Autistic vs normal search metric


Further thoughts on cognition as searching.

Suppose we have a situation, and want to retrieve similar situations that have occurred in the past in order to predict what will happen in the current situation.

Some events in the past are more relevant than others, depending on how similar they are to the current situation. We want the most relevant events, the "best-fit" to the current situation (very strong analogy here to best-fit texture synthesis).

We identify a set of characteristics by which the current situation may be compared to past situations, and assign each characteristic a weight. We then calculate for each past situation

  relevance = weight_0 * discrepancy_0 + weight_1 * discrepancy_1 + ...

I hypothesize that normal people calculate each discrepancy as the squared difference between the current and past value of each characteristic (or thereabouts), whereas people in the autistic spectrum calculate the discrepancy as the absolute difference between values (or thereabouts). Normal people use L2, autisitic people use L1... actually most people are probably somewhere in between... L1.9, L1.2, L1.9, whatever, but with people very close to L2 being disgustingly normal and people very close to L1 highly autistic.

This hypothesis is testable by looking at what analogies people make in abstract problems. For example, create a set of geometric figures with features of varying sizes, lengths, and angles (ie real-valued characteristics), ask people to group them into sets, calculate the weights and norm their grouping is based on (a la k-means).

AD/HD may involve a smaller selection of characteristics than normal -- a smaller number of non-trivial weights... yielding an oblate set of results, a sub-space sharp as a knife ... still thinking about this. Should be orthogonal to autism, as they can co-occur.