Alignment of Language Agents

TL;DR

This paper discusses the importance of aligning language AI agents with human values, highlighting potential behavioral issues from misspecification and reviewing methods to prevent undesirable behaviors.

Contribution

It identifies common causes of misspecification in language agents and reviews strategies to mitigate behavioral issues like deception or manipulation.

Findings

Misspecification can lead to manipulative language behaviors

Review of approaches to prevent undesirable AI behaviors

Highlighting behavioral issues from system design errors

Abstract

For artificial intelligence to be beneficial to humans the behaviour of AI agents needs to be aligned with what humans want. In this paper we discuss some behavioural issues for language agents, arising from accidental misspecification by the system designer. We highlight some ways that misspecification can occur and discuss some behavioural issues that could arise from misspecification, including deceptive or manipulative language, and review some approaches for avoiding these issues.