Association for Behavior Analysis International

Rapid adaptation to changes in the environment aids in the continued acquisition of resources such as food, and eliminates adverse consequences resulting from adherence to strategies that are no longer effective. Such behavioral flexibility is important for survival, and is also thought to be a property of high-level processes that drive rule generation and behavioral selection. When environmental changes occur repeatedly and are associated with multiple cues, animals can often behave in sub-optimal ways by failing to attend to the most relevant cues. Recent research shows that some animals make surprising anticipatory and perseverative errors in a reversal task that provides multiple cues. These errors suggest that behavior is strongly controlled by elapsed time, even though time is a less reliable cue than other concurrently-available stimuli. The present symposium examines possible reasons for reliance on time as a cue for a likely change in contingency, with data from human, pigeon, and chickadee subjects. In particular, the symposium addresses the involvement of response inhibition, category formation, sequential decision-making mechanisms, and contingency discriminability in behavioral flexibility. The symposium considers both adaptation to repeated contingency reversals, and accuracy following prolonged exposure.

It has been shown previously that some animals make surprising anticipatory errors on simultaneous choice tasks in which contingencies of reinforcement reverse midway through each session. We trained pigeons (Columba livia), black-capped chickadees (Poecile atricapillus), and human undergraduates with comparable go/no-go variants of the midsession reversal procedure. Pigeons demonstrated independent errors on the two stimuli, suggesting that incorrect responses were made based on inhibition failures rather than deliberate choice; chickadees showed poor discrimination that was replicated in pigeons on a novel version of the task; and humans generally showed good discrimination, though with errors in the same direction as other animals. Animals flexibility on midsession reversal is affected by anticipation based on interval time, but responding is dependent on the predictability of reversals and can be subverted by response inhibition.

Recent research on midsession reversal learning has revealed that pigeons make systematic anticipatory and perseverative errors around the reversal location, suggesting they are using the time within sessions as a cue for which of two stimuli is likely correct on a given trial. The repetition of the same stimuli across trials, however, means their values are constantly changing and relative to the pigeon's perception of its location within the session. The current experiment tested midsession reversal learning of categories (flowers (S1) and cars (S2)), where every trial presented a unique stimulus pair from each category and, across sessions, these pairs were presented on the same trial (e.g., Trial 1: flower1, car1), allowing us to create stimuli with absolute values of reinforcement. Interestingly, pigeons showed similar errors around the reversal as has been previously found with two stimuli, suggesting that category learning emerges prior to item specific encoding and pigeons anticipate the category shift based on time. Probe tests revealed a strong reliance on the time-based cue, with some evidence that item-specific information is important closer to the reversal. These results have implications for the timing and order by which pigeons learn to use categorical and item-specific features in pictorial discriminations.

The Midsession reversal task involves a simultaneous choice between two stimuli, one of which (S1) is correct for Trials 1-40 and the other (S2) for Trials 41-80. Pigeons have been shown to both prematurely respond to S2 and perseveratively respond to S1, indicating the perceived valence of these stimuli dynamically change within a session. We hypothesized this reflects a decision mechanism adapted to sequential rather than simultaneous encounters in which subjects adjust the latency to respond to the estimated value of each stimulus. The Sequential Choice Model, which has been tested on different protocols and species, follows this logic and proposes that the currently preferred stimulus in simultaneous choice is the option that produced the shorter latency on single option trials. Two predictions of this model are that relative latencies on single option trials should predict the proportion of choices on simultaneous encounters and that choices should not take any extra evaluation time. Using pigeons, we report that, in a midsession reversal task with both single stimulus and simultaneous choice trials, both predictions were confirmed. Further, on single option trials, latencies to S2 decreased as the midsession approached while latencies to S1 increased as failures to obtain reward accumulated.

Discriminability of the relation between stimuli and the availability of reinforcers appears important in behavioral flexibility. Where stimuli associated with contingency reversals are less accurately discriminated, the relation between these stimuli and the correct response cannot be perfectly discriminated. We examined the speed and extent to which pigeons choice came under the control of repeatedly-reversing contingencies when the contingency reversal was accompanied by discrete or continuous change in stimuli. In a modified version of the serial-reversal task, the response more likely to produce a reinforcer changed repeatedly within a single session, according to elapsed time, the number of events, or the color of the keylights. Transitions between conditions involved a change in the local reinforcer differential, or in the point at which contingencies reversed. Discriminability of the stimulus that signaled the response-reinforcer contingency was manipulated in two ways: By changing the time at which the schedule reversal occurred; and in some conditions, by preceding or accompanying the schedule reversal by a keylight color change. Choice reached stable values after fewer sessions, and anticipatory and perseverative errors were reduced, under conditions in which the stimuli signaling the contingency change were more discriminable. Thus, environmental conditions that enhance discriminability of the stimulus-reinforcer relation facilitate behavioral flexibility.