The impact of ChatGPT on reliance on AI advice

Abstract

To improve decision-making, organizations have been increasingly employing AI tools. The use of AI in human decision-making gave rise to algorithm aversion or over-reliance on AI when users do not appropriately rely on AI advice. Literature names the lack of familiarity with AI as a reason for the inability of users to rely on AI advice appropriately. Since ChatGPT has gained global interest, there have been concerns about ChatGPT and other Large Language Models (LLMs) being misused. This thesis argues that familiarity with AI improves appropriate reliance on it. The thesis presents two studies that measure how experience with ChatGPT impacts trust in other AI tools, appropriate reliance, and detection of incorrect information. The results from these studies suggest that experience with ChatGPT improves appropriate reliance on it, which in turn improves decision-making, suggesting that ChatGPT and other AI tools can be used to increase users’ ability to distinguish between correct and incorrect AI advice and to act accordingly, by relying on it when it is correct. Furthermore, seeing ChatGPT err decreases trust in it and improves future verification of the information it provides, further improving appropriate reliance on AI. Nevertheless, it is still easier for individuals to determine when another individual erred than when ChatGPT did.

Date of Award	27 Jun 2023
Original language	English
Awarding Institution	Universidade Católica Portuguesa
Supervisor	Filipa de Almeida (Supervisor)