Vision dominates hearing for location — point a speaker somewhere and the ventriloquist's dummy convincingly relocates sound. But for timing, the hierarchy inverts: auditory temporal resolution is roughly ten times finer than visual. When there's a conflict about how many events occurred, hearing can override sight.
The sound-induced flash illusion demonstrates this. A single flash accompanied by two brief beeps is often perceived as two flashes. You will see something that isn't there — because the auditory signal is more trustworthy about timing, and the brain knows it.
This is a direct corollary of Bayesian cue combination: the brain weights each sense by its reliability for the task at hand. Auditory wins on timing. Visual wins on location. The illusion is the reliability weighting, made visible — or rather, made audible as a visual event.
You'll see 20 trials. In each one, look at the circle and count how many flashes appear. Respond immediately. Use headphones if you have them — the effect is strongest with clear audio.
| condition | trials | reported 1 | reported 2 | reported 3+ |
|---|
The brain continuously asks: how many events just occurred? Vision and hearing each submit an answer. When they agree, integration is straightforward. When they conflict, the brain weights each estimate by reliability — and for temporal event counting, auditory reliability is higher. A flash is a diffuse luminance change; a beep is a sharp pressure transient. The auditory system resolves events separated by as little as 2 ms; visual temporal resolution is closer to 20–30 ms. So when the auditory channel says "two events," the brain recalibrates its visual count upward. The phantom second flash is the recalibration, experienced directly.
The reverse effect — two flashes fusing into one when accompanied by a single beep — is weaker and less consistent. The asymmetry matches the reliability difference: auditory temporal information is more strongly weighted, so it more reliably overrides vision than vision overrides auditory timing.
Shams, Kamitani, and Shimojo (2000) first reported the illusion in Nature. Subsequent work found that the illusion produces measurable changes in early visual cortex activity (V1/V2), suggesting the override happens before visual processing is complete — not just at a later decision stage. The brain doesn't wait to hear a full verdict before updating its visual representation.