A vast body of data supports the notion that animals, including humans, perform statistically optimally in a wide range of tasks, supporting the claim that evolution has shaped the nervous system of organisms in a way that yields maximally adaptive behavior. Optimality is frequently assessed by comparing behavioral output to benchmarks computed via methods derived from statistical decision theory. Such methods have also been used to assess the reliability of sensory neural signals, and have even been invoked as accounts of neural processing.Scientists from the Biopsychology lab have explicitly tested the claim of optimality with pigeons performing a perceptual choice task with varying reward probabilities. Surprisingly, the pattern of choices observed was opposite to that expected under an optimization account. This finding poses important constraints on the class of algorithms useful for modeling adaptive choice behavior. Furthermore, the algorithm which fit the data best was learning exclusively on correct trials ending in reward, as opposed to algorithms learning from both reward and errors. Put simply, the pigeons seem to learn from reward, but not from errors - at least when performing a psychophysical task.

Stüttgen, M.C., Kasties, N., Lengersdorf, D., Starosta, S., Güntürkün, O., Jäkel, F. (2013). Suboptimal criterion setting in a perceptual choice task with asymmetric reinforcement. Behavioural Processes, 96, 59-70.