Skip to Content
  • Article
  • Open Access

19 November 2026

39 Pages

Fiction as Experiment, Part I: Generative AI and the Parametric Study of Social Cognition

,
and
1
Department of Neuroscience, Imaging and Clinical Sciences, University “G. d’Annunzio” of Chieti-Pescara, Via dei Vestini 31, 66100 Chieti, Italy
2
metaLAB (at) Harvard, Cambridge, MA 02138, USA
*
Author to whom correspondence should be addressed.
Big Data Cogn. Comput.2026, 10(11), 398;https://doi.org/10.3390/bdcc10110398 
(registering DOI)
This article belongs to the Section Cognitive System

Abstract

The simulation hypothesis holds that fiction works as a simulator for social cognition, with readers building generative models of characters’ minds. The correlational association has repeatedly been reported, whereas the experimental record is small and unstable, and the most demanding recent test recovered neither. We propose a so far overlooked source of that asymmetry: the simulation that traditional fiction affords is author-constrained, offering one trajectory through the social possibility space from vantage points that the reader cannot vary. Generative AI relaxes the constraint by regenerating the same events from other agents’ positions while varying beliefs, goals, evaluative frames, and observational access. We formalize the reader’s task within Bayesian Theory of Mind, define the divergence between perspectives over the forecasts that each supports, and specify five manipulations by the model component that each varies. The account predicts a single-peaked relation between divergence and cognitive gain whose optimum shifts with the reader’s capacity and with scaffolding, and it locates augmented reading closer to obtaining information about another mind than to simulating one, with consequences for what should be measured. This is the first of two companion papers: its untested predictions are conditional on generative capabilities and instruments whose validation the second specifies.

Article Metrics

Citations

Article Access Statistics

Multiple requests from the same IP address are counted as one view.