Is the sound-induced flash illusion resistant to feedback training, and what does that say about automaticity?

A flash you see twice, or not at all

The sound-induced flash illusion is a well-known trick of multisensory perception: what you hear changes what you think you saw. In the "fission" version, a single flash paired with two quick beeps is often perceived as two flashes. In the "fusion" version, two flashes with one beep can collapse into a single perceived flash. The effect is described as auditory-dominated, meaning the ears tend to override the eyes [1].

Because the illusion is strong and reliable, researchers have treated it as a window into how the brain combines signals from different senses [1]. The practical question behind the training studies is whether that combination process is welded shut or whether practice can loosen its grip.

Practice with feedback makes the illusion weaker

In one study, participants did a standard version of the task before and after a training period. Only the training group completed seven days of behavioral practice with feedback; a control group simply took the same tests twice. The trained group showed reduced responses to both the fission and fusion illusions, and the improvement was linked to better perceptual sensitivity — in plain terms, they got better at telling the real flashes apart from the illusory ones [1].

The gains did not keep growing indefinitely. A plateau appeared during training, with performance tending to stabilize by about the fifth day [1]. That pattern fits a skill that can be sharpened up to a point, rather than one that dissolves with enough effort. The effect was reported as especially clear for the fission illusion [1].

Long-term training of a different kind shows the same direction

A separate line of work asked whether long-term exercise training, rather than lab practice with feedback, also changes the illusion. People with long-term skill-based exercise training showed a smaller illusion than controls, and a group doing physical exercise also showed a lower illusion than controls [2]. The skill group's advantage was tied to a finer temporal binding window — a narrower window of time within which sights and sounds get merged — meaning they were less likely to fuse signals that were actually slightly apart [2].

That study went further and used transcranial direct current stimulation to modulate the cerebellum, a brain region involved in motor skill learning. The results indicated the cerebellum is involved in the illusion, and the authors proposed that the skill group's smaller illusion came from increased cerebellar function built up through long-term training [2]. This is a proposed mechanism rather than a settled cause, but it points the same way as the feedback study: experience can change how strongly the illusion appears.

Does a trainable illusion stop being automatic?

The feedback-training result is the more direct test of the automaticity question, because it shows the illusion shrinking after a week of explicit practice with feedback [1]. If the illusion were entirely automatic and impervious to experience, that reduction should not have occurred. So the honest reading is that the illusion is robust but not rigid: some part of the seeing-and-hearing merge is open to learning.

There is an important limit on how far this goes. The training study measured changes in illusion responses and perceptual sensitivity; it did not test automaticity directly, so the claim that the illusion is "not automatic" is an inference from trainability, not a measured property [1]. The exercise study likewise speaks to training effects and a possible cerebellar mechanism, not to automaticity as such [2]. What the evidence supports is narrower and still useful: the sound-induced flash illusion can be reduced by long-term training, which is hard to square with the idea that it is a completely fixed, untouchable reflex of multisensory integration.

Sources used in this answer

Related research questions