Value Internalization: Learning and Generalizing from Social Reward

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Rong, Frieda, Kleiman-Weiner, Max
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866914879593512960
author Rong, Frieda
Kleiman-Weiner, Max
author_facet Rong, Frieda
Kleiman-Weiner, Max
contents Social rewards shape human behavior. During development, a caregiver guides a learner's behavior towards culturally aligned goals and values. How do these behaviors persist and generalize when the caregiver is no longer present, and the learner must continue autonomously? Here, we propose a model of value internalization where social feedback trains an internal social reward (ISR) model that generates internal rewards when social rewards are unavailable. Through empirical simulations, we show that an ISR model prevents agents from unlearning socialized behaviors and enables generalization in out-of-distribution tasks. We characterize the implications of incomplete internalization, akin to "reward hacking" on the ISR. Additionally, we show that our model internalizes prosocial behavior in a multi-agent environment. Our work provides a foundation for understanding how humans acquire and generalize values and offers insights for aligning AI with human values.
format Preprint
id arxiv_https___arxiv_org_abs_2407_14681
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Value Internalization: Learning and Generalizing from Social Reward
Rong, Frieda
Kleiman-Weiner, Max
Machine Learning
Artificial Intelligence
Multiagent Systems
Social rewards shape human behavior. During development, a caregiver guides a learner's behavior towards culturally aligned goals and values. How do these behaviors persist and generalize when the caregiver is no longer present, and the learner must continue autonomously? Here, we propose a model of value internalization where social feedback trains an internal social reward (ISR) model that generates internal rewards when social rewards are unavailable. Through empirical simulations, we show that an ISR model prevents agents from unlearning socialized behaviors and enables generalization in out-of-distribution tasks. We characterize the implications of incomplete internalization, akin to "reward hacking" on the ISR. Additionally, we show that our model internalizes prosocial behavior in a multi-agent environment. Our work provides a foundation for understanding how humans acquire and generalize values and offers insights for aligning AI with human values.
title Value Internalization: Learning and Generalizing from Social Reward
topic Machine Learning
Artificial Intelligence
Multiagent Systems
url https://arxiv.org/abs/2407.14681