Large Language Models are overconfident and amplify human bias

TL;DR

This paper investigates overconfidence in large language models, revealing they are significantly overconfident compared to humans, especially when uncertain, and that their input influences human decision-making unpredictably.

Contribution

It provides the first systematic evaluation of overconfidence in LLMs, demonstrating their overconfidence levels and effects on human decision-making.

Findings

LLMs are 20-60% overconfident in their answers.

Humans have similar accuracy but lower overconfidence than LLMs.

LLM input increases both accuracy and overconfidence in humans.

Abstract

Large language models (LLMs) are revolutionizing every aspect of society. They are increasingly used in problem-solving tasks to substitute human assessment and reasoning. LLMs are trained on what humans write and are thus exposed to human bias. We evaluate whether LLMs inherit one of the most widespread human biases: overconfidence. We algorithmically construct reasoning problems with known ground truths. We prompt LLMs to answer these problems and assess the confidence in their answers, closely following similar protocols in human experiments. We find that all five LLMs we study are overconfident: they overestimate the probability that their answer is correct between 20% and 60%. Humans have accuracy similar to the more advanced LLMs, but far lower overconfidence. Although humans and LLMs are similarly biased in questions which they are certain they answered correctly, a key difference emerges between them: LLM bias increases sharply relative to humans if they become less sure that their answers are correct. We also show that LLM input has ambiguous effects on human decision making: LLM input leads to an increase in the accuracy, but it more than doubles the extent of overconfidence in the answers.