International Journal of Comparative Psychology

In two experiments, rats received sensory preconditioning treatment in which an auditory conditioned stimulus (CS) X was followed by visual CS A in Phase 1, and CS A was followed by an appetitive US (sucrose) in Phase 2. Rats also received presentations of auditory CS Y unpaired with other events. At test, rats looked for sucrose more following CS X than following CS Y on non-reinforced probe test trials only if the light bulb on which CS A had been presented during training was removed from the chamber at the time of testing. With the light bulb present (but unlit), rats showed nodifference in amount of nose poking between CS X and CS Y. These results suggest that rats distinguish between the explicit absence of events and lack of information

Five experiments investigated the roles of contingency and temporal contiguity in causal reasoning, and the trade-off between them. Participants observed an ongoing, continuous stream of events, which was not segmented into discrete learning trials. Four potential candidate causes competed for explanatory strength with respect to a single dichotomous effect. The effect was contingent on two of these causes, with one of these (A) having a higher probability of producing the effect compared to the other (B), while B was more contiguous to the effect than A. When asked to identify the strongest cause of the effect, participants consistently and reliably selected A, as long as it was not separated from the effect by more than 2.5 s. The extent of preference diminished, however, as the contiguity gradient between A and B increased. Beyond 2.5s, the high-probability, but low-contiguity cause A was seen as equally strong as the low-probability, but high-contiguity cause B, and both reliably stood out compared to the remaining two non-contingent distracter items. This apparent trade-offbetween contingency and contiguity, rooted in contrasting two of David Hume’s (1739/1888) fundamental cues to causality, has important implications for psychological and statistical models of causal discovery, learning theory, and artificial intelligence.

Considerable research has been devoted to investigate the type of information that subjects use to solve tool-using tasks in which they have to avoid certain obstacles (e.g., traps) to retrieve a reward. Much of the debate has centered on whether subjects simply use certain stimulus features (e.g., the position of the trap) or instead use more functionally-relevant information regarding the effect that certain features may have on a moving reward. We tested eight apes (that in a previous study had succeeded in a trap-tube task) with one functional and two nonfunctional traps to investigate the features that they used to solve the task. Four of the eight subjects used functional features. Additionally, we presented 31 apes with a trap task that did not involve tools but required subjects to make an inference about the position of a hidden reward based on its displacement over a substrate with or without a trap. Subjects performed above chance levels (including from the first trial) in the experimental condition (unlike in the control conditions), suggesting that they took into account the effect that a trap may have on a reward. Third, we correlated the subjects performance in four trap tasks (3 involving tool-use and one without tool-use) and found positive correlations between some of the tasks. Our results suggest that apes possess some knowledge about the effects that traps have on slow moving unsupported objects. However, this knowledge was not robust enough to prevent the influence of certain practice and task effects. Moreover, subjects’ knowledge may not have been abstract enough to allow them to establish broad analogies between tasks.

Theories of causal cognition describe how animals code cognitive primitives such as causal strength, directionality of relations, and other variables that allow inferences on the effect of interventions on causal links. We argue that these primitives and importantly causal generalization can be studied within an animal learning framework. Causal maps and other Bayesian approaches provide a normative framework for studying causal cognition, and associative theory provides algorithms for computing the acquisition of data-driven causal knowledge.