In this paper we review different conceptions of rationality as they apply to decision making and problem solving. We defend the notion of enactive problem solving (EPS) and we consider how it applies to the use of therapeutic chatbots. We start by recounting the development of different models of rationality, from expected value maximization, to subjective expected utility theory, and of bounded rationality from prospect theory, to simple adaptive heuristics. We then focus on problem solving. We show how the centre of gravity of problem solving shifts from the model of computational and cognitive processing, to an action-oriented pragmatic one in EPS, that is, to the possible actions that the body-environment interaction allows. We highlight a notion of the agent’s relational autonomy as part of EPS and show that this can be a problem in some chatbot designs in the therapeutic context. EPS suggests that simple early versions of therapeutic chatbots like ELIZA may be more beneficial than more recent sophisticated models involving “Belief-Desire-Intention” (BDI) architecture characterized by predetermined trajectories. The system formed between the agent plus the chatbot tool facilitates a form of enactive problem solving, reducing rigidity on the side of the agent.
Viale et al. (Thu,) studied this question.