Some history

We started the autism project a couple of years ago, when my advisor/collaborator Israel Gelfand and myself meet Alan Leslie at his presentation of the Theory of Mind. I got the idea that the a computer (logical) program can behave exactly the same way (in terms of question answering) as autistic/control children in the experiments, involving such mental states as pretending, believing, observing, etc.

The project started with the questions, what kind of axioms would a human need to reason properly about the mental states of his/her own or the others. I involved some results from the

"reasoning about knowledge", relatively new area in the logical AI. The problem was, the multiple calculi of intention, knowledge and belief are proven to be independent, but just one of them (or some other) could be possibly implemented in the brain. So the machinery of math logic is not very helpful.

I built the axiomatic system based on the concepts of intention, knowledge and belief and computationally showed that any complex mental state or action can be expressed (approximated) in this basis. I tested this axiomatic system in various domains such as natural language understanding, multi-agent constraint satisfaction, modeling of the mental states of investors in finance, etc.

Then I built the program, which represents the question-answering of the autistic/control children in accordance to recent experimental studies. Psychologists focused on the particular mental phenomena, but I was interested to show that the autism boundary separates reasoning about mental states from reasoning about the physical ones. It lead us to planning the experiments, involving all the possible mental states of the given complexity, and checking out what is the scale (granularity) that the autistic brain performs/mistakes.

I have the data of 20 autistic children 4-18 years old as the evidence that the brain does operate on the level of mental axioms: for each child, every axiom is either working or not working. The "proof" comes from the fact, that both normal and autistic brains are too complex for modeling, but the "difference" is represented as a set of about 20 mental axioms.

Another result of these experiments is the training methodology; some children who passed through it a year ago sometimes build associations between mental formulas and people in real life.

back to autism home