Hypothesis: Autism evolved to find Kolmogorov Complexity, implying strengths in utilizing Occam's Razor and Universal Turing Machines. In contrast, neurotypicals may rely more on continuous statistical information approaches.
Motivation for this paper:
Why there would be such a distinct cluster of symptoms for autism?
Why would list-making and stiff facial expressions go together?
Why should tunnel vision be related to overly formal speaking?
Why should autism be anecdotally correlated with physicists and tech founders?
Then, I thought that Kolmogorov Complexity must be the missing connection. It is a formalization of Occam's Razor and would explain Albert Einstein, Isaac Newton, Nikola Tesla. But Kolmogorov Complexity relies on Turing Completeness, so those who would be good at finding shortest programs must also be good minimum viable product programmers, like Elon Musk, Mark Zuckerberg, Vitalik Buterin, and Bill Gates.
If the hypothesis is true, then some amount of autism is necessary for civilization to function, because you need both types of information processing (Shannon and Kolmogorov). The hypothesis is pending empirical verification, but if it is true, autism could be reframed as an alternate, and necessary way of thinking, not as a disorder. Additionally, interventions to help autistic people (and neurotypicals) could be implemented in a more rigorous way, using methods from information theory. The environment could be measured to determine how much Kolmogorov vs Shannon type processing is necessary, then interventions based on the needs of the individual could be addressed. There would be multiple ways to do this, 1. collaboration 2. training 3. accommodation.
collaboration between information types would be like neurotypicals working with autistic people for example at a company. The Hypothesis also implies that neurotypicals benefit from using traditional computers, and I think this is part of the reason why autistic people may have more trouble in some ways in the modern environment. However, it also implies autistic people could benefit from working with Neural Networks, such as LLMs and image generators. Another implication is LLMs like GPT-4 calling traditional code.
Training: If the goal of someone is to become a savant, they can target certain skillsets that increase the ability to compute Kolmogorov Complexity. This may include rote memorization, learning algorithms, and improving focus, while avoiding logical paradoxes. If the goal of an autistic person is to become neurotypical, they may try to accumulate an ever increasing memory of social scripts and information about people, but not quite reach it.
Accommodation: The environment could be changed to support both Kolmogorov and Shannon-type processing. There could be areas where routines and logic are allowed, with other areas supporting unpredictability.
I think if this hypothesis is empirically validated, rigorous interventions could be implemented that improve the lives of autistic people using computer science concepts. What I personally would do (I'm on the spectrum) is try to partner with Neurotypicals and AI's, become a good programmer, and work on improving specific skills, while advocating for a autism by demonstrating it is necessary for robust information processing.
It's much simpler IMO to explain autism as what happens when the brain systematically overreacts to surprising inputs at every level. The reason e.g. stiff faces and list-making would go together is that they're both coping mechanisms to reduce the probability of being surprised by anything. This would also explain social difficulty, the desire for rote social rules, the focus on special interests (e.g. fields of knowledge that can be mastered and therefore predicted), and when combined with the predictive processing theory of motor control, could also explain difficulties in motor control as being a predictive processing failure.
Indeed, this fits with the hypothesis. Neurotypicals prefer a statistical, Shannon Entropy like approach, which can handle random surprises. Autistic people, by overreacting to surprising inputs, may avoid surprise, and end up with the deterministic information processing style (Kolmogorov Complexity).
However, the predictive processing failure doesn't explain the savant skills or large rote memory, or detail orientation based on grids of units.
3 comments
[ 2.9 ms ] story [ 18.8 ms ] threadMotivation for this paper:
Why there would be such a distinct cluster of symptoms for autism? Why would list-making and stiff facial expressions go together? Why should tunnel vision be related to overly formal speaking? Why should autism be anecdotally correlated with physicists and tech founders?
Then, I thought that Kolmogorov Complexity must be the missing connection. It is a formalization of Occam's Razor and would explain Albert Einstein, Isaac Newton, Nikola Tesla. But Kolmogorov Complexity relies on Turing Completeness, so those who would be good at finding shortest programs must also be good minimum viable product programmers, like Elon Musk, Mark Zuckerberg, Vitalik Buterin, and Bill Gates.
If the hypothesis is true, then some amount of autism is necessary for civilization to function, because you need both types of information processing (Shannon and Kolmogorov). The hypothesis is pending empirical verification, but if it is true, autism could be reframed as an alternate, and necessary way of thinking, not as a disorder. Additionally, interventions to help autistic people (and neurotypicals) could be implemented in a more rigorous way, using methods from information theory. The environment could be measured to determine how much Kolmogorov vs Shannon type processing is necessary, then interventions based on the needs of the individual could be addressed. There would be multiple ways to do this, 1. collaboration 2. training 3. accommodation.
collaboration between information types would be like neurotypicals working with autistic people for example at a company. The Hypothesis also implies that neurotypicals benefit from using traditional computers, and I think this is part of the reason why autistic people may have more trouble in some ways in the modern environment. However, it also implies autistic people could benefit from working with Neural Networks, such as LLMs and image generators. Another implication is LLMs like GPT-4 calling traditional code. Training: If the goal of someone is to become a savant, they can target certain skillsets that increase the ability to compute Kolmogorov Complexity. This may include rote memorization, learning algorithms, and improving focus, while avoiding logical paradoxes. If the goal of an autistic person is to become neurotypical, they may try to accumulate an ever increasing memory of social scripts and information about people, but not quite reach it. Accommodation: The environment could be changed to support both Kolmogorov and Shannon-type processing. There could be areas where routines and logic are allowed, with other areas supporting unpredictability. I think if this hypothesis is empirically validated, rigorous interventions could be implemented that improve the lives of autistic people using computer science concepts. What I personally would do (I'm on the spectrum) is try to partner with Neurotypicals and AI's, become a good programmer, and work on improving specific skills, while advocating for a autism by demonstrating it is necessary for robust information processing.
However, the predictive processing failure doesn't explain the savant skills or large rote memory, or detail orientation based on grids of units.