Hanbin Go just recently had his master’s thesis accepted. Congratulations Hanbin!
You can read it here. His work (jointly supervised with James Danckert) dealt with the ideas of mental models, the informal term for the sense we all have of how our world is built: what are the contingenices? What are the odds that one thing happening leads to another? Although this commonsense characterization seems plausible, how do you develop experimental tests of the idea that will also flush out the underlying structure? Hanbin chose eye movements and inspired by prior work constructed a clever protocol where he looked to see if eye movements could be used to probe for expectation, surprise, and the need to update a model.
In short, saccades are faster to expected locations and low probability alone did not equate to surprise. What best indexed participants updating their representations was the time participants looked at a target (aka dwell time). They were not very good at picking up the distributional changes we introduced (that is they did not seem to be Bayesians), but when they told us when they suspected we had made a change (even though we usually had not) we could see a lengthening in time spent fixating the target prior to those reports.