| Paper: | GS-1.1 |
| Session: | Contributed Talks I |
| Location: | Ormandy |
| Session Time: | Thursday, September 6, 11:10 - 12:00 |
| Presentation Time: | Thursday, September 6, 11:10 - 11:35 |
| Presentation: |
Oral
|
| Publication: |
2018 Conference on Cognitive Computational Neuroscience, 5-8 September 2018, Philadelphia, Pennsylvania |
| Paper Title: |
A Generative Model of People's Intuitive Theory of Emotions |
| Manuscript: |
Click here to view manuscript |
| DOI: |
https://doi.org/10.32470/CCN.2018.1128-0 |
| Authors: |
Sean Dae Houlihan, Max Kleiman-Weiner, Joshua Tenenbaum, Rebecca Saxe, Massachusetts Institute of Technology, United States |
| Abstract: |
We present a formal model of emotion predictions that effectively captures observers' nuanced intuitions about the emotional experiences of players in a strategic and socially charged situation---a high-stakes public one-shot prisoner's dilemma. We first incorporate social inequity concerns, reputational considerations, and monetary utility in an inverse planning framework to model observers' intuitions about the latent motivational structure of gameplay. Next, in a novel approach to inverse planning models, simulated agents react to the game's outcomes, generating prediction errors. These reactions, achieved utilities, and counterfactuals are then translated into forward predictions about players' emotions. The emotion predictions generated by the model reflect the richly structured reasoning that observers exhibit when considering players' experiences, including the counterfactual dependencies of Relief and Regret, the social cognitive dependence of Envy, the prosocial dimension of Joy, and the moral content of Guilt and Embarrassment. The model captures these intuitions using a psychologically plausible architecture that resembles observers' direct judgments of the latent parameters |
