One Strike and I'm Out

With the technological breakthroughs in computational power and artificial intelligence (AI) in recent years, intelligent algorithms are increasingly able to assist or even take over tasks from human decision makers. Yet, with the rise and success of AI-based decision-support systems, another big challenge has occurred. Researchers have found that humans reject algorithms due to a general aversion based on mistrust, even when the usage of the algorithms would be beneficial.

This phenomenon – labeled as algorithm aversion – depicts a great challenge for decision making processes. Refusing the help of AI-based decision support systems leads to either misuse, i.e., underuse or overuse of algorithmic help, abuse, i.e., manipulation of algorithm results, or even non-use, e.g., outright refusal to use an algorithm, ultimately causing inferior decisions. Current research has proposed factors that may possibly enhance adoption, e.g., algorithm literacy, human-in-the-loop decision making, and framing the context via behavioral design, but due to the relatively young field of study, many questions remain to be unanswered.

In this project, we investigate in different experiments how initial trust in and subsequent adoption of AI-based decision support systems can be enhanced. In addition, we test how trust and subsequent adoption after seeing an AI-based decision support system err can be restored.

We apply an experimental design with a series of incentivized studies. Participants will be exposed to a prediction task where we can observe real behavior of participants using an AI-based decision support system.

The study aims to raise awareness that not every AI-based decision support system will be adopted smoothly by the decision maker, but barriers to adoption may arise instead.

In some situations, AI-based decision support systems might be able to support decision makers make better decisions. The results of this study can guide managers who are about to introduce such systems in their organizations by comparing and proposing different implementation strategies to overcome initial resistance.

• People tend to reject algorithm-based decision systems even when their use would be beneficial.
• Gaining experience with the system can mitigate algorithm aversion over time.
• Sharing positive experiences of other users can also help reduce algorithm aversion.