Interactive Inference: A Neuromorphic Theory of Human-Computer Interaction

Abstract: Neuromorphic HCI is a new theoretical approach to designing better UX inspired by the neurophysiology of the brain. Here, we apply the neuroscientific theory of Active Inference to HCI, postulating that users perform Bayesian inference on progress and goal distributions to predict their next action (Interactive Inference). We show how Bayesian surprise between goal and progress distributions follows a mean square error function of the signal-to-noise ratio (SNR) of the task. However, capacity to process Bayesian surprise follows the logarithm of SNR, and errors occur when average capacity is exceeded. Our model allows the quantitative analysis of performance and error in one framework with real-time estimation of mental load. We show through mathematical theorems how three basic laws of HCI, Hick's Law, Fitts' Law and the Power Law fit our model. We then test the validity of the general model by empirically measuring how well it predicts human performance in a car following task. Results suggest that driver processing capacity indeed is a logarithmic function of the SNR of the distance to a lead car. This positive result provides initial evidence that Interactive Interference can work as a new theoretical underpinning for HCI, deserving further exploration.