The Intense World Theory of Autism

by Steven Byrnes20 min read27th Sep 202122 comments

75

AutismPsychopathyNeuroscienceWorld Modeling
Frontpage

The first section is well-written and reliable, because I didn't write it, it's a book excerpt. The rest of the post is more prone to errors and speculations; you can skip it if you're not into that kind of stuff—I read a few books and articles about autism, but I'm definitely not an expert, and I didn't spend a lot of time checking everything I wrote. Please comment if anything seems wrong (or dumb, or offensive, etc.) so I can fix it.

Summary / Table of Contents

  • The first section is mostly a book excerpt introducing my favorite theory of how autism works in the brain, namely "Intense World Theory".
  • Then, I'll flesh out how I'm thinking about that theory, by distinguishing "intense world" and "different learning algorithm hyperparameters" as things that happen in different parts of the brain and have different consequences, even if they tend to go together. I'll talk a bit more about each separately, and try to relate them more specifically and mechanistically to the algorithms going on in different parts of the brain.
    • I'll also include how I think this theory connects to other famous aspects/theories of autism, like cerebellar abnormalities, memory, "weak central coherence", etc.
  • At the end, I'll speculate that there's an exact-opposite "dim world theory of psychopathy". Fun!

Introduction to the Intense World Theory of Autism

There are various theories of the root cause of autism in the brain. I'll mention a few as I go. But I'll start right in with the positive case for the theory I like: The Intense World Theory of Autism.

The Intense World Theory of Autism dates to this 2007 article. I first heard about it from the excellent book The Myth of Mirror Neurons, which devotes a whole chapter to it, excerpted below. Then I read Temple Grandin's The Autistic Brain: Thinking Across the Spectrum and found that she also brought it up, and seemed very enthusiastic about it. I've also read a couple review articles on Intense World Theory, including this one by some of the inventors of the theory, and this one and a few others. Everything I've found is pretty positive the theory—please comment if you've seen any substantive criticism.

I can't dream of describing the theory as well as Gregory Hickok does in The Myth of Mirror Neurons, so I'll just excerpt from that. First some motivation:  

…I'm going to suggest the possibility that the dominant neurocognitive theories of autism, which assume that behavioral deficits result from lack of or diminished social sensitivity, have it wrong and in fact have it backward.

"Deficit theories" of dysfunction are reasonable and intuitive. If an individual fails to respond normally to sound, it’s a good bet that the person has a diminished capacity to process and hear sound. He simply isn’t capable of perceiving the signal. Likewise, if another individual fails to respond normally to social stimulation, it’s a reasonable bet that the person has a diminished capacity to process social information. But consider the following thought experiment. Imagine you had a stadium rock concert–type sound system hooked up to your living room television and you attempted to watch the evening news with the sound cranked up all the way. Most likely, you would cover your ears and quickly leave. If you forced yourself to stay, you would run into at least one of three problems as you tried to listen and watch. One, the physical pain would be so extreme that you wouldn’t be able to concentrate on the message. Two, attempts to dampen the sound and ease the pain, say by sticking your fingers in your ears, would filter out many of the fine details you need to hear normally. You would perceive less well. Three, if you did manage to listen, the extreme volume would excite so many nerve fibers that it would drown out the details of the signal itself and again you would miss many things. Excess can be as detrimental to normal function as paucity.…

Behavior does not automatically reveal its cause and can be misleading….

Then he switches to rats:

In fact, Henry Markram, Tania Rinaldi, and Kamila Markram proposed such a theory in a 2007 article aptly titled, "The Intense World Syndrome—An Alternative Hypothesis for Autism." The theory is grounded, oddly enough, in a rat model of autism. I say "oddly enough" because autism has traditionally been considered a uniquely human disorder, with defining symptoms showing up in high-level social and language domains. But it turns out that rats who are exposed prenatally to valproic acid—a compound used in human medications to control seizures and bipolar disorders—develop some key features of autism both neurally and behaviorally, including loss of cerebellar neurons, abnormalities in the serotonergic system, decreased social interactions, increased repetitive behaviors, enhanced anxiety, motor abnormalities, and sensory hypersensitivity. Curiously, the prevalence of autism in humans who are prenatally exposed to valproic acid through maternal use of the medication is substantially higher (one estimate is 11–100 times) than in the general population…

The existence of a rat model makes it possible to explore, in substantial detail, the neural bases of an autistic-like phenotype. The Markrams and their team did precisely that.…

They determined that local neuronal networks in the three brain regions tested in rats are hyperreactive….

The driving force behind the micro network-level hyperreactivity appears to be the direct connections between neurons. Valproic acid-treated animals exhibit a 50 percent increase in the number of direct connections between neurons within a local circuit….

Further, the Markrams and their team found that neural networks in valproic acid–treated rats are also hyperplastic... When connectivity patterns between neurons in local and expanded networks were examined before and after widespread and prolonged (overnight) activation of the entire network, valproic acid–treated rats exhibited an increase in the rate of rewiring, mostly evident in the nonlocal networks, compared to controls…

And back to humans:

Given these kinds of neural changes in the rat model of autism, a plausible story can be told about the neural basis of the range of autistic behaviors. Hyperabilities, such as increased sensory sensitivity or memory in specialized domains, can be explained by hyperreactivity and hyperplasticity of neural circuits. Hyporesponse to social stimuli can be explained in terms of the emotional intensity of the signal, which triggers anxiety and avoidance responses, which means less information is acquired in individual social situations and over time reduced opportunities to learn in the social domain. Theory of mind performance would also be expected to suffer with a hyperactive response to social signals—even if there is no fundamental deficit in mentalizing—because of increased anxiety when interacting with others and/or because avoidance behavior decreases the amount of information perceived or learned. Repetitive behaviors can be viewed as a coping mechanism aimed at regulating the child’s intense world. Motor deficits can be explained by hyperexcitability of the response to sensory stimulation, which has motor consequences, as we’ve discussed, or from hyperreactivity of motor systems themselves. And because language is at some levels a sensorimotor task and at other levels a highly social behavior, abnormalities in the sensorimotor or social domains can be expected to affect language.

All of this is interesting—suggestive even—but how relevant is it, really, to autism?…

Hyper-responsivity leading to avoidance…is observed regularly and uncontroversially in the sensory domain. Autistic individuals often cover their ears when even moderately loud sounds are present in the environment and exhibit other forms of avoidance behavior….

At least one study has confirmed that alternative explanations of the face processing "dysfunctions" in autism may be on the right track. Autistic and nonautistic individuals were scanned using fMRI while they looked at pictures of faces that were either emotionally neutral or emotionally charged. Crucially, using eye-tracking technology, the researchers also monitored which parts of the images their participants were looking at during the experiment. Overall, autistic participants activated their fusiform face region less vigorously than nonautistic controls, replicating previous work. But the eye-tracking data showed that this was simply because they spent less time looking at the most informative region of the faces, the eyes. In fact, when the researchers looked at fusiform activation as a function of time spent fixating on the eyes in the photos, they found a strong positive correlation in the autistic group. This means that the autistic brain is responding quite well to face stimuli, if one takes into account the amount of time spent looking at them...the same study reported that amygdala activation was stronger in the autistic compared to the nonautistic group while looking at faces.

Also consistent with the alternative, emotional hyperreactivity hypothesis are statements from autistic individuals themselves. Here’s a sample gleaned from a paper covering face processing in autism:

It’s painful for me to look at other people’s faces. Other people’s eyes and mouths are especially hard for me to look at. My lack of eye contact sometimes makes people, especially my teachers and professors, think that I’m not paying attention to them. —Matthew Ward, student, University of Wisconsin

Eyes are very intense and show emotions. It can feel creepy to be searched with the eyes. Some autistic people don’t even look at the eyes of actors or news reporters on television.—Jasmine Lee O'Neill, Author

For all my life, my brothers and everyone up 'til very recently, have been trying to make me look at them straight in the face. And that is about the hardest thing that I, as an autistic person, can do, because it’s like hypnosis. And you’re looking at each other square in the eye, and it’s very draining. —Lars Perner, Professor, San Diego State University

These are revealing statements for two reasons. First, they provide a clear indication of an intact theory of mind in these individuals ("my lack of eye contact…makes people…think that…"). And second, active avoidance of eye contact provides just as much evidence for sensitivity to the information contained therein as does active engagement of eye contact. If you can’t recognize that there is information in the eyes, why avoid them?

There's a lot more in the book chapter, but I'm probably already pushing the limits of copyright law. I'll leave it at that. Again, this is from the book Myth of Mirror Neurons, which I heartily endorse.

Honestly, I should probably just end the blog post here. But I couldn't resist adding some more commentary and speculation, particularly trying to shift the level of discussion towards lower-level neuroscience and algorithms.

Deeper dive part 1: "Intense world"

Sensory sensitivity

Here's a chart:

Unexpected sound stimulusNeurotypicalAustistic
Very quiet chirpNot noticedBit of a startle
Moderately loud bangBit of a startleJump-out-of-your-seat surprise!
Firecracker next to your earJump-out-of-your-seat surprise!Aaaaaaaaaa!

I could make a similar chart for vision, touch, or any other sense. Scott Alexander brings up here the example of scratchy tags on the back of shirts, which might repeatedly draw the attention of a person with autism, yet be below the threshold where they draw the attention of most neurotypical people.

What's going on in the brain here? 

The starting point has to be innate, not learned. We're talking here about startle reflexes, orienting reactions (e.g. where you turn your head and body towards a sound), flinching, releasing cortisol, etc. Nobody teaches you how to do those things! Moreover, there's nothing in your sensory input data that says that an unexpected 27 dB sound is sufficiently threatening to warrant startle and cortisol, but an unexpected 19 dB sound is not. This is a heuristic threshold set by the genome. No wonder it's different in different people! (It could also be modified over time by experience, I'm not sure. But anyway, it has to start somewhere.)

I believe that the brainstem (specifically, superior colliculus & inferior colliculus) is where you'll find these innate circuits—circuits that take a sound (or flash or touch etc.), figure out where it's coming from, and if it's sufficiently loud, execute startle reactions and orienting reactions and so on.

But that doesn't mean that these brainstem circuits are necessarily the underlying cause of sensory sensitivity in any particular person. I think other parts of the brain (amygdala, medial prefrontal cortex, cerebellum) function sorta like "eager servants" of the brainstem in this task—they learn to anticipate when the brainstem is going to execute a startle, and they jump the gun, issuing those commands themselves. So if any of those systems are really trigger-happy—i.e. they issue a command in any circumstance where they think there's even the slightest chance that the brainstem would issue that command—you could get the same kind of sensory sensitivity, seems to me.

By the way, this is how I'm thinking about the (apparently) causal role of the cerebellum in autism.

Eye contact

I think eye contact is basically the same as sensory sensitivity. I believe the brainstem (superior colliculus) detects eye contact, and issues (among other things) arousal reactions ("arousal" in the psychology jargon sense, not the sexual sense—e.g. cortisol release, higher heart rate, etc.). There seems to be an arousal sweet spot: too little arousal is boring, too much arousal is overwhelming. (There are many more dimensions to your feelings than just arousal, of course! But it does seem to me that arousal is of central importance.) As in the book excerpt at the top, people with autism apparently often find that eye contact flies way past the sweet spot into "overwhelming" territory. It's aversive, and therefore they avoid it.

This is my model, which I'll be using throughout this post. I'm not proud of it; it's a lousy model. Valence in fact depends on much more than just arousal; they can vary independently. But I'm using this model anyway, because I think it captures some important kernel of truth, and I don't know any better way to express that kernel of truth.

As in the sensory sensitivity section, the higher-than-typical arousal could come from several other sources besides the brainstem itself—particularly the amygdala, medial prefrontal cortex, and cerebellum.

Social interactions more generally

I think eye-contact-detection is just one example: my hunch is that the brainstem (again, the superior & inferior colliculus) has a whole suite of heuristics for a person being near to and interacting with other people—for example, human-speech-sound-detection, and human-touch-detection, human-smell-detection, and so on. All of them create arousal, I assume, and therefore all of them might potentially create an overwhelming level of arousal in people with autism, who may correspondingly try to avoid those things.

Next up is empathetic simulation. As I discussed (speculatively) here, I think the main neurotypical way of understanding and interacting with other people is by empathetically simulating them, constantly and without any deliberate effort. (I call this "little glimpses of empathy"—to be strongly distinguished with the deliberate, slow, effortful "empathy" that people are usually talking about when they say "empathy" in everyday contexts.)

This kind of empathetic simulation is tied up with social instincts, and their corresponding arousal, which as above may be overwhelming and aversive for some people with autism.

I figure what happens as a consequence, at least sometimes, is: the person with autism gradually learns to interact with and understand people, but without using empathetic simulation. Instead, they just take their general intelligence, and leverage it to build a new human model from the ground up, just as people can model any complicated system from the ground up (e.g. a complicated piece of software). So there's still a human model, but it's built on a different foundation—a foundation without that strong innate connection to brainstem circuits.

I think this explains why (IIUC) there are intelligent people with autism who are able to do theory-of-mind-type reasoning (as in the book excerpt at the top), and able to understand social interactions and conventions, but need to deliberately learn aspects of these things that neurotypical people might find intuitive and effortless. The intuitive-and-effortless part is the pathway that goes

      quick effortless empathetic simulation
                → social instincts hardwired into the brainstem
                             → resulting "feeling"

…again as discussed here. (Further clarification in this comment.)

Deeper dive part 2: "Different learning algorithm hyperparameters"

I believe that a very large fraction of the human brain (maybe 98% by volume) exists for the purpose of within-lifetime learning (see discussion of "learning-from-scratch-ism" here). They thus house learning algorithms, or more specifically "learning-and-inference algorithms". (The "learning" part is editing synaptic connections for future use; the "inference" part is using previously-learned content for taking better actions now.)

These learning-and-inference algorithms, like any learning-and-inference algorithms, have what we call "hyperparameters" in machine learning.

Here's an example: If you've seen a pattern "A then B then C" recur 10 times in a row, you will start unconsciously expecting AB to be followed by C. But "should" you expect AB to be followed by C after seeing ABC only 2 times? Or what if you've seen the pattern ABC recur 72 times in a row, but then saw AB(not C) twice? What "should" a learning algorithm expect in those cases?

You can imagine a continuous family of learning algorithms, that operate on the same underlying principles, but have different "settings" for deciding the answer to these types of questions.

And I emphasize that this is one of many examples. "How long should the algorithm hold onto memories (other things equal)?" "How similar do two situations need to be before you reason about one by analogizing to the other?" "How much learning model capacity is allocated to each incoming signal line from the retina?" Etc. etc.

In all these cases, there is no "right" answer to the hyperparameter settings. It depends on the domain—how regular vs random are the environmental patterns you're learning? How stable are they over time? How serious are the consequences of false positives vs false negatives in different situations?

There may be an "optimal" set of hyperparameters from the perspective of "highest inclusive genetic fitness in such-and-such specific biological niche". But there is a very wide range of hyperparameters which "work", in the sense that the algorithm does in fact learn things. Different hyperparameter settings would navigate the tradeoffs discussed above—one setting is better at remembering details, another is better at generalizing, another avoids overconfidence in novel situations, another minimizes energy consumption, etc. etc.

I actually already talked about learning-and-inference algorithm hyperparameters in the "intense world" discussion of the previous section—but without using those words. Specifically I said that the brainstem has innate capability to do startle reactions, but the amygdala, medial prefrontal cortex, and cerebellum all (in different ways) learn to anticipate the brainstem. And this algorithm has a tradeoff between false positives and false negatives. And this tradeoff depends on—you guessed it—hyperparameters. "Intense world" would come from the algorithms being tuned to have almost no false negatives, at the expense of lots of false positives.

In this section, I'm talking about learning-and-inference algorithm hyperparameters more generally, focusing more on memory and cognition.

I can think of two reasons that an autistic brain might wind up with systematically different learning-and-inference algorithm hyperparameters (for memory and cognition) than a neurotypical brain:

  1. Maybe, like in the valproic acid rats above, there is an underlying biophysical cause of "hyperreactive and hyperplastic" neurons. When this root cause impacts the neurons in certain parts of the brainstem / amygdala / cerebellum / whatever, it manifests as the "intense world" stuff above. And when this same root cause impacts the neurons in, say, the fusiform face area, it manifests as "different hyperparameters" in regards to learning and recognizing faces. And so on with other learning areas and modalities.
  2. The brain has a capability of doing real-time hyperparameter variation, in a situation-dependent way, using various neurotransmitters like dopamine, serotonin, norepinephrine, and so on. And there seems to be at least two signals of this type (namely acetylcholine—see here—and serotonin) that more-or-less links arousal level to hyperparameters, I think. So if the "intense world" stuff of the previous section leads to high arousal all the time, that would in turn affect the learning algorithm hyperparameters. (I got that idea from Steve Grossberg's book.)

Finally, here are a couple typical aspects of autism that seem to be in the "different hyperparameters for memory and cognition" category:

  • Autism and memory: Given the same information, I think there are systematic (though not universal) differences between what the memories formed by neurotypical people and people with autism. This can lead the latter to have both "deficits" and "savant-like" memory abilities—even in the same person.
  • Weak central coherence theory: According to this review, it seems likely (though not definite) that the valproic acid rats have a higher ratio of local-connections-vs-long-range connections in the neocortex. (This comes from an increase in local connections, not a decrease in long-range connections.) This seems at least vaguely to go along with the idea that their neocortex would learn and remembering more narrow and specific aspects of things, and fewer cross-domain, analogizing, big-picture aspects of things, e.g. aspects that incorporate multiple senses and abstract context.

Bonus: "Dim world theory of psychopathy"??

To be clear, this part is my own wild speculation. Call it the "dim world theory of psychopathy". Psychopathy, in this theory, is when the brain is hyporeactive and hypoplastic, as opposed to autism where it's hyperreactive and hyperplastic.

When people with autism activate their empathy brain circuits, it reacts so overwhelmingly that they learn early in childhood to avoid activating those circuits in the first place, using techniques like avoiding eye contact, avoiding the use of empathetic simulation as a method of understanding people, and so on.

Conversely, when people with psychopathy activate their empathy brain circuits, there's barely a whisper of activity. Why do psychopathic kids stereotypically torture animals? Because for them to feel anything at all, it takes a situation that would be emotionally overwhelming for neurotypical people.

SituationAutisticNeurotypicalPsychopathic
I'm sitting in
a quiet empty room,
moving my toes
Stimulating
😀
Boring
🙄
Super
boring
😡
I'm talking to
someone, while
making eye contact
and empathetically
simulating them.
Overwhelming
😞
Stimulating
😀
Boring
🙄
I'm lying to someone's
face as part of a
deliberate scheme to
make them
miserable
Super
overwhelming
😱
Overwhelming
😞
Stimulating
😀

I did a little search for studies about sensory sensitivity etc. in psychopaths, and didn't have much luck finding anything very informative. I didn't try very hard, maybe it's out there. (If anyone follows up, let me know what you find. (Wanna coauthor a follow-up post with me? :) ) (Update: here's a possible lead—this wikipedia article talks a bit about "low arousal theory" and psychopathy. The references they provide for that are a bit underwhelming though.) For now, take this as casual speculation by a non-expert, as if we were chatting over drinks.

Hang on, isn't schizophrenia the opposite of autism?

I guess there's a "diametrical" theory where autism and schizophrenia are opposites. I dunno, I know very little about schizophrenia. Here's a random thought though.

People with autism and people with psychopathy are both basically not feeling much effect of their innate social instincts most of the time as they go about their lives. It's for totally opposite reasons! And with radically different consequences! But still, this is a narrow respect in which autism and psychopathy are not opposites. So then we can talk about a different "axis" along which schizophrenia might be the opposite of autism. Maybe people with schizophrenia have their innate social instincts very powerfully active all the time, for whatever reason. Maybe those circuits just randomly trigger all the time for no reason (as opposed to the neurotypical case when they trigger at specific times like eye contact, body contact, empathetic simulation, etc.).

By the way, I don't expect that this is a complete description of schizophrenia; I would assume that (like autism) schizophrenia involves some neuron-related thing that impacts different brain systems in different ways.

My model seems to predict that it should be possible to be both autistic and schizophrenic, whereas it shouldn't be possible to be psychopathic and autistic. As far as I can tell, both those are basically right. Simon Baron-Cohen does say here that he treated a person with both autism and borderline personality disorder (BPD) (if memory serves). BPD is a bit like psychopathy, in that both involve a lack of empathy. Narcissism is in this category too. But BPD is clearly different from psychopathy. By the way, don't ask me for a theory of BPD or narcissism—I have no idea.

75

21 comments, sorted by Highlighting new comments since Today at 9:49 PM
New Comment

The intense world theory sounds similar to Eysenck's theory of introversion vs extraversion from the 1960s. One summary:

Eysenck (2008/1967) believed that differences in extraversion were due to physiological differences in brain systems that caused some people to be more easily aroused than others. Specifically, he proposed that the ascending reticular activating system (ARAS) regulated arousal. People differ in the sensitivity of this system, which causes them to respond differently to their environment. A moderately stimulating environment might cause people with a very sensitive ARAS to feel overstimulated and retreat (i.e., to behave like introverts), but people with a less sensitive ARAS to feel understimulated and seek out additional stimulation (i.e., to behave like extraverts).

The inverted U arousal graph is typically plotted as Performance vs. Arousal and called the Yerkes–Dodson law.

Another summary here; more findable with keywords like "optimal level of arousal", Eysenck, and extraversion.

Great post, as always.

They determined that local neuronal networks in the three brain regions tested in rats are hyperreactive….

Now that's an experimental validation.

Further, the Markrams and their team found that neural networks in valproic acid–treated rats are also hyperplastic... When connectivity patterns between neurons in local and expanded networks were examined before and after widespread and prolonged (overnight) activation of the entire network, valproic acid–treated rats exhibited an increase in the rate of rewiring, mostly evident in the nonlocal networks, compared to controls…

If this applies to autistic humans too then it would have interesting implications.

Intense World

Your sensory sensitivity chart is accurate for autistic people (acknowledging that there is variation from person to person). For me in particular, it slightly underestimates how easily I get startled.

The starting point has to be innate, not learned. We're talking here about startle reflexes, orienting reactions (e.g. where you turn your head and body towards a sound), flinching, releasing cortisol, etc. Nobody teaches you how to do those things! Moreover, there's nothing in your sensory input data that says that an unexpected 27 dB sound is sufficiently threatening to warrant startle and cortisol, but an unexpected 19 dB sound is not. This is a heuristic threshold set by the genome. No wonder it's different in different people! (It could also be modified over time by experience, I'm not sure. But anyway, it has to start somewhere.)

I always like information that helps me distinguish hard-coded from trainable attributes.

I figure what happens as a consequence, at least sometimes, is: the person with autism gradually learns to interact with and understand people, but without using empathetic simulation. Instead, they just take their general intelligence, and leverage it to build a new human model from the ground up, just as people can model any complicated system from the ground up (e.g. a complicated piece of software). So there's still a human model, but it's built on a different foundation—a foundation without that strong innate connection to brainstem circuits.

I think this explains why (IIUC) there are intelligent people with autism who are able to do theory-of-mind-type reasoning (as in the book excerpt at the top), and able to understand social interactions and conventions, but need to deliberately learn aspects of these things that neurotypical people might find intuitive and effortless.

I can confirm this happened for at least . I built a model of human beings from first principles by systematically combing through the scientific, historical and economic literature.

Schizophrenia

Maybe people with schizophrenia have their innate social instincts very powerfully active all the time, for whatever reason.

When I interact with fortune tellers, I feel like their social intuitive sense is supercharged in a way that's very similar to reports of schizophrenia. The difference is that healthy fortune tellers can tell that their intuitions in their head are not "real" in the physical sense whereas schizophrenics can't tell the difference. (It gets confusing because fortune tellers frequently do believe in literal magic [which is not real] but there's a big difference between being a Wiccan and being a schizophrenic.)

Hyporesponse to social stimuli can be explained in terms of the emotional intensity of the signal...

Also: attempting to model another person's mind works much less well if the other person's mind works much differently. (This contrasts with your model in which autistic poeple avoid using empathetic simulation. I propose they use it and it produces poor predictions, but we could both be right.)

Personally I'm overwhelmed by some social signals (especially negative ones, which feel dramatically more intense than positive ones, easily to the point of disrupting my life), and I ambiguously fit the criteria for Asperger's. So I'm curious what surveys of autistic patients, and autopsies, have been done to check this theory.

My (weak!) understanding of psychopathy was that their theory of mind is pretty much fine except that the "empathy" part of it fails in some way. In my model (1st paragraph), if psychopathy were a matter of underconnection, psychopaths should have mispredictions when simulating other people, but they have a reputation for charisma, charm and backstabbing, suggesting high social competence but low empathy. Indeed, low empathy (by itself) might even improve the ability to simulate because higher brain functions that want to do analysis are less interrupted by emotion. (Perhaps a high-functioning psychopath has reduced empathy, but not reduced to zero: enough empathy to predict other people, without the irritation of intensity of feeling.)

My model seems to predict that it should be possible to be both autistic and schizophrenic, whereas it shouldn't be possible to be psychopathic and autistic. As far as I can tell, both those are basically right.

What's the evidence for this?

Also: attempting to model another person's mind works much less well if the other person's mind works much differently.

This is known as the Double Empathy Problem and has been described by Milton (2012) in On the ontological status of autism: the ‘double empathy problem’. Some excerpts:

In a sense it is a 'double problem' as both people experience it, and so it is not a singular problem located in any one person. Rather, it is based in the social interaction between two differently disposed social actors, the disjuncture being more severe for the non-autistic disposition as it is experienced as unusual, while for the ‘autistic person’ it is a common experience.

In analysing the interactions that ‘autistic people’ have with the wider population, it is easy to problematise the definition of autism as a ‘social deficit’ located within an individual’s mind. Differences in neurology may well produce differences in sociality, but not a ‘social deficit’ as compared to an idealised normative view of social reality. Such definitions may help to signpost disability support services, but they are no way of defining autism in any kind of holistic sense.

I want to distinguish two stories about empathy:

(1) (conventional story) The point of empathy is to form factually/predictively accurate beliefs about another person's state of mind and therefore what they will do and feel

(2) (my idiosyncratic story) The point of empathy is that, if you're talking to someone or thinking about someone, and you are aware of how they're feeling, that "awareness" (in neurotypical brains) takes the form of a transient automatic empathetic simulation of the other person, and this mechanism causes automatic "response feelings" in oneself—and these "response feelings" might be pro-social and pro-understanding, but also might not.

For example, if I think someone is angry at me, I might automatically get angry back. If I think someone is feeling condescending towards me, I might automatically get defensive and angry. If I think someone is sad, I might automatically feel sympathetic towards them (or feel triumphant over them! These are complicated circuits!) Etc. etc.

I was talking about (2). My theory is that in autism the "response feelings" in story (2) are so strong that they're aversive, and people with classic autism adopt the coping strategy of avoiding invoking them altogether, by avoiding the specific type of mental operation that I call "empathetic simulation". And in psychopaths, the "response feelings" are so weak as to be barely felt, and psychopaths adopt the coping strategy of seeking out situations that invoke unusually strong "response feelings", like watching people suffer.

As for (1), I do not in fact believe that the neurotypical suite of social instincts (including the transient effortless type of "empathetic simulation") are particularly effective at forming factually/predictively accurate beliefs about another person's state of mind.

(Example neurotypical conversation:
--"Hey Emily, you got mud all over the refrigerator again, c'mon you gotta clean that up!"
--"God, Mom, you're always talking about the fridge, it's like you're totally obsessed with it, we both know you don't REALLY care about the fridge, you're REALLY just pissed off that I found a boyfriend who loves me, and you just want me to be miserable and lonely like you are!"
…Well, not a lot of factually accurate beliefs about mind-states going on in this conversation!)

So in terms of (1), I would strongly agree with your proposal of "attempting to model another person's mind works much less well if the other person's mind works much differently". I would also add something like "all the ways that neurotypical people form factually inaccurate beliefs about other people's minds get treated as 'normal human relations', but all the ways that people with autism form factually inaccurate beliefs about other people's minds get treated as pathological." Some combination of those two things, I figure.

What's the evidence for this?

Umm, maybe 30 minutes of google searching :-P I'm open to being corrected obviously

There's a distinction that is important to make when it comes to empathy, between 'cognitive empathy' and 'emotional empathy', that is a good starting place. The empathy divide goes all ways when it comes to neurology - neurotypicals have a harder time understanding autistics and (the opposite neurotype of autistic folk, who might have been diagnosed psychopaths at some point but don't have a definite entry in the DSM-V).

Cognitive empathy is being able to understand another person's perspective and mental state. Everyone needs to learn this, but if people think similarly to you it is easier because you have some baseline assumptions that are correct. It is often more urgent to learn this for people who don't share your neurotype if you are autistic (or the specific neurotype that is psychopathic?-opposite-of-autism) because you aren't in the majority in most circumstances - and assuming that everyone thinks the same way you do leads to wrong conclusions that cause wrong predictions quite quickly.

Affective or emotional empathy. Responding to another's mental state with our own emotions, when we are affected by them. Sympathy, concern, and personal distress tend to fall into this. We vary as much as anybody on this one, with some potential complications - we may be hyperempathic, and emotionally respond more than neurotypicals to a wider range of stimuli, including perceiving impossible things like the emotional state of a toaster due to hypersensitivity. We may be alexithymic, and not understand our own emotions enough to be able to express them at all without some aggressive introspection, much less appropriately. We may be too overwhelmed by the emotion to use it to find an appropriate response (try literally feeling your pain under intense world theory). We may not know any appropriate response, or the generally appropriate response for our neurotype is not the appropriate one for yours (whereas it may have been a good guess with minds more like ours). If we do know an appropriate response, expressing ourselves with a script, with nonverbal communication, or any other 'odd' way may be more difficult to understand or receive on their end.

Is it possible to make the world "dimmer" by gradually exposing yourself to stronger and stronger stimuli?

Hmm, I was writing recently about how David Burns makes exposure therapy sound like a miracle cure for all sorts of things, including some things that I wouldn't have previously expected to be treatable by exposure therapy (and he says that exposure therapy doesn't have to look like the classic "graduated exposure"). But he doesn't say anything about sensory sensitivity or autism in the book.

So I dunno. I guess it doesn't sound totally crazy off the top of my head, but that's just casual speculation, from someone who is an expert on neither exposure therapy nor autism. :-)

This link seems to be "we tried exposure therapy for sound sensitivity, and it didn't work", but I would be more inclined to describe it as "this isn't proper exposure therapy at all, it's just making loud noises in kids' ears for a minute or two". This seems relevant but I don't know how to find the associated paper if any. This and this makes it sound like exposure therapy for sensory sensitivity is already routinely used for people with autism, at least to some extent, and it works like normal.

Many years ago I read this NYTimes article (paywalled, DM or email me for a PDF) about Applied Behavior Analysis (note: see Ann Brown's reply to this comment), claiming it could "cure" classic autism; if that's true (a big "if"!) I guess we should consider the hypothesis that the underlying mechanism is "basically just lots and lots of exposure therapy for everything, day after day for years".

Many autistic individuals who experienced it have attested that Applied Behavior Analysis is an extremely unpleasant experience that caused them harm. It is a form of operant conditioning, that trains us to behave neurotypically and not behave autistically through reward/punishment. It does not 'cure' but control.

Thanks for sharing, I didn't know that.

My autistic child used to be terrified of the lawn mower, even if he was inside. We couldn't use the mixer, the vacuum, or even the shower without him freaking out. 

He went from terror to thinking these things were cool: if I cut the grass, he comes out to play nearby with his toy mower; he loves to vacuum. And -- glory of glories -- the shower is boring.

So I think for him at least, it's a progression from WAY TOO MUCH to fascinating and fun to bo-ring. 

This leads me to wonder: if it's possible for a stimulus to be overwhelming but not too overwhelming, if exposure therapy might help rather than just making him freak. It's worth a try.

Oops, just thought of this: he loves slurpy noises near his ears. Shouldn't that be way too stimulating? It would be for me or anyone else I know! Seems autism is both about avoiding/muting stimuli and seeking them out.

if it's possible for a stimulus to be overwhelming but not too overwhelming, if exposure therapy might help rather than just making him freak. It's worth a try.

I was chatting about exposure therapy in this other comment. I wonder if we should say that everyday life can sometimes provide "unintentional exposure therapy", in which case it sounds like "exposure therapy" (in that broader sense) has already been working out for y'all, which would be really neat and interesting.

Reminder that I'm extremely not an expert and don't trust me on anything. :-)

he loves slurpy noises near his ears. Shouldn't that be way too stimulating?

Hmm, it might be an example of how that picture I drew—where valence is an inverted-U function of arousal—is a lousy model. Highly-stimulating things aren't always aversive. Like, the best and worst moments of my life were both highly-stimulating. Maybe it's like, highly-stimulating things can be very bad or very good, and not so much in between? And whether it's good or bad depends on other things going on in your brain—for example, how safe you feel. If you feel simultaneously safe and uncomfortable/scared/threatened, the two things mix together like baking soda and vinegar, and the result is…laughter! Or at least, that's what I've been figuring. To be clear, everything I'm saying here is purely coming from introspection / folk psychology, I don't have any more insight than you or anyone else.

We definitely often have both sensory-seeking and sensory-avoiding needs. I am not as sensitive to some elements of the world now as I used to be as a child, but unsure how much of that is because I was a child and now am not, compared to 'getting used to' things (or some partial loss of the actual sense rendering it moot).

Reasons it would not be worth a try are that sensory-avoided experiences are, as mentioned elsewhere, extremely painful. Literally painful, as I was surprised I had to clarify at one point - apparently people without sensory aversions can find sensory experiences metaphorically? painful without them quite registering on the same scale as getting stabbed? But I would definitely say that while I have very much not enjoyed bad ongoing physical pain, bad ongoing sensory-avoid is for my normal experiences of both a great deal worse. A migraine makes me miserable; having to endure a sufficiently unpleasant sound on an ongoing basis without recourse could have me attacking walls / trying to hit my head on things if I didn't know better / other means to try to override and distract with other sensory input.

Great article. Thank you!

I also highly commend reading the original paper  referenced in the article. (full text available here https://www.frontiersin.org/articles/10.3389/neuro.01.1.1.006.2007/full) Reading the original paper was quite a revelation to me. How many things are parsimoniously explained by this hypothesis.... how many things suddenly make sense. As opposed to the jumble of unrelated facts seen in most academic summaries of Autism.

There is of course opposition to the hypothesis but there has also been considerable confirmation as a look at google scholar will easily show.

I'm not a neuroscientist but I think a nice corollary to your theory would be to look into the minds of those with addictions. I've read a lot about addictions online and talked to a few addicts who are addicted to different things (Alcohol, narcotics, etc). They all seem to have induced a dim world within themselves, but a specific one that is usually only acted out upon with their drug of choice, rather than like a psychopathic child killing animals. But late stage addicts do routinely do socially harmful behavior. All addicts seem to have a similar ratchet effect take place where they get their high, the high becomes their baseline, the rest of the world goes dim or less stimulating by default and to get a new high they have to go to a more potent drug / stimulating resource. 

I also noticed they usually seem to hide their warped utility function from themselves, only addressing it when it takes over their entire life (rock bottom). Which I think might be an interesting way to look at alignment problems because addicts are basically misaligned mesa-optimized humans that has their "inner process" hiding their warped true reward function from themselves. From the point of view of addiction recovery it's also interesting to see how a misaligned person tries to change their values to something that produces better long term outcomes for that person, to varying degrees of success. 

I don't know if looking into addictions would be something your interested in, but I figured it was worth bringing up when I read your dim world theory. 

This is interesting and as an autistic it suggests a new option. If this holds out then "taking a chill pill" is likely to tap into and develop those missing social capabilities. And maybe one aspect of it is triaging some of the arousal away so that the remaining functions remain operational.

One could aldso wonder whether co-occurance with depression is because the brain finds use and local optima with the passivity that comes with that. Some people might also try to do the minimum amount of autistic behaviour to make by such as keeping your gaze far enough that you are only panicing and not in terror (with maybe goal of trying to appear as neurotypical as possible). But if it is tought as overstimulation the proper responce would be to go to the most flow-like state, which would mainly be that the more you indulge the aversion the better one tends to do. So instead of being in panic, being stressed would be prerable and even further preference to be relaxed. Which in practical term would be that instead of finding social situations stressful, stop applying the one-size-fits-all shelf ready solutions and find a way to be comfortable in them even if it means your existence is a statistical outlier. You don't do better the lesser your offset is from the median.

I don't think the "on the fly" adjustement is fairly characterised as being hyperparameters. That is more like tuning regular parameters. Imagine you have car that veers to the left. You can still use it to straight if you in fact turn constantly to the right when you in fact want to go straight. The end result is very similar to no veering but your maximum turn left and maximun turn right are effected. If the situation calls for a very different balance that  is far from the hyperparameters optimal then having only 5 or so levers for neurotransmitters the balance is kep int a differnt way (say that you off-set your steering being veering by your brakes being opposite veering so you constantly brake despite applying throttle)

This feels related to Dąbrowski's overexcitability/superstimulatability theory of giftedness. As if there are two dimensions how acutely you are experiencing stuff (how laud is the TV to use your metaphor) and how capable are you to sustain stimulation (how laud can it be for you to hear everything clearly with all the details unheard on lower volumes).

(I feel intrigued by his work on positive disintegration but haven't looked into empirical evidence yet; but my personal experience supports weaker claims about it.  Actually, I don't think that it holds as strongly as the Wikipedia article presents it. But if it is directionally true, today's narrative around trauma be fairly misleading and at least unhelpful. )

Very interesting post! 

1) I wonder what your thoughts are on how "disentangled" having a "dim world" perspective and being psychopathic are (completely "entangled" being: all psychopaths experience dim world and all who experience dim world are psychopathic).  Maybe I'm also packing too many different ideas/connotations into the term "psychopathy". 

2) Also, the variability in humans' local neuronal connection and "long-range" neuronal connections seems really interesting to me. My very unsupported, weak suspicion is that perhaps there is a correlation between these ratios (or maybe the pure # of each), and the natural ability to learn information and develop expertise in a very narrow domain of things (music, math?) vs. develop big new ideas where the concepts are largely formed from cross-domain, interdisciplinary thinking. Do you have any thoughts on this? Depending on what we believe for this, what we believe for question 1) has some very interesting implications, I think?

3) Finally, I wonder if the lesswrong community has a higher rate of "dim world" perspective-havers (or "psychopaths in the narrowly defined sense of having lower thresholds for stimulation), than the base-rate of the general population.

Less Wrong is a text-based forum. It has no audio. Video is rare. It barely even has any pictures. I would be surprised if the userbase wasn't skewed toward people with lower thresholds for stimulation.

Maybe worth differentiating on kinds of simulation? LessWrong provides simulation through high-level cognitive processing, and competes with other sources of simulation also provided through high-level processing. I'm not sure I'd say audio is universally more simulating than text when it comes to this.

When it comes to say simulation of social functions and needs, audio-video does seem a bit more universal in being more simulating than text. Or even basic sensory simulation.