Questionable helpers: people are skeptical about AI, but appreciate its advice

- EN - DE
C: Pexels / Andrea Piacquadio
C: Pexels / Andrea Piacquadio

Positive experiences mitigate aversion to artificial intelligence as advice

Generative artificial intelligence (AI) generates data that is similar to the human way of thinking. If a generative artificial intelligence such as ChatGPT functions as "practice advice" for overcoming personal or social challenges, it is considered by people to be less competent than human experts. However, the advice given was accepted and positive experiences with AI led to more willingness to seek advice from AI. These are the exciting results of a study by an interdisciplinary research team from the University of Vienna and the Emlyon Business School in France, which analyzed how people rate AI compared to human advisors. The work has now been published in the journal Communications Psychology.

Generative AI such as ChatGPT is increasingly finding its way into our everyday lives. Thanks to powerful algorithms and an extensive "knowledge base", it can explain complicated topics in an understandable way and also give advice on how to overcome challenges. However, for AI to actually be accepted in its advisory function, potential users must not only have a positive attitude towards the advice given, but also towards AI as an advisor.

How we humans function

A team of scientists led by Robert Böhm from the Faculty of Psychology at the University of Vienna conducted a series of experiments with over 3,300 test subjects to clarify how people rate "AI advice" compared to human experts. Participants were offered suggestions for solutions to personal and social challenges that came either from ChatGPT or from human experts - with only half of the test subjects knowing the identity behind the advice (AI or human). The test participants then rated the competence of the advice and the quality of the tips.

First author of the study Robert Böhm on the exciting results of this experiment: "We were able to establish clear evidence of a subjective aversion to AI advice that did not correspond to the objective evaluation of the recommendations given." The test subjects devalued the competence of the AI advice if they knew that the tip came from the AI - while the evaluation of the quality of the tips given, the willingness to follow them or share them with others was not influenced by this knowledge.

How we humans vote

Böhm and his team carried out further research to determine how the experience gained affected the choice of advice - whether from human experts or from an AI. The results showed that among those who had already had positive experiences with AI and were aware of this, the decision in favor of advice from AI was significantly higher (32%) than among the group that did not previously know whether the advice was AI-generated (23%).

"Our results suggest that a possible skepticism towards AI as advice for everyday problems could decrease the more experience people have with such applications," says test manager Böhm, interpreting the results. The advice from ChatGPT is perceived as very comprehensible, yet precisely formulated. According to the authors, this offers a real opportunity for AI to be integrated into our everyday lives. "Why not try it out for yourself," says Robert Böhm.

Original publication:

Robert Böhm, Moritz Jörling, Leonhard Reiter, Christoph Fuchs: People devalue generative AI-s competence but not its advice in addressing societal and personal challenges. In: Communications Psychology, 2023.
DOI: 10.1038/s44271’023 -00032-x