"So," you ask, "you do 'research'. What the heck does that mean?"

At Yale, I do research on visual learning. Subjects sit down in front of a computer and are shown displays like the one to the right. They have to look through these displays until they find the sideways 'T', and then press either the left or right arrow key, depending on which way the bottom of the T is pointed. This ensure that they actually found the T. The experiment is grouped into 20 "blocks," each of which contains 32 of these displays where subjects must find the T.

What's special about this experiment is that half of the displays from the first block (16 of them) are repeated again in every single block afterwards. The other half in each block are new, random, displays. What we find is that although subjects in the experiment are unaware of the displays being repeated, and, when asked, cannot tell you which displays repeated and which didn't, they are actually learning these displays implicitly. They are significantly faster on the displays they have seen before compared to the ones that are new to them, even if the target is in the same location in both. They have learned that a particular configuration of Ls predicts where the T is going to be, and use this information to find the T more quickly. We call this effect contextual cueing -- the context you find something in is cueing you to its identity and/or location.

Interestingly, we don't need to repeat the entire configuration of Ls in order for subjects to find the T more quickly. Simply repeating the location of a group of Ls near the T and changing the rest to different locations still allows subjects to find the T more quickly. In fact, people find the T just as quickly when we repeat only the two items in the same 'quadrant' of the display as the target instead of repeating the entire display. It seems as though people are only learning the local context around the target in order to find the target more quickly!

We discovered this fact by using a model of contextual cueing that I created. This computer model attempts to act just like a person, learning the displays and guessing where the target is in each of them. What we discovered when we were making this model was that to account for some previous data about how people do contextual cueing, we needed to limit how far away from the target the model could learn. This model then predicted that people were probably only learning the locations of the Ls that were very near the target -- and when we tested this, we found it was true.