Scaling Unstructured Category Learning

Scientists, like all humans, are attracted to structure. Scientists studying category learning have spent most of their time and attention on structured categories. The structure underling the categories could be either rule-based or feature relationship-based. However, there are many categories that are unstructured. One example of an unstructured category is "my passwords." Hopefully, there is more than one exemplar in that category, and none of them are "password." "My passwords" are most likely alphanumeric strings, but there is no underlying rule or feature relationship uniting members of the group.

Very little is empirically known about unstructured category learning. One aspect I have recently investigated is the finite human capacity to learn unstructured categories. I was manipulating the number of exemplars per category to validate a particular set of stimuli. For this set of stimuli as I increased the the number of exemplars per category, performance could be characterized as "very easy," "easy," "kinda of hard," and "impossible." I'm not unclear what underlies this particular anecdotal result. It would better serve the field to better understand this empirical phenomenon and other unstructured category learning properties before assuming neurobiological mechanisms.