Today's post is by David Thorstad on his recent paper “Why bounded rationality (in epistemology)?” published in 2023 in Philosophy and Phenomenological Research.
 |
| David Thorstad |
Bounded rationality gets a bad rap in epistemology. Critics argue that theories of bounded rationality are too context-sensitive, conventionalist, or reliant on ordinary language. My aim in this paper is to make sure that bounded rationality gets the rap sheet it deserves.
The first order of business is to lay out an approach to bounded rationality inspired by traditional theories of bounded rationality in cognitive science. I explain and defend this theory more fully in my book, Inquiry under Bounds, under contract with Oxford University Press. My approach has five core commitments.
First, bounds matter. Paradigmatic bounds such as limited cognitive abilities and the cost of computation bear on how it is rational for us to cognize.
Second, rationality is procedural, moving upwards from the lower-order question of what an agent should believe to the higher-order question of which processes of inquiry she should use to form and revise beliefs.
Third, it is often rational for agents to form judgments using a particular kind of cognitive process: fast-and-frugal heuristics.
Fourth, rational is ecological, or environment-relative. Because all heuristics perform well in some environments and badly in others, we cannot ask whether a heuristic is rational or irrational full-stop. We must always ask instead: in which environments would this process be rational to use?
Finally, the right approach should ground a program of vindicatory epistemology which aims to show how many seeming irrationalities are in fact instances of boundedly rational cognition.
The next order of business is to show how this approach answers some recent criticisms of bounded rationality by Jennifer Carr (2022). Carr argues that theories of bounded rationality are unacceptably conventionalist, relying on arbitrary epistemic conventions formed by communities of inquirers. But my approach is not conventionalist in any way.
Car also argues that theories of bounded rationality are unable to specify which bounds matter to rational cognition. Focusing on internal cognitive bounds, Carr offers several examples of bounds that matter and bounds that do not, but argues that no plausible theory of bounded rationality can distinguish them. I propose a traditional Bayesian understanding on which the bounds that matter are those determined by an agent’s fixed cognitive architecture. I argue that this approach correctly sorts the bounds that matter from those that don’t on Carr’s lists.
An enduring question about bounded rationality is whether and to what extent traditional approaches to bounded rationality are compatible with Bayesian theorizing. Recent work has suggested that bounded rationality may be not only compatible with, but essential to Bayesian theorizing: careful attention to cognitive bounds has led to innovations such as decisionmaking by sampling, probability heuristics, and resource-rational analysis that have improved the descriptive and normative plausibility of Bayesian theories.
On this basis, I argue, bounded rationality should be viewed as a welcome addition to existing theories of Bayesian epistemology.