Autistic people tend not to use metaphors. I have a possible model of this in terms of (naive) Bayesian mixture model classifiers, that is, normal people will forgive a large difference in one dimension if other dimensions match.
Ok, so... It seems to me the tendency will be for normal people to prefer sticking new things in existing categories (assimilation), whereas autistic people will be unable to ignore discrepancies and be forced to invent new classes (accommodation).
My suspicions about this are somewhat enhanced by the fact that one of the principal revolutionaries in "Wonderful Life" is reported to like sitting in the library wearing a cloak.
Ok, so... One way to test this would be to look at the authors pre-revolutionary and revolutionary papers on some scientific subject, and see if the revolutionaries are more autistic.
Prediction: the sentence lengths in revolutionary (accommodative) papers will have sentence lengths having a higher characteristic exponent than those in assimilative papers. The fields of paleontology and taxonomy would be good places to look for such papers.
[ Note that science works best in those fields in which this is the best approach -- in which a few odd datums truly do falsify an existing theory. In the "soft" sciences, such as psychology, one would expect a scientific approach to initially fail to account for some really obvious phenomena, then later perhaps produce predictively accurate but overly complex theories, with more classes than are necessary. ]