Cause-effect perception in an object place task

In this paper, we conducted an exploratory study in virtual reality to investigate whether people can discover causal relations in a realistic sensorimotor context and how such learning is represented at different processing levels (conscious-cognitive vs. sensorimotor). Additionally, we explored the relationship between human causal learning and state-of-the-art causal discovery algorithms. The task consisted of placing a glass on a surface. To enhance ecological validity, the setup included haptic rendering to simulate the glass's weight and contact force. The glass would break if the contact force exceeded its breakability threshold, determined by the causal structure weight → breakability ← color . Participants were asked to repeatedly transport and place glasses of varying weights and colors on a surface without breaking them. Therefore, to accomplish the task, participants had to discover the underlying causal structure. The trials were conducted over three separate sessions, each aimed to capture a different behavior [(i) naive and causally unaware, (ii) exploratory, and (iii) consolidated and causally aware]. After each session, participants completed a questionnaire providing a measure of their conscious understanding of the task's causal structure. Sensorimotor representations were inferred by applying three causal-discovery algorithms (PC, FCI, FGES) to the recorded trial-by-trial variables, and conditional mutual information was used to quantify the strength of causal influence on the sensorimotor level. Results show that (i) participants identified the weight-breakability link (≈76% correct after the final session) and, to a lesser extent, the color-breakability link (≈43% correct), but they could not reliably infer causal direction. (ii) Sensorimotor analysis revealed a robust weight-force coupling that increased across sessions, whereas the color-force coupling was weak and noisy, yet mutual information indicated an attempted learning. (iii) Discovery algorithms recovered the underlying structure across the three sessions. Together, these findings indicate that humans can, to some extent, perceive the causal structure of the task and that conscious and sensorimotor representations are partially dissociated.