The brain weights sensory cues by their reliability

Correct: C — They weight each cue by its inverse variance, like a near-optimal Bayesian observer

Ernst & Banks (2002) showed that human visual–haptic integration follows a maximum-likelihood (Bayesian) rule: each cue's contribution is weighted by 1/variance, so noisier cues count less. When they manipulated visual noise, participants relied more on touch — and the integrated estimate had lower variance than either single-cue estimate, exactly as Bayes predicts. This statistical optimality has since been replicated for visual–auditory, visual–vestibular, and proprioceptive–visual integration. The brain treats sensory cues like a Kalman filter — not by deferring to a 'dominant' modality, but by reliability-weighted averaging.