Repeated reversal of training in Octopus

Octopuses were trained by successive presentations to discriminate between two figures by means of a combination of food rewards and electric shocks. This was followed by a series of up to nine daily sets of trials in which reward and punishment were systematically reversed on alternate days. It was found that performance became progressively less accurate with successive reversals until it reached a random level; this appeared largely due to a continual fall in the total number of attacks at the figures. At the same time, such attacks as were made were mainly at the positive figures and there was evidence that performance as judged by this criterion became more accurate during each day and perhaps progressively more accurate on successive days. Octopuses lacking the vertical lobe showed a fall in the total number of attacks but did not show this improvement within each day or over successive days. A theoretical interpretation of the results is put forward.