Truth Machines: Synthesizing Veracity in AI Language Models

TL;DR

This paper examines how large language models like InstructGPT and ChatGPT synthesize and operationalize different notions of truth, highlighting challenges and proposing directions for improving AI's veracity through social and contextual enhancements.

Contribution

It analyzes the mechanisms behind truth production in language models and introduces the concept of truth as a social practice, suggesting ways to improve AI truthfulness.

Findings

InstructGPT synthesizes conflicting claims into truth-statements.

ChatGPT exhibits similar truth operationalization issues.

Enriching sociality may enhance AI's truth-evaluating capacities.

Abstract

As AI technologies are rolled out into healthcare, academia, human resources, law, and a multitude of other domains, they become de-facto arbiters of truth. But truth is highly contested, with many different definitions and approaches. This article discusses the struggle for truth in AI systems and the general responses to date. It then investigates the production of truth in InstructGPT, a large language model, highlighting how data harvesting, model architectures, and social feedback mechanisms weave together disparate understandings of veracity. It conceptualizes this performance as an operationalization of truth, where distinct, often conflicting claims are smoothly synthesized and confidently presented into truth-statements. We argue that these same logics and inconsistencies play out in Instruct's successor, ChatGPT, reiterating truth as a non-trivial problem. We suggest that enriching sociality and thickening "reality" are two promising vectors for enhancing the truth-evaluating capacities of future language models. We conclude, however, by stepping back to consider AI truth-telling as a social practice: what kind of "truth" do we as listeners desire?