Reversing the Knobe Effect: AI Takes Less Blame for Harmful Decisions

When a decision leads to harmful side effects, observers typically judge the action as more intentional than if it results in beneficial side effects, an asymmetry known as the “Knobe effect.” Using an online experiment, we examine whether this effect extends from human decision-makers to artificial intelligence (AI) systems. We also investigate the effects on the attribution of praise or blame to the decision-maker or the company. The results show that, with humans, harmful side effects led to greater perceived intentionality than beneficial side effects and resulted in more blame than praise. For AI systems, however, the effect reversed—beneficial outcomes led to greater perceived intentionality than harmful outcomes. We also find that the traditional Knobe effect holds true for company attribution regardless of who or what makes the decision. Our research contributes to the Knobe effect by showing that the effect reverses for AI and highlighting that organizations, not algorithms, are held publicly accountable for AI missteps.