AI has become increasingly efficient in anticipating our behavior. Will this impact, in the near future, how much we feel control over events generated with AI-assistance? In everyday life, our sense of agency over events occurring at various delays after our actions has adapted to accommodate these delays. Here we investigate whether our sense of agency can also adapt to a highly unusual situation, in which a consequence precedes an action. We used an online game where players tried to beat the computer at finding and clicking on a target to trigger an animation, while in fact an algorithm triggered the animation before the players9 click. The animation was not randomly controlled by the algorithm, but rather based on the history of the players9 past movements and on the beginning of their current movement. We used modeling and machine learning decoding approaches to capture how players compute their reported sense of agency over the animation. We found evidence that, in less than an hour, players implicitly learned, despite the unusual timing, that they were controlling the animation and adapted their sense of agency accordingly. Such findings may help us to anticipate how humans will integrate AI-assistance to guide their behavior.

Humans progressively feel agency over events triggered before their actions