The Capacity for Moral Self-Correction in Large Language Models

Abstract

We test the hypothesis that language models trained with reinforcement learning from human feedback (RLHF) have the capability to "morally self-correct" -- to avoid producing harmful outputs -- if instructed to do so. We find strong evidence in support of this hypothesis across three different experiments, each of which reveal different facets of moral self-correction. We find that the capability for moral self-correction emerges at 22B model parameters, and typically improves with increasing model size and RLHF training. We believe that at this level of scale, language models obtain two capabilities that they can use for moral self-correction: (1) they can follow instructions and (2) they can learn complex normative concepts of harm like stereotyping, bias, and discrimination. As such, they can follow instructions to avoid certain kinds of morally harmful outputs. We believe our results are cause for cautious optimism regarding the ability to train language models to abide by ethical principles.

Cited by in corpus (6)

• Generative AI
• Potential Benefits of Employing Large Language Models in Research in Moral Education and Development
• Contextual Agent Security: A Policy for Every Purpose
• Should AI Mimic People? Understanding AI-Supported Writing Technology Among Black Users
• A challenge in A(G)I, cybernetics revived in the Ouroboros Model as one algorithm for all thinking
• Meta-ethics and AI: exploring the novel meta-ethical questions in the era of AI