Rational process models help to connect computational and algorithmic levels of analysis: the principle of rationality at the computational level motivates algorithms that approximate probabilistic inference at the algorithmic and implementational levels. From @2024griffithsBayesian:
Rational process models take a different approach from traditional strategies for making computational models in cognitive psychology, which start with a hypothesized set of psychological mechanisms and examine how those mechanisms can be combined to model behavior. In a rational process model, we begin with an algorithm for approximating probabilistic inference, ask whether the components of the algorithm are consistent with what we know about cognitive processes, and then examine how well the model fits behavior. The result is a class of models that are guaranteed to approximate probabilistic inference but deviate from ideal solutions in ways that can be instructive about the processes that underlie human judgments. Indeed, the hope is that the ways in which a rational process model deviates from perfect rationality will turn out to be the very ways in which human behavior deviates from the ideal rational solution. (287)
Rational process models assume that the mind approximates optimal inference and decision-making given constraints on time and computation. Here, rationality refers to the assumption that the algorithm used in cognition will converge to the optimal inference or decision as amounts of time and computation increase.
In contrast to rational process models, resource-rational analysis aims to uncover the single most effective approximation strategy, given constraints on time and computation.