve got lab animal levels of control, suddenly obesity is fairly treatable while you still have no practical way to treat stupidity.”

          Perhaps many SSC commenters will agree with this proposition … yet to the best of my knowledge and understanding, no very strong body of scientific evidence supports this proposition, whereas considerable scientific evidence contradicts it.

          For example, “lab animal levels of control”, applied in service of increasing linguistic ability, have been found to remarkably increase the cognitive capacities of nonhuman primates.

          Obviously, ethical considerations obstruct scientists from applying “lab animal levels of control” to humans. Still the extended, all-controlling training programs of the USMC are widely believed to effectively increase the cognitive capacities for which “g”-measures purportedly are surrogates:

          [T]he grassroots of our country believes … that our Corps is downright good for the manhood of our country; that the Marines are masters of a form of unfailing alchemy which converts unoriented youths into proud, self-reliant stable citizens — citizens into whose hands the nation’s affairs may safely be entrusted.
            — USMC Gen Victor Krulak, 1957

          In summary, the USMC’s extended, all-controlling training programs are demonstrably effective in reducing BMI by remodeling the associated musculoskeletal connective tissues. It therefore is reasonable (isn’t it?) to enquire, whether there are there any substantial scientific grounds to reject the proposition, that comparably extended, all-controlling training programs can be comparably effective, in increasing IQ by remodeling the associated cognitive connectome tissues.

          Needless to say, “SAT Book Camps” are pallid reflections of real boot camps … perhaps this is why they are only partially effective? With broader regard to real-world IQ plasticity, SSC commenter Janet’s recent remarks (of Sept. 28, 2017) are admirably well-considered (as they seem to me).

    • Eponymous says:

      A key question here is how complicated human intelligence is.

      The human muscular system is not all that complicated. Our genes say where to put muscle fibers and how big to make them. Then stresses cause damage to muscle fibers, the body overcompensates in its repairs, and makes muscle fibers larger (and therefore stronger) to deal with higher stresses. The size of the response is governed by chemicals (including steroids / HGH) and available nutrients.

      Thus natural variation is due to base size, and your body’s natural production of these growth factors and how it responds to them. You can increase strength by increasing stresses (working out), making sure you get enough nutrients, and by injecting synthetic steroids.

      Is intelligence similar? Is there a simple quantitative factor analogous to “size of muscle fibers”, like “number of neurons” or “number of neural connections”? And is there a neural growth factor analogous to steroids? Then are differences in IQ just determined by how much your brain responds to the stresses of working on mental tasks?

      If it’s just that, and the main costs of high IQ are metabolic, then maybe there is something akin to synthetic steroids, a synthetic neural growth factor, that would allow people to bulk up their IQs, at least to the upper bound of normal human variation, and a bit past it.

      Incidentally, my money is on human intelligence not being that complicated — that we basically just have chimp brains with more neurons, in much the same way that gorillas basically have human muscles with larger muscle fibers.

    • Douglas Knight says:

      Humans gain and lose muscle over the course of their life. Probably farmers cycle through it every year. Hunter-gatherers maybe more erratically. So there is some switch to gain muscle. All you have to do is turn it on. Whereas, humans largely do not gain and lose brain. As Eponymous says, the brain is complicated, so building it up is a lot more expensive than adding cells, so it isn’t worth cycling it down and back up. Of course, you could ask, why are muscle and brain different in this way. Why is one simple and one complicated? Probably more calories in childhood, the one important brain-building phase, does lead to more intelligence, although we’ve probably saturated this pathway.

      • bbartlog says:

        Brain building likely has more to do with fat consumption, just as muscle building requires protein. Of course you also need enough calories to survive, but adding more calories without the requisite fatty acids is not going to do much of anything for brain growth.

  7. Garrett says:

    > platelet clumping

    Wait? What? Everything until then made some degree of sense to me, at least in that everything was focused around experience of the brain. But why would platelet clumping, involved in the blood clotting process, be involved? Platelets aren’t even whole cells, exactly. What’s up with that?

  8. pistachi0n says:

    90% of the serotonin in the body is secreted by EC cells (cells that line the intestinal tract and coordinate with the enteric nervous system, especially peristalsis). Most signals for serotonin release come from stimulation of EC cells in the gut. There are a few chemicals that consistently trigger serotonin release when detected by EC cells, and mechanical stress also triggers serotonin production and release, which is how poop moves.

    I don’t actually know a lot about the brain– I study gut serotonin and the microbiome–but I just want to put this out there if we’re talking about the mystery of serotonin.

    • Lambert says:

      Am I correct in thinking that serotonin can’t cross the blood brain barrier, so there’s no direct relation between those two sources of serotonin?

      • pistachi0n says:

        Serotonin can’t cross the blood brain barrier, but 5HT and tryptophan can, both of which are serotonin precursors that are regulated by EC cells, the enteric nervous system, and probably the gut microbiome.

        • cuke says:

          A related question… I understand GABA also doesn’t cross the blood brain barrier even though it’s sold as a supplement for calm. Is it possible that with so many neurotransmitters in our gut that these chemicals don’t need to cross the blood brain barrier in order to have an effect on mood? Or is that not how they work? Like does serotonin in the gut improve mood? Does a damaged gut contribute to mood problems in part because it can’t hold onto serotonin as well as a healthy gut? Do we know the answers to these questions?

        • cuke says:

          Oh and one more: if SSRIs produce neurogenesis in the brain, do they do that in the gut as well?

  9. timujin says:

    I am on 5mg of Zyprexa daily. How long should I abstain from it, until I get it out of my system, so I could trip on LSD?

  10. c0rw1n says:

    Scott, have you ever thought that maybe Algernon’s law is a very stupid heuristic? You know that our brains have evolved under constraints that no longer apply (like limited availability of food), and actively try to minimize the amount of energy they have to spend to predict their environment accurately and not just by learning the best possible models, but the best possible models constrained by how tolerably short they are to discover/learn and also stop caring about refining the models once they seem good enough.

    Algernon’s Law is an observation that evolution does not optimize for the things that an intelligent system with human values would optimize for, what with evolution being a basically random walk with very short steps. It is not a hard limit of what biology can do even in theory.

    • Saint Fiasco says:

      But you still have to figure out which old constraint that no longer applies is involved. For example, if we figure out that keeping some receptor at maximum makes you smarter but reduces fine motor dexterity, knock yourself out. But if we mess with receptors without knowing what exactly they do and why they are not maxed all the time, we might end up with all sorts of weird platelet clumping disorders.

    • Michael Arc says:

      There’s an obvious exception to Algernon’s law for anything that increases the consumption of energy or of any scarce metabolite which is abundant for us, but that doesn’t seem likely in this case. The largest single-gene boosts to IQ seem to all be associated with costly side-effects. Also, intelligence in general seems like it might be associated with depression.

      • Nancy Lebovitz says:

        usually, when people talk about a connection between intelligence and depression, it’s followed up with “of course– it’s because the world is so awful”. You didn’t do that, but now I’m wondering whether there’s a connection which has nothing to do with what the world is like, it’s just some physical factor in the brain.

        Also, it isn’t obvious that intelligence in this world would lead to depression– it’s as plausible that intelligence would take the edge off of pain.

        • Michael Arc says:

          I’m assuming that intelligence is a measure of strength and selectivity of inhibitory mechanisms, but that general inhibition leads to an over-all increase to our measures of intelligence.

        • hlynkacg says:

          it’s followed up with “of course– it’s because the world is so awful”.

          …and here Scott was just discussing different sorts of coping receptors.

          It occurs to me that this sort of thing would be a natural candidate for one of the things that “intelligence” is traded against. It’s a classic evil genie gimmick. Sure, I’ll rewire your brain to make you smarter but the process will also make you miserable and/or psychotic.

        • Antistotle says:

          The world is not awful, but as Douglas Adams (RIP) pointed out, the one thing a man *cannot* have is a sense of perspective:

          “Have some sense of proportion!” she would say, sometimes as often as thirty-eight times in a single day.
          And so he built the Total Perspective Vortex — just to show her.
          And into one end he plugged the whole of reality as extrapolated from a piece of fairy cake, and into the other end he plugged his wife: so that when he turned it on she saw in one instant the whole infinity of creation and herself in relation to it.
          To Trin Tragula’s horror, the shock completely annihilated her brain; but to his satisfaction he realized that he had proved conclusively that if life is going to exist in a Universe of this size, then the one thing it cannot afford to have is a sense of proportion.”

          The thing about being, say, in the 2nd and 3rd standard deviation portion of the bell curve is that you know pretty much how intelligent you are, but more importantly how intelligent you *aren’t*. You also start to understand why so many problems haven’t been solved, which is to say they are REALLY difficult to intractable.

          And you also interact with enough people who are both “normal” and have no interest in the complexities of the world who insist it’s *simple*. You just build a wall, you just ban guns, you just put the government in charge of health care, you just make all drugs legal, you just give drug dealers the death penality, and they know this one guy who was gay and got “cured” and if you’re not in favor a law MANDATING that you call trans people by their preferred pronoun you are are RACIST NAZI.

          Which either has you in the back room punching your bore and loading magazines, or in the front room popping thorazine and reading magazines.

  11. jebbyderinger says:

    Sounds like I have a 1A problem or SCS (Stoic Coping Disorder). Guess I need some psychadelics 😉

  12. benwave says:

    Is it reasonable to ask questions such as “Why not set the anti-depression dial to maximum and leave it there” when it concerns the results of evolution? I don’t think there are good reasons to assume that ‘dialing up’ whatever process has evolved to play the role of suppressing depression would actually result in more suppression of depression, or that it would Only have that effect. Puts me in mind of asking ‘teeth help you hunt and eat better. Why not just have bigger teeth?’ and in this case we have the sabre-tooth tiger and friends to help illustrate the result of that.

    I guess what I’m trying to say is that brains don’t have anti-depression widgets. They have hacked together organs and structures, some of which have the effect (among others) of suppressing depression when used in particular ways. Capacity for being increased is not something we should necessarily expect from thusly evolved structures.

  13. sheelba of the eyeless face says:

    2A activation as loosening an “idea/thought filter” makes sense. This would be adaptive in the active coping context, explain the sense of profound insight (and variable degree of actual quality of insight) in recreational 2A agonist trips, and why blockers can have antipsychotic action. Might even explain some of the non-figuative visual effects of trips as well – a still object “flowing”, trails for the paths of moving objects, etc as thoughts about motion that are normally filtered out.

  14. The Nybbler says:

    The brain has to be too complicated for poking around rather bluntly on a single-digit number of receptor types to have good effects, right? Not just Algernon’s law, but it seems to me like trying to improve the functioning of an electronic device by reducing the value of all the 10k resistors or something.

  15. Michael Arc says:

    SSRIs cause Hippocampal growth, and their effect on depression seems to happen subsequent to that growth, not immediately. Stress causes hippocampal shrinking, so there’s a logical connection there. It’s generally difficult to accurately remember what happens during a psychedelic experience. There might be a connection there.

    My guess is that stress response calls for reacting to threats in real-time, while the hippocampal consolidation of memory takes a longer period and is thus neglected. 5-HT1A down-regulates stress, enabling it to be ignored and enabling memory consolidation to recover. 5-HT2B up-regulates it and enables accurate awareness of the situation rather than the suppressed awareness that is usually maintained in order to preserve belief in a just world.

  16. J says:

    A passive/active coping model suggests that bipolar could be the result of an under-damped control system.

    • Douglas Knight says:

      Bipolar sure sounds like a parameter bouncing between two poles, but does it sound like this system? This control system is about psychosis, not mania.
      (I also wouldn’t be so specific as to suggest underdamping.)

  17. Montfort says:

    I’ve heard just-so stories about alleged benefits of depression keeping it around when it (naively) seems like it should’ve been selected out of the gene pool if possible. Not that I really believed them, but if depression does serve a purpose, evolutionarily speaking, then there’s an obvious reason we don’t just keep the anti-depression receptors at max all the time.

    • hlynkacg says:

      Remember that the “purpose” may in fact be a trade off where “x% chance of depression” is the price being paid for some other trait that matters a lot more in the evolutionary scheme of things.

  18. TomA says:

    You seem to be asking . . . why does our brain work this way?

    The answer is that what exists is what works (as in, an evolved trait based upon the long-term evolutionary history of our species). If our mental processing incorporates a two-step, sequenced hierarchy of behavior, then that is what maximized our ability to survive and thrive in the ancient ancestral environment of our predecessors.

    • leoboiko says:

      Evolution doesn’t maximize things; it just cuts off that which isn’t good enough. Anything that works well enough to survive and reproduce will survive and reproduce, no matter how clunky or far from the optimum (both of which are super plentiful in nature).

      • toastengineer says:

        No, it maximises too. You’ll have as big a litter size as you can get away with; sure having fewer would be “good enough” but the strain that has six pups will overtake the strain that has five pups, assuming there isn’t some significant drawback to the extra pup.

  19. chridd says:

    I don’t know if I’m not understanding something correctly, or other people aren’t, or what, but this confuses me every time I hear about the subject (also this kind of relates to The Nybbler’s comment above)… Why would a neurotransmitter have a single function? (Or, at least, why is that an assumption people make?)

    My understanding is that neurotransmitters are for transferring information across a single synapse, seems like it should make them more like a 1 bit in a computer than some overall “happiness” or “confidence” or whatever variable. A 1 bit represents very different things depending on where it is in memory; in one place it represents part of my mouse cursor location, in another it represents part of a character in a document, in another it represents the color value of a pixel, etc.; so by analogy, why wouldn’t the same neurotransmitter represent “red somewhere near the right side of my field of vision”, “I want to move my little finger now”, and “that sounds kind of like a duck”, depending on which exact synapse it’s in (or something along those lines)?

    (If my understanding is correct, then it would provide another possible answer to “why would there be a receptor that if you stimulate it enough makes you psychotic”: artificially increasing neurotransmitter receptivity like that seems like it should be the equivalent of telling the brain things like “every possible sense input is occurring” or “everything is happening right now” or “you are now experiencing all possible emotions at once”… just like artificially making more bits in RAM 1 is going to corrupt stuff, even though some bits should be 1)

  20. Kevin says:

    In the optimization literature, one of the central problems is often characterized as “exploitation vs exploration” or “earn vs learn” and is frequently modeled as the multi-armed bandit problem (MAB).

    See https://www.cs.mcgill.ca/~vkules/bandits.pdf for an overview.

    The basic setup is that you walk into a huge casino with a bunch of slot machines and are told that some of them are positive expected value to varying degrees. The challenge is to pull levers to maximize returns. Obvious issues are: (1) a losing bandit randomly pays off early so you think it’s a winner, (2) a winning bandit has a bad run early so you think it’s a loser, (3) you find a winner but there are better winners.

    It’s a really hard problem with a literature that goes through a lot of dense math to reach the conclusion, “this is a really hard problem.” But one of the conclusions is that you always have to allocate _some_ resources to earning and some to learning. Then there’s another arm (sorry, couldn’t help myself) of the literature with papers that say, “Look, here’s a real world problem that behaves a lot like the MAB model.”

    Then there’s the variation where payoffs change over time, called the Restless Bandit. Now, this sounds like a really good approximation of most non trivial real world problems. Unfortunately, there’s a paper that shows determining optimal play for this problem is PSPACE hard (probably harder than NP): https://arxiv.org/pdf/0711.3861.pdf.

    So to optimize your personal outcome, you have to optimize a bunch of problems that are well modeled by a nigh unoptimizable problem. This certainly sounds like a situation where you’d want a mechanism for varying the propensity for exploitation vs exploration. Moreover, it would explain why you’d see a lot of variation in peoples’ genetic/epigenetic tendency to do one vs the other.

    • taktoa says:

      I’m sure you know all this, but for anyone else reading I thought it might be worth pointing out that, as stated in the paper, it’s not just determining optimal play for the Restless Bandit that’s PSPACE-hard, it’s PSPACE-hard even to just approximate optimal play. Note that this says nothing of greedy play (i.e.: play that attains a local optimum, even if it is not the globally optimal strategy), and it also may be the case that the Restless Bandit is a strictly more general problem than the instances of it that we see in the real world, in which case these complexity bounds may be relaxed for much the same reason that SAT solvers are useful in industry despite 3-SAT being NP-hard (the problems we give to our SAT solvers are not simply randomly chosen from the set of all formulas in propositional logic, and so there are regularities that can be exploited in the form of heuristics). The fact that humans regularly fail at making the right choice in explore vs. exploit tradeoffs does indicate that this is fundamentally hard, though (respectively, humans are pretty good at doing math, which is why the SAT solver thing makes sense).

      • Kevin says:

        Yep, just didn’t want to split hairs for people who don’t care about the details of computational complexity.

        After thinking about it, I believe there’s a strong parallel here with the GABA signal processing system.

        You have a hard processing problem with a clear tradeoff. So you use the same signalling molecule and two different receptors coding for the two directions of the tradeoff.

        I wonder if something codes for collision in the serotonin system like dopamine does for GABA. Actually, I wouldn’t be terribly surprised if it were dopamine as well. Why use a different tolerance mechanism if you can just reuse one?

        Does dopamine correlate with bipolar disorder? I could see how mania (or at least hypomania) could be characterized as an extreme form of exploration. The depression is caused by the failure of an exploitation strategy, so the brain tries to switch to an exploration strategy and goes too far.

        To your observation on Restless Bandit as an approximation, I’ve actually got a couple of thoughts on why it’s is an upper bound on the complexity of real world problems, as applied to the economics of the firm:


        Basically, if there’s some sort of probabilistic dependence structure among the bandit payoff distributions, you can try to exploit that structure.

  21. Antistotle says:

    > Active coping is like…an LSD trip? It’s some kind of grabbing the brain and shaking it, in the hopes that
    > maybe when it settles it will be in a state that’s better able to deal with whatever’s going on.

    Note that this ALWAYS worked when my tape player (Cassette, VHS. etc.) wasn’t work.

    Pick it up, shake it, plug it back in. PERFECETION.

  22. carsonmcneil says:

    Well, I want to start off by saying that I think it’s good that somebody has proposed this theory, and I understand that taking baby steps in understanding is an important thing to do.

    However, I do wish that in conversations about the function of a particular neurotransmitter, we stayed a bit more conscious of the fact that there really doesn’t have to be a describable (or even consistent across the brain!) high-level function at all. Neurotransmitter release in a natural setting is spatially precise, it happens at the level of a single synaptic release site. Neurons may have a tendency towards some kind of function based on their morphology, and the receptor types they express, but their connectivity matters also! There is no reason that seratonin release in one place in the brain has to encode at all the same thing as seratonin release in a different place. Even if those neurons were right next to one another and use exactly the same receptor types. The connectome is crucial!

    Essentially, I worry that asking “what is the function of seratonin in the brain?” is like asking “what is the function of electrons in a microprocessor?”. You can “increase the number of electrons” and it will do something, but whatever it does in terms of behavior is going to have much less to do with the physical properties of electrons and much more to do with how the microprocessor is wired up.

    Now, you can of course object that seratonin isn’t like electrons. Seratonin neurons make up only a very small percentage of all the neurons in the brain, and there are some nice stories that sort of seem consistent with the data we have. But I think it’s important to acknowledge that when we induce a brain-wide non-natural state (i.e. non-selectively changing the amount of extracellular neurotransmitter), the effect we get doesn’t have to have anything to do with the endogenous function of that neurotransmitter. The brain is highly nonlinear, and the fact that increasing the amount of seratonin dramatically causes hallucinations does not have to be at all related (in the “purpose of this transmitter” sense) to the effect of increasing the amount mildly.

  23. petterpan says:

    So what about SSRI being “passive coping”? Can we not check wether a group treated with SSRIs does better than an untreated group (prefferably a control group) 10 years later?

    If taking the medication is a passive coping option (reducing brain plasticity that needs changing for solving life problems), then we should expect to see a “worse outcome” of the SSRI group years later, should we not?

  24. Ranorith says:

    O God and Heavenly Father, Grant to us the A1 activation to accept that which cannot be changed; A2 activation to change that which can be changed, and the serotonin levels to know the one from the other. Amen.

  25. Reasoner says:

    Why have a receptor for treating depression and anxiety, rather than just always keep the receptor at maximum so you’re never depressed or anxious?

    Maybe depression and anxiety play evolutionarily useful functions?

    • Nancy Lebovitz says:

      It seems reasonable to me that depression is too much of the “not worth doing” impulse and mania is too much of “Great idea!”. They need to be balanced or moderated somehow rather than either one being knocked out of the system.

Leave a Reply