Abstract
We investigated how human subjects plan a sequence of rapid movements in a foraging-like task and compared performance to an optimal model based on maximum expected gain. On each trial, three targets (disks) carrying monetary rewards coded by color were presented for 1600ms in fixed locations on a computer screen. The subject had to choose which targets to attempt to hit and what sequence to attempt them in. The key challenges for the subject were, what sequence to choose and whether to try for all three targets. The key question for us was, do subjects select the sequence that maximizes their expected gain?
To compare subjects' performance to optimal, we have to first measure how accurately they can execute each possible sequence of length 2 or 3 on three targets. The experiment consisted of two sessions. In the first session, four naïve subjects were trained until their performance was stable. We then measured each subject's performance for all 12 possible sequences of lengths 2 or 3. We refer to the targets as A,B,C. The dependent variable was the probability of hitting each target (pA, pB, pC) and is a function of sequence. It might be (.9,.9,.3) for sequence ABC but (.9,.1,.6) for sequence ACB. The sequence ABC offers the higher probability of hitting all three targets but, if C is made valuable enough, the sequence ACB has higher expected gain. In the second session, the values of the targets were altered by amounts that varied from trial to trial. Subjects were told the values of the color-coded targets before each trial and were free to choose any sequence they wished. The values in the second session were chosen so that an ideal mover maximizing expected gain would change sequence for some value combinations. Two out of four subjects varied their movement sequences so as to maximize expected gain. The other two subjects exhibited highly consistent strategies across different reward values but did not maximize expected gain.
Grant EY08266 from the National Institute of Health