The influence of algorithmic advice on decision-making strategies in a hypothesis testing task.

The proposed research project sets out to examine the influence of role-playing AI as an advice giver on individual decision-makers. This is of interest due to the challenge presented to human decision-makers in executing evidence-based recommendations (Amason and Schweiger, 1994). A convincing body of evidence exists regarding how particular decision-making (DM) challenges can be positively influenced by assigning specific roles to advice-providers (De Dreu and West, 2001; Nemeth, Brown and Rogers, 2001). However, in practice, human advisors do not execute such roles consistently or to full effect (Xiao, 2017). This has been attributed to the influence of social conformity and conflict-avoidance behaviours (Cialdini and Goldstein, 2004). Progress in the capabilities of modern AI programs has resulted in the emergence of interactive AI interfaces. Such programs present novel experimental opportunities. This project investigates how algorithmic advice from a role-playing python-based program influences individuals' DM strategies. The primary research question of the project is: What effect does taking algorithmic advice have on individuals performing DM tasks? The secondary research question is: What different effects do the three advice roles have on the DM strategies and accuracy? To study these questions, participants will take part in a computerised medical diagnosis game under two conditions: with and without advice. The hypothesis testing strategies and test accuracy under these two conditions will be examined to provide insight into how the advice influences human decision-makers. The aims of the study are to: 1. Examine the effect algorithmic advice provision on DM information search strategies. 2. Measure the effect of algorithmic advice provision on the task accuracy.