Friday, April 26, 2013

What Autism Tells Us About Artificial Intelligence

As I think more and more about the creation of artificial intelligence, I can see how the study of autism might have very important implications for the design of intelligent agents. My ideas of how to create an artificially intelligent being all revolve around a learning system that can model and systemize its environment. At one point though, I realized that what I was designing, after training and learning, would amount to an autistic agent. The agent may learn how to systemize its environment, but it wouldn’t have the social inclinations necessary to develop social skills, empathy and theory of mind. For this reason I think that “strong AI” necessitates a computer equivalent of the mammalian social modules. This means that AI researchers will need to acquaint themselves with concepts like oxytocin and vasopressin signaling, their effect on the nucleus accumbens, the endogenous opioid system, the HPA axis, the cingulate and orbitofrontal cortex and the way these are all affected by social encounters, social constraints and social expectations. Research into the neurological, cellular and molecular basis of mammalian social neuroscience may provide tremendous insight into how best to organize AI efforts such as pattern recognition, analytics, prediction, adaptive control, decision making, and response to query.


  1. Am following your blogs re autism, though I cannot follow all the science. Since listening to an npr report on the increasing number of births of autistics, I have wondered if this is an evolution for survival..for what, and why? Since I work with plants and insects, I am fascinated with nature's adaptability. Will continue following your work, and urge other interested persons to do so...thanks

  2. though there may be ways to implement them it in a more elegant and concise fashion using Lua's unique strengths.speech recognition software