Learning Theory

Learning theory is one of several consequentialist modes of explanation in the social sciences, along with functionalism, expected utility, game theory, and conflict theory. In consequentialist explanations, actions are explained in terms of the outcomes they produce. An obvious problem is [Page 437]that the explanatory logic runs in the opposite direction from the temporal ordering of events. Actions are the explanandum and their outcomes the explanans. This explanatory strategy collapses into teleology unless mechanisms can be identified that bridge the temporal gap. While expected utility theory and game theory posit a forward-looking and analytic causal mechanism, learning theory provides a backward-looking and experiential link.

In forward-looking rationality, the link from actions to their explanatory consequences is the analytical ability of purposive actors to reliably predict the outcomes of alternative choices. With a perfect grasp of the logical or mathematical structure of a well-defined problem and complete information about inputs to the model, the likely consequences of alternative courses of action can be known before the fact. The ideal type is “the neoclassical economic model in which rational agents operating under powerful assumptions about the availability of information and the capability of optimizing can achieve an efficient reallocation of resources among themselves through costless trading” (Axelrod 1997:4). The consequences that matter are not the actual ones (which have not yet occurred), but those that are predicted. Outcomes that arise behind the backs of the actors, such as the unintended collective benefits of the “Invisible Hand,” cannot attract the choices that produce them, an insight made famous by Adam Smith.

Forward-looking calculation is mainly applicable to skilled entrepreneurs, political strategists, military leaders, or game theorists. In everyday life, decisions are often highly routine, with little conscious deliberation. These routines can take the form of social norms, protocols, habits, traditions, and rituals. Learning theory explains how these routines emerge, proliferate, and change in the course of consequential social interaction, based on experience instead of calculation. In these models, repetition, not prediction, brings the future to bear on the present, by recycling the lessons of the past. Through repeated exposure to a recurrent problem, the consequences of alternative courses of action can be iteratively explored, by the individual actor (reinforcement learning) or by a population (evolutionary learning). Individual learning alters the probability distribution of routines competing for an actor's attention. Population learning alters the frequency distribution of routines carried by individuals competing for survival, reproduction, or social influence. Reinforcement learning is not limited to human actors but may be applied to larger entities such as firms or organizations that adapt their behavior in response to environmental feedback. And evolutionary learning is not limited to genetic propagation. In cultural evolution, norms, customs, conventions, and rituals propagate via role modeling, occupational training, social influence, imitation, and vicarious learning.

Whether the process is individual-level reinforcement, genetic propagation, or cultural evolution, the underlying learning principle is the same: adaptation to environmental feedback. Positive outcomes increase the probability that the associated routine will be repeated or reproduced, while negative outcomes reduce it. For example, a firm's problemsolving strategies improve over time through exposure to recurrent choices, under the relentless selection pressure of market competition. Suboptimal routines are removed from the repertoires of actors by learning and imitation, and any residuals are removed from the population by bankruptcy and takeover. The outcomes may not be optimal, but we are often left with well-crafted routines that make their bearers look much smarter than they really are (or need to be), like a veteran outfielder who catches a fly ball as if he or she had calculated its trajectory.

...