In today’s fast-changing environments, effective learning of current information has important economic consequences and is essential for good decision making. Despite the importance of quick and effective learning, research suggests that people do not integrate new information unbiasedly. Instead, people tend to show a preferential integration of positive information into their existing beliefs while disregarding negative information that is inconsistent with their positive self-concept, their core beliefs, or their preferences. This phenomenon, which is referred to as valence-dependent belief updating, optimistic belief updating bias, or good news/bad news effect, is one of the most consistent, prevalent, and robust biases documented in research. The bias promotes harmful behaviors such as smoking, over-spending, and irrational risk-taking and is considered one of the core causes of the financial downfall in 2008. It might well be considered a key contributor to individual failures to take precautions during the health crisis in 2020. Despite the effort, attempts to reduce the bias by educating participants about risk factors have led to little change and, in some instances, even increased the bias.

Across four online experiments and one field study, we provide robust evidence that AI advisors can help mitigate valence-dependent learning. In particular, our results repeatedly show that people tend to update their beliefs valence-dependent when receiving advice from a human but less valence-dependent when receiving it from AI. Thus, people are more likely to learn negative information when the information is provided by AI rather than a human. These results have important implications for organizations in the public and private sector that need to confront customers with negative information, but also for individuals who aim to make better decisions and learn more effectively in our fast-paced environments.

A Project by Prof. Anne Scherer and Cindy Candrian