Learning with delayed rewards in Octopus

Abstract

1. Octopuses can be trained to discriminate between white and black discs given at successive trials when the food rewards and shocks are delayed for as much as 30″. There are slight signs of learning even with delays of 60″. 2. The amount of information acquired from each attack (at white or black) decreased with delay in reward. 3. The tendency to attack the positive figure declined with delay in reward, especially beyond 30″. 4. If rewards were then given without delay attacks increased, showing that there is a positive process of learning to attack. 5. Octopuses also show some memory storage allowing distinct reactions by touch to rough and smooth spheres with delay of up to 30″ in reward. 6. The amount of information gained from each occasion of taking a sphere decreased progressively with delay. 7. There is evidence from both visual and tactile fields that discrimination learning under these conditions of successive presentation involves the formation of distinct representations ensuring take (attack) with one object and rejection (retreat) with another. The two may show different rates of information storage, which do not alter in the same way when changes are made in the delay before reward or punishment. 8. With longer delays (60″ and 120″) some individuals made “discriminant” scores, but there are reasons for believing that this can be attributed to extraneous circumstances, including alternate presentation, and a fall in the probability of taking within sessions. 9. After removal of the vertical lobe octopuses were unable with rewards delayed 15″ to learn a black-white discrimination against the preference. 10. Animals without vertical lobes trained with tactile discriminations showed less than normal capacity to learn with 10″ delay. With longer delays the individuals were characteristically variable. Some quickly came to take the positive sphere on nearly all occasions, others performed randomly, taking both spheres very often. 11. Two characteristics of animals without vertical lobes are thus to swing to extreme preferences and to be unable to learn not to take objects that yield shocks. 12. Animals with the supraoesophageal lobes bisected learn rather less well than normals when rewards are delayed, but do not show the aberrations characteristic of those without vertical lobes.