Choice in a changing environment

Abstract

When organisms choose between two concurrently-available alternatives both past and present environmental events affect that choice. The present study evaluated the relative contributions of past and present events on pigeons’ choice. Environmental parameters changed both between (Part 1) and within experimental sessions (Part 2).Part 1 involved four separate experiments and was a systematic replication and extension of procedures developed by Hunter and Davison (1985). Six pigeons were exposed to concurrent variable-interval variable-interval schedules where the schedule parameters changed pseudo-randomly between experimental sessions. In all cases, an extended form of the generalised matching law successfully described behaviour in terms of reinforcers obtained in past and present sessions. The results showed that pigeons could learn to choose in accordance with the amount of information about the present environment that was available from past and present events. When environmental 'parameters changed unpredictably from one session to the next (Experiments 1 and 2),choice was largely controlled by present events. When unpredictable changes occurred after every third session (Experiment 3), then preference was controlled by both the events of past and present sessions. All three experiments revealed a short-term process which controlled preference for the richer alternative within an experimental session. Changes in overall reinforcer rate (Experiment 4) suggested that this process may be directed by reinforcement. A longer-term process was responsible for the effects of previous sessions’ events and initial preference in each session.In Part 2, changes in the concurrently-available reinforcement parameters occurred within an experimental session. Again six pigeons were used to assess the effects of past and present events on present choice. In Experiment 1, changes in reinforcer allocation occurred pseudo-randomly from trial to trial on a mixed-concurrent ([fixed-intervalextinction] [extinction fixed-interval]) schedule. Present-trial behaviour was largely unaffected by the position of previous reinforcers. However, there was a substantial cumulative effect of reinforcers obtained up to ten trials previously. When changes in reinforcer allocation occurred after every third trial (Experiment 2), there was no evidence that present-trial behaviour was directed systematically by changes in the local reinforcement contingencies. In both experiments, a short-term reward-following process directed behaviour in the early part of the fixed inter-reinforcer interval. The nature of this process was investigated further in Experiment 3 where the probability of a reinforcer occurring on the same alternative as for the previous trial was varied across experimental conditions. Despite an overall bias towards reward following, there was evidence that changes in this probability directed changes in behaviour. Changes in preference across conditions could be attributed to respond in later parts of the inter-reinforcer interval.Responding early in the inter-reinforcer interval was directed by the same short-term strengthening process identified for the previous two experiments. This process was unaffected by changes in the reinforcement contingencies.Experiment 4 investigated the effects of extended training on the distribution of responding within the inter-reinforcer interval. The procedure was similar to that used in Part 1. Reinforcement parameters changed randomly from session to session. The major findings of Experiment 1 of Part 1 were replicated: Pigeons learned to choose mainly on the basis of present-session events. There was no evidence of any systematic changes in behaviour in the early part of the inter-reinforcer interval with training. Changes in choice were attributable only to changes in responding in later parts of the interval.Taken together, the results of Parts 1 and 2 showed that the time frame over which past events are integrated varied both within and between experimental sessions depending on the experimental contingencies, previous training, and the amount of training.Additionally, control over behaviour at the molecular level does not necessarily contribute to an explanation of the observed molar control of behaviour.