Personally I share this strategy but have trouble implementing it. e.g. last spring I noticed myself maybe getting depressed again and knew that I should take some action to stave off falling into further depression ASAP, since I knew from past experience that when I am more depressed than that it's much harder to take such actions; however, I failed to do that effectively and did end up getting more depressed.
IIUC, attractor states "emerge" out of other characteristics, like to what degree being tired makes you leave the house less. If something changes the degree to which being tired makes you work less (for example, you get a dog and you have to walk it regardless of how tired you are), then the attractor state will also change.
Thanks. Using this metaphor of balls and basins, what do we mean when somebody’s genetically prone to depression? Are the genetic factors just a few more variables or do we mean they are completely different dynamical systems (say with different update equations)? Your article is making feel like depression can only be managed but not cured :( Is that too hopeless?
>What do we mean when somebody’s genetically prone to depression?
"Genetically prone" can be thought of as having a bigger and/or deeper basin around depression. Which amounts to saying, there are more positions of the ball that roll into the "depression" attractor, and it's harder to get the ball out once it goes in.
> Are the genetic factors just a few more variables or do we mean they are completely different dynamical systems?
There's a bit of gray area between "more variables" and "different equations": you can always squeeze multiple models into one equation by just setting variables to 0 when you don't need them. I imagine that while most of the "mood equation" is shared by most people, there are some pieces that may be individual-specific, especially as you consider neuro-atypical people.
This puts me in mind of a 1991 essay, in which the philosopher David Stove bemoans our lack of an ontology of the disorders of intellectual philosophical thought. I would be very interested in how you would put together such an ontology.
So Alice lost her job and her insurance and she broke up with Bob which now keeps him inside all day and away from making big plays in commodities. I know it's a vast oversimplification, but the aliens should just materialize universal healthcare and give Bob a new girlfriend. The gallon of prevention that's worth a barrel of cure.
Is Bob's new girlfriend a real person, with the own hopes and dreams and capacity to desire? If so, would it be effective to "give" Bob a new girlfriend? Putting even the ethics aside, what if she broke up with him because the aliens guessed wrong about their compatibility?
Similarly, but far bigger and more complicated, "giving" everyone health care involves millions of choices and interactions per year. An alien species cannot just "give" everyone health care, because health care is provided by a series of people (doctors, nurses, etc.).
I'm not saying that either situation isn't fixable, but that mandating that the obvious immediate fix go into place doesn't mean it will actually work. Bob needs to find his own girlfriend.
In the hypothetical, Alice gets a job as soon as her health improves. (We ignore Medicaid for simplicity, but that leaves room for alien health care in the hypothetical).
But hypothetically we've determined that isn't the only outcome possible.
"But suppose Alice gets a serious pneumonia which knocks her health down to 70. Now she can’t work, she loses insurance, her health starts going down 1 each day, and she eventually goes down to 0. This is another stable state; as long as the system keeps evolving according to the rules, it will never change."
She could get to a point where health is not 0, but health also is not 75 (the point hypothetically needed to maintain job and health insurance).
Since we're talking about aliens with a materialization ray sophisticated enough to produce millions of BBL of West Texas Intermediate on the White House lawn, let's get Alice some healthcare and get Bob a nice real person (the materialization ray can produce real oil...I think they got this) so he can start shorting the market again and WTI goes down.
Free healthcare for Alice and a magic girlfriend for Bob might not be perfect solutions sure, but they might just be enough to get both of them out of their rut and rolling towards the axis of stability.
Joe feels sad sometimes so he goes to his room and refuses to talk to his friends. After a few days he feels fine and is ready to deal with people again, even if they are kind of assholes.
"SAM-e and tetrahydrobiopterin are cofactors for the enzymes tryptophan hydroxylase and tyrosine hydroxylase, which change the amino acids tryptophan and tyrosine into serotonin and dopamine, respectively. And serotonin and dopamine are known to be heavily involved in mood and energy level."
So is this why I sometimes get intense cravings for cheese? As in "I really need to eat some cheddar NOW"? Tyrosine in cheese -> dopamine?
Now you've got me wondering if this is why pizza is one of the only things that lifts my mood when I'm depressed. Nothing has ever tasted better than dripping, greasy cheese when I'm depressed. Looking at foods containing tyrosine too, cheese is one of the only foods I eat that contains much...
If Scott is right (and Scott is nearly always right), then it's just good old "fat plus carbs are tasty, render you drugged-out in bliss after consuming a ton of them, and if it's unhealthy your body will love it" rather than anything as redeeming as "I'm low on dopamine and just recharging" 😀
I once met someone who claimed that cheese was addictive, though they attributed it to dairy farmers adding morphine. Maybe there was a kernel of truth, but for the wrong reason.
The "help my toddler is a picky eater" internet genre definitely contains claims that dairy (and gluten) are effectively addictive. For example: https://healthhomeandhappiness.com/why-does-my-child-seem-addicted-to-dairy-andor-wheat.html I haven't seen the claim of added morphine, just that casein (and gluten) are sufficiently opioid-like to be addictive in their own right.
Now that I think of it, the morphine-cheese claim was probably a confused telephone-game version of the fact that casein does metabolize into an opioid: https://en.wikipedia.org/wiki/Casomorphin
I would be surprised if it was anything that simple. Almost any elegant theory with lots of good real life correlates in biology usually turns out to be wrong.
In writing about complex adaptive systems (CAS) we fall into a couple attractor states - either they're still tractable because of the inherent predictability of *some* macro patterns, or they're so complicated because of the number of moving parts that you're shooting in the dark. And yet, despite the billions of variables that affect each part of the economy, we do tend to repeat certain patterns. Even if inadequate that still has some explanatory power.
My contention is that to get to a dynamical understanding, simulations are criminally underused as explanatory vehicles, especially in business/ economic settings (medicine too, though I don't know enough here). We'd need them to help us figure out the right level of abstraction to understand these attractors and the overall shape. (I explored this recently again (https://www.strangeloopcanon.com/p/simulating-understanding). It remains the biggest annoyance I've had against economics since grad school. I want to see the "biochemical pathways" flow diagram equivalent for money, so we can run scenarios and hone assumptions. They'll help us get a better understanding of the strange attractors within the system, while analytical/ predictive models tend to focus more on getting to an answer, which fails in cases of CAS. (This might be a false dichotomy.)
The mind seems a far more complex system than the economy in this instance, and I wonder how you'd even go about the problem of trying to create/ simulate something to get an understanding here.
Don't simulations of large complicated systems have a butterfly effect type problem where if you get any of the parameters even slightly wrong, or miss a variable you shouldn't have missed, they tell you nonsense?
(now I'm realizing I must be at least sort of wrong about this, because we simulate the weather okay, but I don't know how much work that took or how replicable it is)
Weather may be a special case, because it's highly visible and very important to every day life. Also what rkris said about predictability due to macro patterns. In particular the macro pattern of observing wind direction and looking upwind to determine what's already happening there. A snowstorm in Indiana is soon going to be a snowstorm in Ohio.
For systems that have less obvious cause and effect, or that aren't visible and fine tuned by billions of observers, many models may in fact tell us nonsense.
If I remember correctly, weather was precisely an example of a system with butterfly effects, but the reason we can sort of predict it is that butterfly effects take some time to get going: you can predict it a day or two ahead pretty well, but there's a limit.
Yes, exactly this. In mathematics, the "time to get going" can be measured by the Lyapunov exponent.
It's also worth pointing out systems can have different levels of chaos at different scales. This is the usual "climate versus weather" thing: while you can't predict local weather much more than a week or two before butterfly effects ruin it, larger-scale variables like average surface temperature are more stable and can be forecast much longer.
If you're interested in the difference between climate vs weather, and what sorts of things we can expect to predict and why, I highly recommend Weather, Macroweather, and the Climate by Lovejoy.
They do, but they manifest differently. In weather it was to do with how far in the future you could predict, and that boundary has been slowly and painstakingly been moved forward. We also use simulations in engineering all over the shop, for rocket design, and for pharma research etc.
But the main benefit as I see it is not predictive. It's explanatory. It's also a better way to understand the moving parts that actually affect a system and then, if lucky, try short term predictions before the changes overwhelm us. Not saying they're the best, or only, choice, but that we're barely using them and we should do more!
A complex system contains a number of variables. The more variables of these you can observe, the further throughout time you can predict, within error bounds, how the system will evolve. One way of becoming better at predicting is to build better models, but usually building a good model is relatively easy if you have the variables and enough data in front of you. The most meaningful increase in predictability is in being able to obverse more and more variables. That's why hedge funds are paying ridiculous sums for private data sources. Creating more data streams and making them public is the best way to advance humanity's modeling of complex systems like the brain.
The internet and its data did more for artificial intelligence than any model did.
The butterfly effect is not a problem that results from large complicated systems. The original example only had three dynamical variables. We can even see it in one dynamical variable (in discrete time). Of course, it can occur in large complicated systems, but all that is needed is growing errors.
The simplest example is to take an infinite decimal, shift all the digits to the left by one, and drop anything to the left of the decimal. As an equation, this is: x(t+1) = 10*x(t) mod(1) . We can watch how an error, in this case to decimal representation of 5/7, grows.
x(0) = 0.714295714285714285... Error is 0.00001 (the 9 should be an 8)
x(1) = 0.142957142857142857... Error is 0.0001
x(2) = 0.429571428571428571... Error is 0.001
x(3) = 0.295714285714285714... Error is 0.01
x(4) = 0.957142857142857142... Error is 0.1
And then the error goes away because none of the subsequent digits were messed up (and this example isn't one-to-one). If the subsequent digits were messed up, then an initial accuracy of 10^-5 would only allow us to predict for t < 5.
In order to meaningfully observe such a complex system as the economy, you need access to many variables that are a part of this system. In our current world of walled gardens, quarterly earnings, and paywalls, getting the data necessary to modelling the system is harder than actually modeling it.
The best thing you can do to advance the modeling of the economy, or the human body, or the brain, or anything really, is to fight to increase the amount of high-quality data that's publicly available. If the data is on the internet, nerds will come and use it. Pushing companies towards real-time earnings would be a dramatic improvement for example. I believe that advocating for more data should be a priority for the rationalist movement.
I don't disagree at all here. I think data is crucial. However I also think we're at a point where even if we were to get some data, we don't really have a lattice to hang it all onto beyond smaller scale and tightly defined analytic models. In fact it's worse, we use the data we have as input into ml models with an assumption that the hidden layers will somehow intrinsically recreate the necessary framework. Even if that's true, without it being made explicit, it's unclear if we can gain a better understanding of the system.
Homocysteine, huh? Interesting, Sarah was talking about that just the other day on Twitter in the context of mortality rather than depression, and how it's possible that supplementing folate to reduce could be a good idea just in general (but how nobody's done a proper study of this): https://twitter.com/s_r_constantin/status/1356643626785378304
The Lorien Psychiatry website has a whole bunch of essays on various psychiatric things, and they've slowly been making there way over here. I guess Scott built up a pretty big buffer while setting up his new practice.
One possibly overly-pedantic query... Where it says, "I think the mind is at least this complicated," surely the mind can't be anywhere near as complicated as the global economy, given that billions of minds are themselves part of the economy?
I'm not sure that follows, since most of the complexity of each mind has negligible effect on the economy. The patterns of traffic flow through a city might be less complicated than the flow of electrons through the circuits of a single computer chip, even though each car contains many such devices.
Similarly, I wouldn't say the ideal gas law is more complex than multibody quantum mechanical calculations, even though one is an approximate sum of many iterations of the other.
But what’s analogous to the ideal gas law in this case is not the global economy, it’s a *theory* of the global economy, which you’re certainly right is far less complicated. The actual state of the gas itself, though, is wildly more complex than that the quantum state of any of its constituent atoms, since it is composed of those states. What (I thought) Scott was comparing was the global economy and the human mind as systems.
The traffic / electron analogy is a better one I think, but I still don’t think it’s right, because I don’t agree that the complexity of individual minds has a negligible effect on the economy. I think they have a huge, constitutive effect, and I suspect that much of the failure of economics as a predictive science comes down to its (perhaps unavoidable) analysis of people as simple agents.
Ok, fair, I agree that's strictly, literally true. It's also, I think, kind of the whole point. With the possible exception of whatever the True Theory of Fundamental Physics is, everything else we can discuss involves a series of abstractions that screen off the things we don't care about, because they don't meaningfully affect the system's macroscopic behavior. The more we have to screen off, the more often we should expect the model to fail to make a correct prediction.
I think your line of thinking implies a measure of complexity that just isn't very useful. If I magically removed a nanogram of marble from the top of the head of the statue of David, then in terms of physical forces it has become super-exponentially less complex, and yet it is extremely unlikely this will have a net measurable effect on future events. It might, but on average whatever tiny effects it has will probably not add up to anything noticeable at human or societal scale. Discussions of complex systems tend to focus on chaotic unpredictability, but in practice understanding attractors is a big part of how we can best intervene in and reason about their behavior.
That said, I do think there is a whole lot of room for economics to incorporate *more* aspects of human complexity in its models, and get useful insights out of it.
That’s well put. You’re right, while physically a system can’t be less complex than one of the sub-systems that comprise it, it can still be meaningful to ask whether its emergent
behaviour is.
Presumably for the sake of fairness we’d also need to “screen off” lower-level sub-systems of the mind at some point... e.g. is all the information flow, feedback etc. within each neuron allowed to count towards the overall mental complexity? The global economy might have grounds for complaint if it did...
My mind is hugely complex. It still always chooses "more money", "more leisure time", "more pleasurable experiences" and so forth.
Things can be complex without being impactful at the chosen level of discussion. A plate of mashed potato will have extremely complex surface geometry, but this complexity doesn't affect high-level questions like "is it healthy?", "will it poison me?", "does it fit my macros?" etc. One plate is like another.
Yes, I think I misinterpreted it as a statement about complexity in the physics/information theory sense rather than about predicative models.
For the record, though, it’s not so obvious that people’s psychology can be reduced anywhere near as drastically as mashed potato topology without impacting the accuracy of the model. After all, there’s a whole branch of economics that takes this problem seriously and tries to build models that don’t do it so much.
Sadly my mind does NOT reliably choose “more money”. I can’t quantify how much that’s affected the global economy, but it’s definitely disappointed my mother.
I think a cell is more complicated than a bone along the relevant predictability dimension, even though bones are made of billions of cells.
Or to think of it in a different way - I am much better at predicting my girlfriend than I am at predicting protein folding, even though my girlfriend contains billions of proteins and the exact way each one folds is necessary to keep her alive and generate her behavior.
Partly this is because I don't care about predicting my girlfriend on the molecular level. But I don't care about predicting the economy on the mind-by-mind level. Mostly I just want to know whether it will go up or down or whatever.
These are nice, but I think it's a mistake to call them ontologies. They are models which, as you point out, capture a few salient points, but not the whole shtick. I think "ontology" in information science is best used to describe the underlying framework of a humanly-created system, where it can, in theory, be complete and even offer guidelines for further development. But "ontology" is too highfalutin a term for an approximate model of a large and complicated system that we have encountered from the outside and continue to study, many of whose fundamental inner details are unknown and will be for a long time, maybe forever. Using the term "ontology" gives the models more credence than they deserve, since an ontology is supposed to embody the axiomatic basis of the system is associated with, and those are unknown.
Thank you for allowing me to carp on the use of the term. Except for that, I liked your models and the point you are making.
Agree. A nice definition in Wikipedia https://en.wikipedia.org/wiki/Ontology_(information_science) -- "In computer science and information science, an ontology encompasses a representation, formal naming and definition of the categories, properties and relations between the concepts, data and entities..." What Scott is discussing is not an ontology.
I'm not familiar with information science. I'm using it as it's used in philosophy, which Wikipedia describes as "the questions of how entities are grouped into basic categories and which of these entities exist on the most fundamental level."
I don't think anything in your "ontologies" refers to something that exists on "the most fundamental level"; that is, in the organisms and populations. I feel that they are abstract models which capture some aspects of the phenomena they apply to. And I'm all for models and theories that capture 90% of the behavior with 10% of the work that a more complete model would supply, so Iike the simple models. But to say that they capture fundamentals seems an exaggeration. They capture salient aspects. They are strictly phenomenological and not about deep underlying causes. You assume a depressed person; what would be more fundamental would be the ability to predict that some person will be depressed. It's good to have models like these, but an ontology is supposed to represent, as you say, fundamental understanding. And it is not even clear that deeper understanding will arise from elaborating these specific models forward.
The Ptolemaic theory was quite predictive, until new measurements and eventually new thinking led to a simpler description, a better model. Perhaps Ptolemy thought of his theory as an ontology; but it was replaced with a better theory, which in turn had to be modified at large scales when Relativity became known. It didn't turn out to be an ontology after all, but rather a model that was superceded. And when we are describing complicated systems that we encounter (as opposed to systems that we build from scratch), I am dubious that any description can be termed an ontology, because any of our models are in danger of being shown to be lacking in the face of new knowledge. So how could any of them ever have been an ontology?
In philosophy, ontology is a branch of metaphysics, not physics, or chemistry, or biology. In science, in contrast, we have models. But the term has been adopted in information science to refer to the basic assumptions, principles and axioms that a constructed system is built upon. But I think there are good reasons for scientists who build models to avoid the term ontology. It's too impressive and even misleading a term for what we do. All IMO, that is.
P.S. It's great to see that we now can post replies easily!
Why wouldn't the attractor states make the taxometrics show a taxa, instead of a trait? Shouldn't the attractor states create multiple modes at the attractors?
Back in the taxometrics post, you had an example of an obvious taxa, humans vs bunnies. But if you included all the species along a continuous traversal through the evolutionary tree from humans to bunnies (i.e. back from humans to our common ancestor, then forward to bunnies), you'd have a smooth curve from human to bunny (smooth i.e. continuous, but not monotonic). Unless I've misunderstood taxometrics, I would imagine that including all those species would cause the statistics to lose the distinction of humans and bunnies as separate taxa, and instead to just conclude they were various parts of a rather large and skewed distribution.
The dynamic system, in this case, would be the environmental selection pressures that evolved the common ancestor towards humans and towards rabbits. It has the attractor states of human, and rabbit, which we can now statistically differentiate into separate taxa. If depression really does have attractor states that we can think of as "has depression" and "doesn't have depression", shouldn't that show up in the taxometrics, just like humans and rabbits do?
I agree with the prediction that depressed vs. nondepressed states, within an individual, should be taxonic (at least in some kinds of depression). I don't know if anyone has ever studied this question.
I'm unclear what it means in this context to predict that "has depression" and "doesn't have depression" are taxonic within an individual. Don't we observe around us that in and out of depression or in and out of anxious states is pretty much always a matter of degree, as in mild, moderate, severe or with varying scores on assessments, from one day or week or month to the next within an individual? I think I must be misunderstanding.
This was a delight to read, and felt very natural to me. I've had similar thoughts and discussions about my own mental health, and my experiences with meditation. It also reminds me of everything I've read on SSC and elsewhere describing the mind in terms of a potential energy landscape.
Mostly, though, it reminded me of a recent discussion I had with a coworker about Zvi's lesswrong covid post where they (he?) used the phrase "pretend to pretend to try to try." Her reaction was, "I love that there's a forum out there with an audience that reads that sentence and nods along like it's a perfectly normal thing to say."
Thanks for creating a space where this kind of discussion is a perfectly normal thing to have.
if psychiatrists weren’t aware that some people have persistent depressive symptoms that occasionally get more severe until “double depression” came into the lexicon we are all fucked
Scott is slightly incorrect saying, “An enzyme in your body called MTHFR changes vitamin B9 into the active form l-methylfolate.” In reality, MTHFR changes the intermediate called methylenetetrahydrofolate (a form of folate three metabolic steps downstream from vitamin B9/folic acid) into l-methylfolate.
Why people are paying so much attention to MTHFR when any of the multiple folate cycle enzymes can go wrong? That is because of the so called genetic polymorphism 677C>T or, systematically, rs1801133, which in its homozygous form is present in 12% of Europeans. Homozygous rs1801133 (TT) results in a lower activity MTHFR enzyme (60% of normal activity).
Despite the hype surrounding companies like Genetic Genie, a very simple routine test any doctor (including Scott) can prescribe will tell you all you need to know about the state of your folate system. The test is called plasma total homocysteine, and it is very sensitive to the disruptions of folate metabolism. For example, a person with normal genotype has homocysteine level at 6-7 microM, but a person with rs1801133 (TT) has homocysteine at 10-11 microM. If you treat such a person with with 4 mg of folic acid per day their homocysteine goes down to 7 microM (see figure 3 in MTHFR 677C>T genotype is associated with folate and homocysteine concentrations in a large, population-based, double-blind trial of folic acid supplementation. https://pubmed.ncbi.nlm.nih.gov/21508090/ )
Since you seem to know a lot about folate, do you know anything about the supposed increase in some kinds of cancer risk (prostate cancer) with folate supplementation? I found one study that mentioned it, but it was mostly composed of people who took regular folate rather than l-methylfolate.
My homocysteine was 12.5 a few years ago on a blood test, so I wonder if I have the gene defect. Never explicitly tried folate supplementation until very recently.
It's unclear. There are arguments that high folate supplementation increases intestinal cancer while lower level of plasma folate increases some other cancers. I would not go overboard. Increasing the folate supplement dose will not lower your homocysteine much.
Appreciate the response. I'm generally quite cautious about what I supplement and ruthlessly purge things that don't work for me, so I'll be ok. I'm mainly taking it on a trial basis to see if it improves my mood. Scott had previously mentioned it here: https://slatestarcodex.com/2014/06/15/fish-now-by-prescription/
My homocysteine 2023-03-27 was 62.95 µmol/l. Next time I checked, 2023-12-08, it was 26.11 µmol/l. Also B12 was 177.00 pg/ml and folic acid at 5.90 ng/ml -- both on a second blood test; I didn't test it the first time. But I often had a feeling like half of my left palm was half-numb - I assume that was B12 deficiency. I don't have such symptoms anymore, so...
I had my genome sequenced, and Promethease reports I've got
I've been supplementing B vitamins and B12 specifically, so I assume homocysteine went down because of that. But I didn't take the mutation into account.
Question: why might my value be so high? Did I get confused with units somehow?
Also, would the following stack make sense, to work around the mutation + increase serotonin?
- L-5-MTHF-Ca 500ug
- 200mg (mix of Methylcobalamin and Adenosylcobalamin)
- Trimethylglycine (betaine) 1000mg
- 5-HTP 100mg + EGCG
- Fluoxetine (20mg)
Could it somehow disrupt synthesis of other catecholamines?
I think there's a difference between "trait rather than dynamical system" and "unchangeable". The first example that comes to mind is that height is a trait, but cutting my legs off (or just waiting for me to get old and decrepit) will change it. I admit I don't have a great way to think about this philosophically.
The earliest fossil evidence for Homo Sapiens is about 300,000 years ago, so you'd need to go back a few interglacials. And even that might not be enough - the control of fire is almost a million years old, so maybe Homo Erectus was still smart enough to need mental health care.
"Some mental disorders are pure traits without much dynamism – I think personality disorders, ADHD, and autism fall into this category." Our seven-year old son has been diagnosed with combined type ADHD, ODD and "mild ASD" (exact words the paediatrician used). From my perspective, there is heaps of dynamism in his symptoms. Right now he is in a good phase, the school tells us he is following the rules, doing the work, even having positive social interactions with other students (he is in a mainstream classroom, not special education). But he has also had his bad phases. Last year, there were weeks when the school called us almost every day to come pick him up early because he has been physically violent against the school staff, and the school was trying to encourage us to pull him out of mainstream and send him to special education instead. (His private Catholic school only has mainstream classes, to go to special education we'd have to enrol him in another school.) There have been periods when I haven't been able to work because I've had to remove him from the home to stop him from physically attacking his mother. But, none of that right now, we are hoping and praying the current good phase lasts, yet we've had good phases that didn't last before. Is that anecdotal evidence against the idea that ADHD and autism are "pure traits without much dynamism"?
This is a difficult question and I don't have a great answer to it, besides the "trait-like" vs. "unchangeable" distinction I made in the comment above.
Simon K, this comment is inspired by your noting the variation in your son's experience rather than directly responding to it specifically. I hope that's okay.
To my mind, stress is the biggest single source of dynamism in mental disorders whether we're talking about ADHD, anxiety/depression, PTSD, OCD, personality disorders, bipolar disorder, or psychosis.
Sometimes the stress is obvious from the outside (job loss, illness) and sometimes it's not (subtle changes in performance demands, unreported bullying) or is due to internal changes (developmentally normal hormone fluctuations, undiagnosed auto-immune disorder).
Stress isn't a free-standing event, but is what results from a new force being added (from within or outside the body) to the current ingredients of the individual. By current ingredients I mainly mean how well-resourced or not the person is in flexing with the added force. Well resourced can mean "good mental tools to handle curve balls without freaking out" or "really great support system to lean on" or "consistent vigorous daily exercise," among others (good night's sleep, eating well, etc).
Kids in particular are extra dynamic systems because they are growing all the time and being faced with and mastering new developmental challenges all the time. Later, we can say about adults, "oh she's going through a divorce, she's under a lot of stress." But for kids, it's so hard to say because something that was unbelievably stressful two weeks ago, they may have mastered today or something that they've been riding on mastery of for awhile may have just shifted out from under them as a result of new demands in their environment or changes in their internal chemistry. They can't generally explain this to us as it's happening so I think we miss a lot of the stressors.
Anyway, I don't see any mental disorders that are pure traits without dynamism (I'm a psychotherapist).
Completely foreign and exotic aliens which have no corresponding concepts for "War" and "Money", but can perform sentiment analysis to figure out likes/dislikes from facial expressions? They might not even have faces!
I'm just gonna add a little to the dynamical system and attractors perspective - sometimes a dynamical system has attractors which are not points in phase-space (ie, the multidimensional space of possible states of the system) but rather closed lines (limit cycles). This would mean that the system travels along those lines in a periodic way, and any other orbit in phase space eventually reaches these closed lines. A physical example of this is a mass on a spring with friction that sometimes slows it down and other times speeds it up (van der Pol oscillator).
Do you think this could be a good way to understand disorders which have periodic behavior? Say, is bipolar manic and depressive episodes better understood as "external shocks move the system between stable points", or "the internal dynamics of the system moves it along a periodic trajectory between two extremes"?
I have definitely seen bipolar disorders in both categories. Some bipolar patients have episodes whenever they travel (because the jet lag messes up their circadian rhythm) - this is a clear external shock. Other bipolar patients have cycles you can set your watch to. Sometimes almost literally - some ultradian cycling patients will predictably have a mood crash at a certain time of day.
Tangent - "But also, the refinery workers might strike in sympathy with the truckers" is a problem that a robust system of social castes solves! If refining and transportation are done by two separate tribes who don't interact socially, don't intermarry, and don't try to horn in on each other's economic domains, then the odds of one tribe paying a cost to show sympathy to the other at the same time that that other tribe is purposefully making things difficult for the first one are essentially nil.
Leaving aside for a second how weird a tangent this is, I'm not even sure it's true. Black people and white people are pretty socially segregated, but white people have been protesting in solidarity with black people a lot recently, and I doubt it's just that the white protesters personally know a lot of blacks.
I'd wager it's more the opposite and they don't know very many blacks. Baizuo and all that. There's probably few whiter things you could do in the U.S. than social justice activism.
I wonder if a dice roll is a dynamic system with 6 attractor states? It seems like there is some graph that maps a dice leaving the player's hand to its resting state on one face, but the graph is so large that we can't understand it as a system and so think of it as random.
I don't mean to be cynical. It's more important to understand the economy and our mental health than it is to understand which way a dice will fall. Maybe it's even more important to understand just how hard that endeavour is and remind ourselves that anyone who is too confident in their interventions is (at best) a benevolent alien.
In theory you could perfectly predict a dice roll, given the exact speed and position of the dice leaving the hand, the topography and materials of the rolling surface, the forecast for local micro air currents for the next few seconds, etc.
In other news, my new casino will replace dice with a list of people with bipolar disorder. You just bet then call one and ask for their current state.
Yes, a die about to land is more or less a system with 6 attractor states. (You could nitpick about the details and how the angles matter and how the 6 states are more like 6 attractor subsets, but more or less this is right.) And it isn't even a complicated and mysterious system at all: it's extremely well-understood, as far as systems go. The reason we treat it as hard to predict as random is that it has a property, that if you measure the starting state with an even very slight error (and you always will), the error at some points can start growing very fast and render your long-term prediction wrong. (Do the systems behind you mental processes also have this hard to predict property? We don't know for sure, but it seems very plausible.)
I imagine it's harder to do this for dice, but people have successfully developed computers that can predict the outcome of a roulette wheel, by measuring the speed and position of the ball. (One group used a computer hidden in their shoe, tapping their foot when the ball came around to measure the speed.)
It's not perfect, obviously, but it's good enough to beat the house, so casinos now keep an eye out for these shenanigans.
“large stimulus packages can sometimes shift the system from recession to normal growth”
To economists out there: is there an example of this actually happening? I know the Great Depression is the era most associated with Keynesian economics, but my understanding is that the increase in taxes (starting under Hoover) made policy not all that Keynesian, and it lasted a long time (even having another recession prior to full recovery).
I'm attaching a celebrated paper which talks about the neuroscience of memory retrieval. It also takes the dynamical systems approach in order to explain how we retrieve memories in our brain. The strength of the paper lies in the fact that it does not make strong assumptions about the brain, and most parameters can be "random". https://www.pnas.org/content/79/8/2554
As an English computer scientist, I was confused both by your use of "gas" to mean "petrol" (I assumed by "gas lines" you meant "pipes for natural gas", not "petrol queues") and your use of "i" and "j" for boolean variables rather than indices.
Just wanted to note that Scott's drawings may be slightly misleading. Critical slowing down is a feature that appears only when the system is going from one macro state to the other (the attractors in Scott's language, although the phenomena occurs in non dynamical systems too), but not in the vicinity of the basin of attraction, so no slowing down near an attractor, even though the drawing might suggest that due to the ground being level there.
It's also funny that the link toward an example of critical slowing down in the global economy is an article claiming there is no evidence for it, which seems to conform to a brief internet search on the terms.
On the very misleading side is calling something fractally complicated due to having causes that have causes,etc. The defining feature of a fractal is self-similarity, so that you do not have to know the underlying cause, everything you need to know is at whichever scale you're looking at and you can understand this scale while completely oblivious to finer details (which are just more of the same).
I'm happy that Scott is back and double happy he is posting very regularly, but throughout the first parts I felt being constantly threatened with math and links to peer-review until I stopped to review some things.
As an expert in dynamical systems, I like this way of looking at the world. Most things are dynamical, and have many variables. A dynamical model of the brain would probably have to contain as many variables as there are neurons. That being said, a surprisingly large number of complicated things behave as if they only have a low number of variables. For example, the transition to turbulence can (usually) be well understood using ideas of chaos theory with a few variables. These sorts of reductions require some sort of continuity or dissipation, which are lacking in the brain or the economy.
I want to correct one error: the opening example has only one dynamical variable. The system can be written in one line, using the sign function (which is +1 if the input is positive and -1 if the input is negative - and for this example, we also want sgn(0) = +1).
h(t+1) = h(t) + sgn( h(t) - 75 )
If it did have more than one dynamical variable, you wouldn't have been able to understand what it does using a one dimensional picture. The picture would be more accurate as a tent, with straight lines going up to a point above 75.
On a tangential note I suffer from atypical depression, though it's likely that my attractors are shallow and I'm moving among those faster than classic MDD (I suspect this being linked to my Delayed Sleep Phase Syndrome messing up the circadian rhytm). My diagnosis could also be incorrect since it's more similar to cyclothymia
This is an interesting and illuminative post. I wonder though about the overly exclusive physiological perspective. I agree the physiological complexity is staggering but there seem to be 2 additional areas left out though you do refer to them obliquely.
The feedback loops created by the external environment and the recursive element of the mind whether conscious or unconscious. Your distinction that the difference between sadness and depression might be that depression is a "stable" attractor state is one I'm not sure I agree with. It could be that there is some variable susceptibility but could it not be the feedback between all 3 systems that deepens the state so that it's difficult to emerge from it without help. If that was true it would perhaps explain why there is no substantive difference between sadness and depression at the outset. I am extending the inference of course that duration is the defining difference, as suggested by taxometrics, and that it's the complex dynamic feedback loops that deepens and creates your attractor state?
I don't think it's typical, but I agree it's a good idea.
Personally I share this strategy but have trouble implementing it. e.g. last spring I noticed myself maybe getting depressed again and knew that I should take some action to stave off falling into further depression ASAP, since I knew from past experience that when I am more depressed than that it's much harder to take such actions; however, I failed to do that effectively and did end up getting more depressed.
Is it also possible for the attractor states to evolve over time so they change in their shape and number, or are they genetically predetermined?
IIUC, attractor states "emerge" out of other characteristics, like to what degree being tired makes you leave the house less. If something changes the degree to which being tired makes you work less (for example, you get a dog and you have to walk it regardless of how tired you are), then the attractor state will also change.
Thanks. Using this metaphor of balls and basins, what do we mean when somebody’s genetically prone to depression? Are the genetic factors just a few more variables or do we mean they are completely different dynamical systems (say with different update equations)? Your article is making feel like depression can only be managed but not cured :( Is that too hopeless?
>What do we mean when somebody’s genetically prone to depression?
"Genetically prone" can be thought of as having a bigger and/or deeper basin around depression. Which amounts to saying, there are more positions of the ball that roll into the "depression" attractor, and it's harder to get the ball out once it goes in.
> Are the genetic factors just a few more variables or do we mean they are completely different dynamical systems?
There's a bit of gray area between "more variables" and "different equations": you can always squeeze multiple models into one equation by just setting variables to 0 when you don't need them. I imagine that while most of the "mood equation" is shared by most people, there are some pieces that may be individual-specific, especially as you consider neuro-atypical people.
Nice article.
You've done it again Scott. Bookmarked.
Just wanted to mention that this article has not appeared in my RSS feed as of this comment. All the other articles have. Maybe it's just me.
It appeared both in my RSS feed and email.
Are you using https://astralcodexten.substack.com/feed ? It appears for me.
This puts me in mind of a 1991 essay, in which the philosopher David Stove bemoans our lack of an ontology of the disorders of intellectual philosophical thought. I would be very interested in how you would put together such an ontology.
https://web.maths.unsw.edu.au/~jim/wrongthoughts.html
So Alice lost her job and her insurance and she broke up with Bob which now keeps him inside all day and away from making big plays in commodities. I know it's a vast oversimplification, but the aliens should just materialize universal healthcare and give Bob a new girlfriend. The gallon of prevention that's worth a barrel of cure.
Is Bob's new girlfriend a real person, with the own hopes and dreams and capacity to desire? If so, would it be effective to "give" Bob a new girlfriend? Putting even the ethics aside, what if she broke up with him because the aliens guessed wrong about their compatibility?
Similarly, but far bigger and more complicated, "giving" everyone health care involves millions of choices and interactions per year. An alien species cannot just "give" everyone health care, because health care is provided by a series of people (doctors, nurses, etc.).
I'm not saying that either situation isn't fixable, but that mandating that the obvious immediate fix go into place doesn't mean it will actually work. Bob needs to find his own girlfriend.
And Alice needs to find a new job.
In the hypothetical, Alice gets a job as soon as her health improves. (We ignore Medicaid for simplicity, but that leaves room for alien health care in the hypothetical).
But hypothetically we've determined that isn't the only outcome possible.
"But suppose Alice gets a serious pneumonia which knocks her health down to 70. Now she can’t work, she loses insurance, her health starts going down 1 each day, and she eventually goes down to 0. This is another stable state; as long as the system keeps evolving according to the rules, it will never change."
She could get to a point where health is not 0, but health also is not 75 (the point hypothetically needed to maintain job and health insurance).
Since we're talking about aliens with a materialization ray sophisticated enough to produce millions of BBL of West Texas Intermediate on the White House lawn, let's get Alice some healthcare and get Bob a nice real person (the materialization ray can produce real oil...I think they got this) so he can start shorting the market again and WTI goes down.
Free healthcare for Alice and a magic girlfriend for Bob might not be perfect solutions sure, but they might just be enough to get both of them out of their rut and rolling towards the axis of stability.
Joe feels sad sometimes so he goes to his room and refuses to talk to his friends. After a few days he feels fine and is ready to deal with people again, even if they are kind of assholes.
Be like Joe.
That is an absolutely incredible metabolism diagram even though it's incomplete. Imagine having one for the brain.
"SAM-e and tetrahydrobiopterin are cofactors for the enzymes tryptophan hydroxylase and tyrosine hydroxylase, which change the amino acids tryptophan and tyrosine into serotonin and dopamine, respectively. And serotonin and dopamine are known to be heavily involved in mood and energy level."
So is this why I sometimes get intense cravings for cheese? As in "I really need to eat some cheddar NOW"? Tyrosine in cheese -> dopamine?
Now you've got me wondering if this is why pizza is one of the only things that lifts my mood when I'm depressed. Nothing has ever tasted better than dripping, greasy cheese when I'm depressed. Looking at foods containing tyrosine too, cheese is one of the only foods I eat that contains much...
The carbs from the crust might be helping, too. It's not as though just cheese makes the difference, it's pizza.
If Scott is right (and Scott is nearly always right), then it's just good old "fat plus carbs are tasty, render you drugged-out in bliss after consuming a ton of them, and if it's unhealthy your body will love it" rather than anything as redeeming as "I'm low on dopamine and just recharging" 😀
I once met someone who claimed that cheese was addictive, though they attributed it to dairy farmers adding morphine. Maybe there was a kernel of truth, but for the wrong reason.
The "help my toddler is a picky eater" internet genre definitely contains claims that dairy (and gluten) are effectively addictive. For example: https://healthhomeandhappiness.com/why-does-my-child-seem-addicted-to-dairy-andor-wheat.html I haven't seen the claim of added morphine, just that casein (and gluten) are sufficiently opioid-like to be addictive in their own right.
Now that I think of it, the morphine-cheese claim was probably a confused telephone-game version of the fact that casein does metabolize into an opioid: https://en.wikipedia.org/wiki/Casomorphin
I would be surprised if it was anything that simple. Almost any elegant theory with lots of good real life correlates in biology usually turns out to be wrong.
So basically I'm just addicted to fermented dairy products? Ah well, such is life!
You probably just have French or Dutch genes :P
…that originated from Swiss mercenaries.
The Dutch used many Swiss mercenaries during their Golden Age, so perhaps there are some Swiss genes in the Dutch...
In writing about complex adaptive systems (CAS) we fall into a couple attractor states - either they're still tractable because of the inherent predictability of *some* macro patterns, or they're so complicated because of the number of moving parts that you're shooting in the dark. And yet, despite the billions of variables that affect each part of the economy, we do tend to repeat certain patterns. Even if inadequate that still has some explanatory power.
My contention is that to get to a dynamical understanding, simulations are criminally underused as explanatory vehicles, especially in business/ economic settings (medicine too, though I don't know enough here). We'd need them to help us figure out the right level of abstraction to understand these attractors and the overall shape. (I explored this recently again (https://www.strangeloopcanon.com/p/simulating-understanding). It remains the biggest annoyance I've had against economics since grad school. I want to see the "biochemical pathways" flow diagram equivalent for money, so we can run scenarios and hone assumptions. They'll help us get a better understanding of the strange attractors within the system, while analytical/ predictive models tend to focus more on getting to an answer, which fails in cases of CAS. (This might be a false dichotomy.)
The mind seems a far more complex system than the economy in this instance, and I wonder how you'd even go about the problem of trying to create/ simulate something to get an understanding here.
Don't simulations of large complicated systems have a butterfly effect type problem where if you get any of the parameters even slightly wrong, or miss a variable you shouldn't have missed, they tell you nonsense?
(now I'm realizing I must be at least sort of wrong about this, because we simulate the weather okay, but I don't know how much work that took or how replicable it is)
Weather may be a special case, because it's highly visible and very important to every day life. Also what rkris said about predictability due to macro patterns. In particular the macro pattern of observing wind direction and looking upwind to determine what's already happening there. A snowstorm in Indiana is soon going to be a snowstorm in Ohio.
For systems that have less obvious cause and effect, or that aren't visible and fine tuned by billions of observers, many models may in fact tell us nonsense.
Not to mention, it's a recurring joke about how often the weatherman is wrong.
If I remember correctly, weather was precisely an example of a system with butterfly effects, but the reason we can sort of predict it is that butterfly effects take some time to get going: you can predict it a day or two ahead pretty well, but there's a limit.
Yes, exactly this. In mathematics, the "time to get going" can be measured by the Lyapunov exponent.
It's also worth pointing out systems can have different levels of chaos at different scales. This is the usual "climate versus weather" thing: while you can't predict local weather much more than a week or two before butterfly effects ruin it, larger-scale variables like average surface temperature are more stable and can be forecast much longer.
If you're interested in the difference between climate vs weather, and what sorts of things we can expect to predict and why, I highly recommend Weather, Macroweather, and the Climate by Lovejoy.
They do, but they manifest differently. In weather it was to do with how far in the future you could predict, and that boundary has been slowly and painstakingly been moved forward. We also use simulations in engineering all over the shop, for rocket design, and for pharma research etc.
But the main benefit as I see it is not predictive. It's explanatory. It's also a better way to understand the moving parts that actually affect a system and then, if lucky, try short term predictions before the changes overwhelm us. Not saying they're the best, or only, choice, but that we're barely using them and we should do more!
A complex system contains a number of variables. The more variables of these you can observe, the further throughout time you can predict, within error bounds, how the system will evolve. One way of becoming better at predicting is to build better models, but usually building a good model is relatively easy if you have the variables and enough data in front of you. The most meaningful increase in predictability is in being able to obverse more and more variables. That's why hedge funds are paying ridiculous sums for private data sources. Creating more data streams and making them public is the best way to advance humanity's modeling of complex systems like the brain.
The internet and its data did more for artificial intelligence than any model did.
The butterfly effect is not a problem that results from large complicated systems. The original example only had three dynamical variables. We can even see it in one dynamical variable (in discrete time). Of course, it can occur in large complicated systems, but all that is needed is growing errors.
The simplest example is to take an infinite decimal, shift all the digits to the left by one, and drop anything to the left of the decimal. As an equation, this is: x(t+1) = 10*x(t) mod(1) . We can watch how an error, in this case to decimal representation of 5/7, grows.
x(0) = 0.714295714285714285... Error is 0.00001 (the 9 should be an 8)
x(1) = 0.142957142857142857... Error is 0.0001
x(2) = 0.429571428571428571... Error is 0.001
x(3) = 0.295714285714285714... Error is 0.01
x(4) = 0.957142857142857142... Error is 0.1
And then the error goes away because none of the subsequent digits were messed up (and this example isn't one-to-one). If the subsequent digits were messed up, then an initial accuracy of 10^-5 would only allow us to predict for t < 5.
In order to meaningfully observe such a complex system as the economy, you need access to many variables that are a part of this system. In our current world of walled gardens, quarterly earnings, and paywalls, getting the data necessary to modelling the system is harder than actually modeling it.
The best thing you can do to advance the modeling of the economy, or the human body, or the brain, or anything really, is to fight to increase the amount of high-quality data that's publicly available. If the data is on the internet, nerds will come and use it. Pushing companies towards real-time earnings would be a dramatic improvement for example. I believe that advocating for more data should be a priority for the rationalist movement.
I don't disagree at all here. I think data is crucial. However I also think we're at a point where even if we were to get some data, we don't really have a lattice to hang it all onto beyond smaller scale and tightly defined analytic models. In fact it's worse, we use the data we have as input into ml models with an assumption that the hidden layers will somehow intrinsically recreate the necessary framework. Even if that's true, without it being made explicit, it's unclear if we can gain a better understanding of the system.
Homocysteine, huh? Interesting, Sarah was talking about that just the other day on Twitter in the context of mortality rather than depression, and how it's possible that supplementing folate to reduce could be a good idea just in general (but how nobody's done a proper study of this): https://twitter.com/s_r_constantin/status/1356643626785378304
I'm curious, did you write most of these articles in the last couple of months, or the last couple of weeks?
A mix, but this one was months ago.
The Lorien Psychiatry website has a whole bunch of essays on various psychiatric things, and they've slowly been making there way over here. I guess Scott built up a pretty big buffer while setting up his new practice.
How do I go about getting a prescription for Zxyxon? Also, you misspelled Vraylar®.
It should be called V®@yl@®.
This is a fascinating post.
One possibly overly-pedantic query... Where it says, "I think the mind is at least this complicated," surely the mind can't be anywhere near as complicated as the global economy, given that billions of minds are themselves part of the economy?
I'm not sure that follows, since most of the complexity of each mind has negligible effect on the economy. The patterns of traffic flow through a city might be less complicated than the flow of electrons through the circuits of a single computer chip, even though each car contains many such devices.
Similarly, I wouldn't say the ideal gas law is more complex than multibody quantum mechanical calculations, even though one is an approximate sum of many iterations of the other.
But what’s analogous to the ideal gas law in this case is not the global economy, it’s a *theory* of the global economy, which you’re certainly right is far less complicated. The actual state of the gas itself, though, is wildly more complex than that the quantum state of any of its constituent atoms, since it is composed of those states. What (I thought) Scott was comparing was the global economy and the human mind as systems.
The traffic / electron analogy is a better one I think, but I still don’t think it’s right, because I don’t agree that the complexity of individual minds has a negligible effect on the economy. I think they have a huge, constitutive effect, and I suspect that much of the failure of economics as a predictive science comes down to its (perhaps unavoidable) analysis of people as simple agents.
Ok, fair, I agree that's strictly, literally true. It's also, I think, kind of the whole point. With the possible exception of whatever the True Theory of Fundamental Physics is, everything else we can discuss involves a series of abstractions that screen off the things we don't care about, because they don't meaningfully affect the system's macroscopic behavior. The more we have to screen off, the more often we should expect the model to fail to make a correct prediction.
I think your line of thinking implies a measure of complexity that just isn't very useful. If I magically removed a nanogram of marble from the top of the head of the statue of David, then in terms of physical forces it has become super-exponentially less complex, and yet it is extremely unlikely this will have a net measurable effect on future events. It might, but on average whatever tiny effects it has will probably not add up to anything noticeable at human or societal scale. Discussions of complex systems tend to focus on chaotic unpredictability, but in practice understanding attractors is a big part of how we can best intervene in and reason about their behavior.
That said, I do think there is a whole lot of room for economics to incorporate *more* aspects of human complexity in its models, and get useful insights out of it.
That’s well put. You’re right, while physically a system can’t be less complex than one of the sub-systems that comprise it, it can still be meaningful to ask whether its emergent
behaviour is.
Presumably for the sake of fairness we’d also need to “screen off” lower-level sub-systems of the mind at some point... e.g. is all the information flow, feedback etc. within each neuron allowed to count towards the overall mental complexity? The global economy might have grounds for complaint if it did...
My mind is hugely complex. It still always chooses "more money", "more leisure time", "more pleasurable experiences" and so forth.
Things can be complex without being impactful at the chosen level of discussion. A plate of mashed potato will have extremely complex surface geometry, but this complexity doesn't affect high-level questions like "is it healthy?", "will it poison me?", "does it fit my macros?" etc. One plate is like another.
Yes, I think I misinterpreted it as a statement about complexity in the physics/information theory sense rather than about predicative models.
For the record, though, it’s not so obvious that people’s psychology can be reduced anywhere near as drastically as mashed potato topology without impacting the accuracy of the model. After all, there’s a whole branch of economics that takes this problem seriously and tries to build models that don’t do it so much.
Sadly my mind does NOT reliably choose “more money”. I can’t quantify how much that’s affected the global economy, but it’s definitely disappointed my mother.
I think a cell is more complicated than a bone along the relevant predictability dimension, even though bones are made of billions of cells.
Or to think of it in a different way - I am much better at predicting my girlfriend than I am at predicting protein folding, even though my girlfriend contains billions of proteins and the exact way each one folds is necessary to keep her alive and generate her behavior.
Partly this is because I don't care about predicting my girlfriend on the molecular level. But I don't care about predicting the economy on the mind-by-mind level. Mostly I just want to know whether it will go up or down or whatever.
These are nice, but I think it's a mistake to call them ontologies. They are models which, as you point out, capture a few salient points, but not the whole shtick. I think "ontology" in information science is best used to describe the underlying framework of a humanly-created system, where it can, in theory, be complete and even offer guidelines for further development. But "ontology" is too highfalutin a term for an approximate model of a large and complicated system that we have encountered from the outside and continue to study, many of whose fundamental inner details are unknown and will be for a long time, maybe forever. Using the term "ontology" gives the models more credence than they deserve, since an ontology is supposed to embody the axiomatic basis of the system is associated with, and those are unknown.
Thank you for allowing me to carp on the use of the term. Except for that, I liked your models and the point you are making.
Agree. A nice definition in Wikipedia https://en.wikipedia.org/wiki/Ontology_(information_science) -- "In computer science and information science, an ontology encompasses a representation, formal naming and definition of the categories, properties and relations between the concepts, data and entities..." What Scott is discussing is not an ontology.
I'm not familiar with information science. I'm using it as it's used in philosophy, which Wikipedia describes as "the questions of how entities are grouped into basic categories and which of these entities exist on the most fundamental level."
I don't think anything in your "ontologies" refers to something that exists on "the most fundamental level"; that is, in the organisms and populations. I feel that they are abstract models which capture some aspects of the phenomena they apply to. And I'm all for models and theories that capture 90% of the behavior with 10% of the work that a more complete model would supply, so Iike the simple models. But to say that they capture fundamentals seems an exaggeration. They capture salient aspects. They are strictly phenomenological and not about deep underlying causes. You assume a depressed person; what would be more fundamental would be the ability to predict that some person will be depressed. It's good to have models like these, but an ontology is supposed to represent, as you say, fundamental understanding. And it is not even clear that deeper understanding will arise from elaborating these specific models forward.
The Ptolemaic theory was quite predictive, until new measurements and eventually new thinking led to a simpler description, a better model. Perhaps Ptolemy thought of his theory as an ontology; but it was replaced with a better theory, which in turn had to be modified at large scales when Relativity became known. It didn't turn out to be an ontology after all, but rather a model that was superceded. And when we are describing complicated systems that we encounter (as opposed to systems that we build from scratch), I am dubious that any description can be termed an ontology, because any of our models are in danger of being shown to be lacking in the face of new knowledge. So how could any of them ever have been an ontology?
In philosophy, ontology is a branch of metaphysics, not physics, or chemistry, or biology. In science, in contrast, we have models. But the term has been adopted in information science to refer to the basic assumptions, principles and axioms that a constructed system is built upon. But I think there are good reasons for scientists who build models to avoid the term ontology. It's too impressive and even misleading a term for what we do. All IMO, that is.
P.S. It's great to see that we now can post replies easily!
I think questions like "are mental disorders real things that exist, or are they made up of smaller parts" are ontological questions.
I agree, but I don't think your models address or embody a solution to that ontological question.
I think the taxometrics one was very directly about whether mental disorders "exist" as "things", and this one is adding some epicycles to that.
Why wouldn't the attractor states make the taxometrics show a taxa, instead of a trait? Shouldn't the attractor states create multiple modes at the attractors?
Back in the taxometrics post, you had an example of an obvious taxa, humans vs bunnies. But if you included all the species along a continuous traversal through the evolutionary tree from humans to bunnies (i.e. back from humans to our common ancestor, then forward to bunnies), you'd have a smooth curve from human to bunny (smooth i.e. continuous, but not monotonic). Unless I've misunderstood taxometrics, I would imagine that including all those species would cause the statistics to lose the distinction of humans and bunnies as separate taxa, and instead to just conclude they were various parts of a rather large and skewed distribution.
The dynamic system, in this case, would be the environmental selection pressures that evolved the common ancestor towards humans and towards rabbits. It has the attractor states of human, and rabbit, which we can now statistically differentiate into separate taxa. If depression really does have attractor states that we can think of as "has depression" and "doesn't have depression", shouldn't that show up in the taxometrics, just like humans and rabbits do?
I agree with the prediction that depressed vs. nondepressed states, within an individual, should be taxonic (at least in some kinds of depression). I don't know if anyone has ever studied this question.
I'm unclear what it means in this context to predict that "has depression" and "doesn't have depression" are taxonic within an individual. Don't we observe around us that in and out of depression or in and out of anxious states is pretty much always a matter of degree, as in mild, moderate, severe or with varying scores on assessments, from one day or week or month to the next within an individual? I think I must be misunderstanding.
For anyone who's interested, I wrote a short post about how Borsboom's ideas might apply to chronic pain, which can very complex. https://www.bettermovement.org/blog/2017/pain-networks
This was a delight to read, and felt very natural to me. I've had similar thoughts and discussions about my own mental health, and my experiences with meditation. It also reminds me of everything I've read on SSC and elsewhere describing the mind in terms of a potential energy landscape.
Mostly, though, it reminded me of a recent discussion I had with a coworker about Zvi's lesswrong covid post where they (he?) used the phrase "pretend to pretend to try to try." Her reaction was, "I love that there's a forum out there with an audience that reads that sentence and nods along like it's a perfectly normal thing to say."
Thanks for creating a space where this kind of discussion is a perfectly normal thing to have.
if psychiatrists weren’t aware that some people have persistent depressive symptoms that occasionally get more severe until “double depression” came into the lexicon we are all fucked
Psychiatrists: "Depression can some times flare up into a more severe form officially known as Real Bad Depression."
Scott is slightly incorrect saying, “An enzyme in your body called MTHFR changes vitamin B9 into the active form l-methylfolate.” In reality, MTHFR changes the intermediate called methylenetetrahydrofolate (a form of folate three metabolic steps downstream from vitamin B9/folic acid) into l-methylfolate.
Why people are paying so much attention to MTHFR when any of the multiple folate cycle enzymes can go wrong? That is because of the so called genetic polymorphism 677C>T or, systematically, rs1801133, which in its homozygous form is present in 12% of Europeans. Homozygous rs1801133 (TT) results in a lower activity MTHFR enzyme (60% of normal activity).
Despite the hype surrounding companies like Genetic Genie, a very simple routine test any doctor (including Scott) can prescribe will tell you all you need to know about the state of your folate system. The test is called plasma total homocysteine, and it is very sensitive to the disruptions of folate metabolism. For example, a person with normal genotype has homocysteine level at 6-7 microM, but a person with rs1801133 (TT) has homocysteine at 10-11 microM. If you treat such a person with with 4 mg of folic acid per day their homocysteine goes down to 7 microM (see figure 3 in MTHFR 677C>T genotype is associated with folate and homocysteine concentrations in a large, population-based, double-blind trial of folic acid supplementation. https://pubmed.ncbi.nlm.nih.gov/21508090/ )
Since you seem to know a lot about folate, do you know anything about the supposed increase in some kinds of cancer risk (prostate cancer) with folate supplementation? I found one study that mentioned it, but it was mostly composed of people who took regular folate rather than l-methylfolate.
My homocysteine was 12.5 a few years ago on a blood test, so I wonder if I have the gene defect. Never explicitly tried folate supplementation until very recently.
It's unclear. There are arguments that high folate supplementation increases intestinal cancer while lower level of plasma folate increases some other cancers. I would not go overboard. Increasing the folate supplement dose will not lower your homocysteine much.
Appreciate the response. I'm generally quite cautious about what I supplement and ruthlessly purge things that don't work for me, so I'll be ok. I'm mainly taking it on a trial basis to see if it improves my mood. Scott had previously mentioned it here: https://slatestarcodex.com/2014/06/15/fish-now-by-prescription/
Correcting a typo - it should read: Increasing the folate supplement dose beyond 5 mg per day will not lower your homocysteine much.
Thank you, I have slightly edited the post.
My homocysteine 2023-03-27 was 62.95 µmol/l. Next time I checked, 2023-12-08, it was 26.11 µmol/l. Also B12 was 177.00 pg/ml and folic acid at 5.90 ng/ml -- both on a second blood test; I didn't test it the first time. But I often had a feeling like half of my left palm was half-numb - I assume that was B12 deficiency. I don't have such symptoms anymore, so...
I had my genome sequenced, and Promethease reports I've got
"gs192" - "677CT/1298CA 19.8% - 2 heterozygous mutations (compound heterozygous)"
which is composed of
rs1801133(C;T) - copy of C677T allele of MTHFR = 65% efficiency in processing folic acid
rs1801131(A;C) - Possibly impaired folate metabolism
----
I've been supplementing B vitamins and B12 specifically, so I assume homocysteine went down because of that. But I didn't take the mutation into account.
Question: why might my value be so high? Did I get confused with units somehow?
Also, would the following stack make sense, to work around the mutation + increase serotonin?
- L-5-MTHF-Ca 500ug
- 200mg (mix of Methylcobalamin and Adenosylcobalamin)
- Trimethylglycine (betaine) 1000mg
- 5-HTP 100mg + EGCG
- Fluoxetine (20mg)
Could it somehow disrupt synthesis of other catecholamines?
"Some mental disorders are pure traits without much dynamism – I think personality disorders, ADHD, and autism fall into this category."
What about the famous anecdotes of fevers relieving autism symptoms?
I think there's a difference between "trait rather than dynamical system" and "unchangeable". The first example that comes to mind is that height is a trait, but cutting my legs off (or just waiting for me to get old and decrepit) will change it. I admit I don't have a great way to think about this philosophically.
Imagine that these aliens had arrived at Earth during an earlier interglacial period. It would have been so much easier!
How many millions of years do we have to go back to make mental health that easy? Or would the previous interglacial be far enough?
The earliest fossil evidence for Homo Sapiens is about 300,000 years ago, so you'd need to go back a few interglacials. And even that might not be enough - the control of fire is almost a million years old, so maybe Homo Erectus was still smart enough to need mental health care.
Cats are often pretty mad. I don't think the aliens' life would really have been that much easier.
"Some mental disorders are pure traits without much dynamism – I think personality disorders, ADHD, and autism fall into this category." Our seven-year old son has been diagnosed with combined type ADHD, ODD and "mild ASD" (exact words the paediatrician used). From my perspective, there is heaps of dynamism in his symptoms. Right now he is in a good phase, the school tells us he is following the rules, doing the work, even having positive social interactions with other students (he is in a mainstream classroom, not special education). But he has also had his bad phases. Last year, there were weeks when the school called us almost every day to come pick him up early because he has been physically violent against the school staff, and the school was trying to encourage us to pull him out of mainstream and send him to special education instead. (His private Catholic school only has mainstream classes, to go to special education we'd have to enrol him in another school.) There have been periods when I haven't been able to work because I've had to remove him from the home to stop him from physically attacking his mother. But, none of that right now, we are hoping and praying the current good phase lasts, yet we've had good phases that didn't last before. Is that anecdotal evidence against the idea that ADHD and autism are "pure traits without much dynamism"?
This is a difficult question and I don't have a great answer to it, besides the "trait-like" vs. "unchangeable" distinction I made in the comment above.
Simon K, this comment is inspired by your noting the variation in your son's experience rather than directly responding to it specifically. I hope that's okay.
To my mind, stress is the biggest single source of dynamism in mental disorders whether we're talking about ADHD, anxiety/depression, PTSD, OCD, personality disorders, bipolar disorder, or psychosis.
Sometimes the stress is obvious from the outside (job loss, illness) and sometimes it's not (subtle changes in performance demands, unreported bullying) or is due to internal changes (developmentally normal hormone fluctuations, undiagnosed auto-immune disorder).
Stress isn't a free-standing event, but is what results from a new force being added (from within or outside the body) to the current ingredients of the individual. By current ingredients I mainly mean how well-resourced or not the person is in flexing with the added force. Well resourced can mean "good mental tools to handle curve balls without freaking out" or "really great support system to lean on" or "consistent vigorous daily exercise," among others (good night's sleep, eating well, etc).
Kids in particular are extra dynamic systems because they are growing all the time and being faced with and mastering new developmental challenges all the time. Later, we can say about adults, "oh she's going through a divorce, she's under a lot of stress." But for kids, it's so hard to say because something that was unbelievably stressful two weeks ago, they may have mastered today or something that they've been riding on mastery of for awhile may have just shifted out from under them as a result of new demands in their environment or changes in their internal chemistry. They can't generally explain this to us as it's happening so I think we miss a lot of the stressors.
Anyway, I don't see any mental disorders that are pure traits without dynamism (I'm a psychotherapist).
I like the humility of this post. The brain, society, etc. are _grand_ things.
"Or a recession caused by deregulation of banks causing them to offer too many subprime loans causing all of them to go belly-up at once"
Were any of the pro-regulation people saying the banks needed to be regulated so they gave <I>fewer</I> subprime loans before the recession happened?
For those interested in dynamical systems and the brain - check out Chris Eliasmith and his lab http://watarts.uwaterloo.ca/~celiasmi/cv.html
> What we know about metabolism so far looks like this...
As complex and intimidating as that chart seems, it's really not even the half of it. It's missing:
- Bile acids
- Prostaglandins
- NO signaling and more detailed arginine metabolism (e.g., S-Nitrosoglutathione)
- Advanced glycation end-products and their metabolism (e.g., glucosepane)
- Endocannabinoids (which overlaps with endogenous TRPV agonist metabolism, also not on the chart.)
And much, much more.
A full chart, even one that's restricted only to what's currently known, would be at least five times the size. Which, I suppose, sharpens your point.
Completely foreign and exotic aliens which have no corresponding concepts for "War" and "Money", but can perform sentiment analysis to figure out likes/dislikes from facial expressions? They might not even have faces!
I'm just gonna add a little to the dynamical system and attractors perspective - sometimes a dynamical system has attractors which are not points in phase-space (ie, the multidimensional space of possible states of the system) but rather closed lines (limit cycles). This would mean that the system travels along those lines in a periodic way, and any other orbit in phase space eventually reaches these closed lines. A physical example of this is a mass on a spring with friction that sometimes slows it down and other times speeds it up (van der Pol oscillator).
Do you think this could be a good way to understand disorders which have periodic behavior? Say, is bipolar manic and depressive episodes better understood as "external shocks move the system between stable points", or "the internal dynamics of the system moves it along a periodic trajectory between two extremes"?
I have definitely seen bipolar disorders in both categories. Some bipolar patients have episodes whenever they travel (because the jet lag messes up their circadian rhythm) - this is a clear external shock. Other bipolar patients have cycles you can set your watch to. Sometimes almost literally - some ultradian cycling patients will predictably have a mood crash at a certain time of day.
Tangent - "But also, the refinery workers might strike in sympathy with the truckers" is a problem that a robust system of social castes solves! If refining and transportation are done by two separate tribes who don't interact socially, don't intermarry, and don't try to horn in on each other's economic domains, then the odds of one tribe paying a cost to show sympathy to the other at the same time that that other tribe is purposefully making things difficult for the first one are essentially nil.
Leaving aside for a second how weird a tangent this is, I'm not even sure it's true. Black people and white people are pretty socially segregated, but white people have been protesting in solidarity with black people a lot recently, and I doubt it's just that the white protesters personally know a lot of blacks.
I'd wager it's more the opposite and they don't know very many blacks. Baizuo and all that. There's probably few whiter things you could do in the U.S. than social justice activism.
This was a fantastic read. Thank you.
I wonder if a dice roll is a dynamic system with 6 attractor states? It seems like there is some graph that maps a dice leaving the player's hand to its resting state on one face, but the graph is so large that we can't understand it as a system and so think of it as random.
I don't mean to be cynical. It's more important to understand the economy and our mental health than it is to understand which way a dice will fall. Maybe it's even more important to understand just how hard that endeavour is and remind ourselves that anyone who is too confident in their interventions is (at best) a benevolent alien.
Thanks again!
In theory you could perfectly predict a dice roll, given the exact speed and position of the dice leaving the hand, the topography and materials of the rolling surface, the forecast for local micro air currents for the next few seconds, etc.
In other news, my new casino will replace dice with a list of people with bipolar disorder. You just bet then call one and ask for their current state.
Heh, thanks. This is terrifying.
Yes, a die about to land is more or less a system with 6 attractor states. (You could nitpick about the details and how the angles matter and how the 6 states are more like 6 attractor subsets, but more or less this is right.) And it isn't even a complicated and mysterious system at all: it's extremely well-understood, as far as systems go. The reason we treat it as hard to predict as random is that it has a property, that if you measure the starting state with an even very slight error (and you always will), the error at some points can start growing very fast and render your long-term prediction wrong. (Do the systems behind you mental processes also have this hard to predict property? We don't know for sure, but it seems very plausible.)
Thank you. Your point about the problem of measuring the starting state is great. I had not thought about that piece of the puzzle.
I imagine it's harder to do this for dice, but people have successfully developed computers that can predict the outcome of a roulette wheel, by measuring the speed and position of the ball. (One group used a computer hidden in their shoe, tapping their foot when the ball came around to measure the speed.)
It's not perfect, obviously, but it's good enough to beat the house, so casinos now keep an eye out for these shenanigans.
I saw that metabolism diagram and my immediate reaction was "Wow, it's like the US military's procurement system!" https://graylinegroup.com/wp-content/uploads/2017/05/atl_wall_chart-1.jpg
They even keep the three letter acronyms and have a Krebs Cycle stand-in!
They're better at color-coding, though.
Typo:
"People often ask: is depression the just same thing as normal sadness?" -> "just the same thing"
>MTHFR
Can I be the only one who mentally expands this into Samuel L Jackson's favourite epithet?
I probably was, but maybe now others will suffer with me.
Have I created a new attractor state?
I've had it with these MTHFR states on this MTHFR brain!
I did that too when I read it. Glad to know I'm not alone
“large stimulus packages can sometimes shift the system from recession to normal growth”
To economists out there: is there an example of this actually happening? I know the Great Depression is the era most associated with Keynesian economics, but my understanding is that the increase in taxes (starting under Hoover) made policy not all that Keynesian, and it lasted a long time (even having another recession prior to full recovery).
The APA joke got me; I literally laughed out loud. Weirdly enough, this is the most true description of trying to treat depression I've ever seen.
Scott, can you please write something about ketamine, which is revolutionizing mental health?
https://www.google.com/search?client=firefox-b-1-d&q=slatestarcodex+ketamine
I wonder what you make of this one:
"Bupropion Normalizes Cognitive Performance in Patients With Depression"
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1924979/
Bogus or legit?
First time here. Great article!
I'm attaching a celebrated paper which talks about the neuroscience of memory retrieval. It also takes the dynamical systems approach in order to explain how we retrieve memories in our brain. The strength of the paper lies in the fact that it does not make strong assumptions about the brain, and most parameters can be "random". https://www.pnas.org/content/79/8/2554
As an English computer scientist, I was confused both by your use of "gas" to mean "petrol" (I assumed by "gas lines" you meant "pipes for natural gas", not "petrol queues") and your use of "i" and "j" for boolean variables rather than indices.
Just wanted to note that Scott's drawings may be slightly misleading. Critical slowing down is a feature that appears only when the system is going from one macro state to the other (the attractors in Scott's language, although the phenomena occurs in non dynamical systems too), but not in the vicinity of the basin of attraction, so no slowing down near an attractor, even though the drawing might suggest that due to the ground being level there.
It's also funny that the link toward an example of critical slowing down in the global economy is an article claiming there is no evidence for it, which seems to conform to a brief internet search on the terms.
On the very misleading side is calling something fractally complicated due to having causes that have causes,etc. The defining feature of a fractal is self-similarity, so that you do not have to know the underlying cause, everything you need to know is at whichever scale you're looking at and you can understand this scale while completely oblivious to finer details (which are just more of the same).
I'm happy that Scott is back and double happy he is posting very regularly, but throughout the first parts I felt being constantly threatened with math and links to peer-review until I stopped to review some things.
As an expert in dynamical systems, I like this way of looking at the world. Most things are dynamical, and have many variables. A dynamical model of the brain would probably have to contain as many variables as there are neurons. That being said, a surprisingly large number of complicated things behave as if they only have a low number of variables. For example, the transition to turbulence can (usually) be well understood using ideas of chaos theory with a few variables. These sorts of reductions require some sort of continuity or dissipation, which are lacking in the brain or the economy.
I want to correct one error: the opening example has only one dynamical variable. The system can be written in one line, using the sign function (which is +1 if the input is positive and -1 if the input is negative - and for this example, we also want sgn(0) = +1).
h(t+1) = h(t) + sgn( h(t) - 75 )
If it did have more than one dynamical variable, you wouldn't have been able to understand what it does using a one dimensional picture. The picture would be more accurate as a tent, with straight lines going up to a point above 75.
I loved Borsboom papers when they were published, but I've seen criticism about their replication (https://psych-networks.com/network-models-do-not-replicate-not/ and https://psych-networks.com/7-new-papers-network-replicability/); I guess in order to strenghten the model, unsurpisingly, more research is needed.
On a tangential note I suffer from atypical depression, though it's likely that my attractors are shallow and I'm moving among those faster than classic MDD (I suspect this being linked to my Delayed Sleep Phase Syndrome messing up the circadian rhytm). My diagnosis could also be incorrect since it's more similar to cyclothymia
This is an interesting and illuminative post. I wonder though about the overly exclusive physiological perspective. I agree the physiological complexity is staggering but there seem to be 2 additional areas left out though you do refer to them obliquely.
The feedback loops created by the external environment and the recursive element of the mind whether conscious or unconscious. Your distinction that the difference between sadness and depression might be that depression is a "stable" attractor state is one I'm not sure I agree with. It could be that there is some variable susceptibility but could it not be the feedback between all 3 systems that deepens the state so that it's difficult to emerge from it without help. If that was true it would perhaps explain why there is no substantive difference between sadness and depression at the outset. I am extending the inference of course that duration is the defining difference, as suggested by taxometrics, and that it's the complex dynamic feedback loops that deepens and creates your attractor state?