Decision Making as a Dynamic Process in Semantic Space
Formal theories of choice typically assume that value attaches to outcomes, so that deliberation amounts to computing, learning, or sampling the consequences of actions. Yet human deliberation is dominated by reasons, justifications, identities, and moral rules: rich semantic representations that derive their motivational force from their meanings and associations rather than from outcomes in the world. I present Value Semantics, a framework in which value is defined as a function over a high-dimensional semantic space and deliberation is a stochastic traversal of that space. At each moment, a representation is retrieved with probability that depends on the options, the context, previously retrieved representations, and the current state of preference. Retrieved representations are bound to the options they concern, and each option’s value accumulates as they are sampled. Preference thus emerges as a dynamic, path-dependent quantity, shaped by the trajectory deliberation takes through semantic space. The framework accounts for many phenomena that are anomalous or difficult to explain within outcome-centered theories, and it recovers existing frameworks, including sequential sampling models, as special cases. Because representations can be embedded with language models, the framework can be fit to think-aloud protocols, allowing the dynamics of deliberation to be measured directly.
There is nothing here yet. Be the first to create a thread.
Cite this as: