What do we actually want to experience? A computational metric for assessing reward values

TL;DR

This paper introduces a new method to measure how rewarding people find experiences without relying on self-reporting, showing it better predicts behavior than explicit ratings.

Contribution

A novel computational metric for inferring reward values of experiences using reinforcement learning and modeling.

Findings

Participants showed higher cRV for positive images and more attractive faces.

cRVs were sensitive to context and individual differences.

cRVs predicted behavior better than explicit self-reported values.

Abstract

People’s motivation to have different experiences is predicated on how much they find those experiences rewarding or not, and these reward values are not always fully accessible to our consciousness. In two studies, we demonstrate that using a combination of reinforcement learning (RL) paradigms and computational modeling, we can measure computationally inferred reward values (cRV) of experiences, which do not rely on conscious self-report. Consistent with motivational reward theory, convenience samples of participants exhibited higher cRV (greater reward value of that experience) to viewing positive vs. negative images (subject pool; Study 1) and to viewing more vs. less attractive faces (online sample; Study 2). Further, these cRVs were sensitive to context (familiarity vs. novelty of images, Study 1) and to individual differences (attraction preference, Study 2). Lastly, although cRVs were mildly correlated with explicit self-report values, which demonstrates their validity, they were better predictors of behavior than were the explicit values, which suggests that cRVs are capturing reward processes that are not represented by explicit value judgments. This method of measuring cRV holds great promise for understanding the motivation driving people’s choices of a variety of experiences across a wide array of fields of study.

Figures