UCLA

The ability to learn and effectively manipulate causal structures in the world is an important aspect of cognition that allows us to interact with the environment in meaningful and often biologically significant ways. While many have considered such a key cognitive faculty to be one that divides humans from animals, recent evidence has suggested that rats are capable of learning causal structures and making correct inferences of these structures in a way that is effective (e.g., to obtain a desired outcome), whether from passive observation of the causal structure, or intervention on an event within the causal structure (Blaisdell, Sawa, Leising, & Waldmann, 2006). Furthermore, this ability appears to be more than merely an approximation via basic associative processes. However, the nature of causal interventions is not well understood. The observation that rats that intervene to evoke an outcome of a common cause do not also expect the other co-outcome to occur (i.e., the attribution of the outcome to their own action as opposed to the occurrence of the common cause, referred to in this dissertation as the causal intervention effect) suggests that causal interventions may require sensitivity to action-outcome contingencies. If so, brain areas that are known to govern goal-directed action (i.e., the dorsomedial striatum) should be necessary for causal interventions. After determining experimental perimeters required to obtain some basic effects (namely, the causal intervention effect and goal-directed/habitual instrumental behavior), an experiment was done wherein rats were either pre-trained to lever-press in a manner that was either outcome-sensitive (i.e., goaldirected) or outcome-insensitive (i.e., habitual) prior to training on a common cause model (i.e., Tone <-- Light --> Food). During test, rats either intervened on the lever to produce, or merely observed Tone. If causal interventions require outcome sensitivity of actions, then only rats that behaved on the lever in a goal-directed manner (i.e., outcome-guided) should have shown the causal intervention effect and thus discounted Food; whereas rats that behaved habitually (i.e., outcome-insensitive) on the lever, as well as all subjects that merely observed the Tone should have attributed it to the common cause (i.e., Light) and thus expected Food. This study was followed up with another involving a direct neural manipulation, wherein subjects were given either a lesion of the posterior dorsomedial striatum (pDMS) or a sham lesion before being trained on the common cause model, and subsequently tested as before. Since the pDMS has been shown to be necessary for the acquisition and expression of action-outcome associations in instrumental conditioning (e.g., Yin, Ostlund, Knowlton, Balleine, 2005), pDMS-lesioned rats that intervened on (as well as all rats that merely observed) the Tone should not have discounted Food. Conversely, sham-lesioned rats that intervened on the Tone should have discounted Food, thus showing the causal intervention effect. Unfortunately, neither predictions were supported by the data from both experiments. Future avenues of research about causal interventions, inferences, and reasoning are discussed.