When and Why Implementing Bad Ai Algorithms Feels Worse Than Self-Initiating Them

Algorithm regulation – i.e., the question of who is authorized to decide which algorithms to implement, portrays a key ethical debate regarding AI. We focus on AI programmers, who play a central role in AI deployment processes. Findings of 4 studies indicate that compared to Low-agency programmers who are instructed which algorithm (deontological or utilitarian) to implement, high agency programmers who freely choose which algorithm to employ, engage in a pre-justification process, and are thus better equipped to overcome cognitive-dissonance when encountering fatal unfavorable outcomes. Consequently, they feel less responsibility, guilt, and regret about both their programming and the outcome. We further show that this effect has an impact on their future programming decisions – high agency programmers are more likely to stick with their previous programming choices even when they learn that these choices led to fatal outcomes.