In human-machine collaboration, external assistance can improve task performance but may also diminish the agent’s sense of agency (SoA). SoA can be defined as the feeling of initiation and control over one’s actions and their consequences. This study examines how the SoA is affected by external assistance within different task goals, specifically when assistance substitutes for the agent’s own information‑gain process. In this regard, two variants of a multi‑armed bandit task that share a common generative process but differ in the nature of their goal (exploitative versus explorative) were implemented. The results suggest that under assistance, goal‑dependent divergence emerge. In the explorative condition, increasing information disclosure significantly decreased SoA whereas in the exploitative condition, SoA was not weakened across increasing information disclosure levels. These findings highlight that assistance which delivers knowledge exogenously undermines perceived epistemic progress when success depends on belief improvement. However, it is less detrimental when success depends on producing preferred outcomes through pragmatic action. These results motivate goal‑sensitive principles for the affective engineering of assistive systems that preserve the SoA.
Briard et al. (Thu,) studied this question.