Contextual Moral Value Alignment Through Context-Based Aggregation

TL;DR

This paper introduces a system for aligning large language models with multiple moral values by contextually aggregating responses based on user input features, improving alignment with human values.

Contribution

It proposes a novel context-based aggregation method for moral value alignment in LLMs, enhancing multi-value adaptability and alignment accuracy.

Findings

Outperforms existing methods in moral value alignment.

Effective in integrating responses aligned with diverse moral values.

Improves adaptability of LLMs to different moral contexts.

Abstract

Developing value-aligned AI agents is a complex undertaking and an ongoing challenge in the field of AI. Specifically within the domain of Large Language Models (LLMs), the capability to consolidate multiple independently trained dialogue agents, each aligned with a distinct moral value, into a unified system that can adapt to and be aligned with multiple moral values is of paramount importance. In this paper, we propose a system that does contextual moral value alignment based on contextual aggregation. Here, aggregation is defined as the process of integrating a subset of LLM responses that are best suited to respond to a user input, taking into account features extracted from the user's input. The proposed system shows better results in term of alignment to human value compared to the state of the art.