Precision not prediction: Body-ownership illusion as a consequence of online precision adaptation under Bayesian inference

Abstract

Humans can experience body ownership of new (external) body parts, for instance, via visuotactile stimulation. While there are models that capture the influence of such body illusions in body localization and recalibration, the computational mechanism that drives the experience of body ownership of external limbs is still not well understood and under discussion. Here, we describe a mathematical model of the dynamics of this phenomenon via uncertainty minimization. Using the Rubber Hand Illusion (RHI) as a proxy, we show that to properly estimate ones arm position, an agent needs to infer the least uncertain world model that explains the observed reality through online adaptation of the signals relevance, i.e., its precision parameters (the inverse variance of the prediction error signal). Our computational model describes that the illusion is triggered when the sensory precision estimate quickly adapts to account for the increase of sensory noise during the physical stimulation of the rubber hand due to the occlusion of the real hand. This adaptation produces a change in the uncertainty of the body position estimates, yielding a switch of the perceived reality: the "rubber hand is the agents hand" becomes the most plausible model (i.e., it has the least posterior uncertainty). Overall, our theoretical account, along with the numerical simulations provided, suggests that while the perceptual drifts in body localization may be driven by prediction error minimization, body-ownership illusions may be a consequence of estimating the signals precision, i.e., the uncertainty associated with the prediction error.