Finally, estimated trial-by-trial PEs from participants’ responses behaved similarly as insight experiences (Kullback-Leibler Divergence).
Yesterday at #ICON2025 I got to present our poster with #EttoreAmbrosini, @palencianoap.bsky.social, and @mruz.bsky.social on flexible task representations. I loved the chance to share our preliminary results and hear such thoughtful feedback!
Recent research proposed a relationship between insight experiences and the resolution of prediction errors (PEs) during sudden performance increases. So, more incorrect predictions, more intense insights?
Not quite! Bayesian inference highlights the role of the (un)certainty of predictions.
Indeed, in 3 datasets, the intensity of insight was best explained by an interaction between the accuracy and precision of initial predictions. More precision led to stronger insight for incorrect, but to less intense insight for correct predictions, corroborating ideas of insight as a readout of PE
According to Bayesian inference, predictions are weighted by their respective uncertainty, influencing levels of surprise: violating more certain (vs. uncertain) predictions is more surprising.
We hypothesised a similar interaction for insight experiences if they are related to PEs.
We presented people ambiguous images for which they tried to guess the correct label (to derive prediction accuracy) and rated their confidence in that label (for prediction certainty). Insight intensity was rated as how strong subject’s “Aha!” was once the solution image was shown.
🚨Preprint! “Bayesian surprise tracks the strength of perceptual insight” - Work with @lindedomingo.bsky.social & @gonzalezgarcia.bsky.social
Ever wondered what factors influence the subjective experience of suddenly understanding a previously unclear input?
Click below:
doi.org/10.64898/202...