Beyond Turing Test: Can GPT-4 Sway Experts' Decisions?

TL;DR

This paper investigates how GPT-4-generated text influences decision-making among both experts and amateurs, proposing a new evaluation approach based on audience reactions rather than traditional indistinguishability metrics.

Contribution

It introduces a novel perspective on evaluating LLMs by analyzing their impact on human decisions and releases a dataset for future research in this area.

Findings

GPT-4 can significantly sway expert and amateur decisions.

High correlation between audience reactions and multi-dimensional evaluation metrics.

Proposes a new evaluation paradigm based on reader responses.

Abstract

In the post-Turing era, evaluating large language models (LLMs) involves assessing generated text based on readers' reactions rather than merely its indistinguishability from human-produced content. This paper explores how LLM-generated text impacts readers' decisions, focusing on both amateur and expert audiences. Our findings indicate that GPT-4 can generate persuasive analyses affecting the decisions of both amateurs and professionals. Furthermore, we evaluate the generated text from the aspects of grammar, convincingness, logical coherence, and usefulness. The results highlight a high correlation between real-world evaluation through audience reactions and the current multi-dimensional evaluators commonly used for generative models. Overall, this paper shows the potential and risk of using generated text to sway human decisions and also points out a new direction for evaluating generated text, i.e., leveraging the reactions and decisions of readers. We release our dataset to assist future research.