What seems to have surprised Mario Treviño most in this review is how flimsy the old story starts looking once you line up the evidence. The classic idea says your eyes mostly lunge toward whatever is bright, loud, flashy, or generally behaving like a toddler with a drum set. But the newer picture is messier and way more interesting: your visual system seems to look around like it already has bets on what matters, what might happen next, and what would be worth checking before reality embarrasses it.

The old model was neat. The brain was not.

For years, visual attention was often explained with "salience" - basically, the stuff in a scene that pops out. High contrast, sudden movement, weird color, shiny object, brain go brrr. And yes, that still matters. If a fire alarm goes off, your eyes do not hold a committee meeting.

But Treviño’s review argues that salience alone does a lousy job explaining real gaze behavior. People look differently depending on what they expect, what they are trying to do, how uncertain they feel, and what they just learned. In other words, your eyes are not just reacting. They are sampling.

That is the predictive-processing version of attention: the brain builds guesses about the world, then uses eye movements to test those guesses and reduce uncertainty. Not "What is the brightest thing here?" so much as "What would settle the argument fastest?"

Saccades: the brain’s frantic refresh button

This idea makes eye movements look less like random darting and more like strategy. A fixation can linger because the system is extracting useful evidence. A saccade can fire because the current guess is shaky and the brain wants receipts. Even scanpath entropy - how orderly or chaotic your looking pattern is - may reflect hidden states like surprise, learning, or confidence.

That is the fun twist here. Looking is not just seeing. Looking is asking.

Your brain loves a prediction almost as much as it loves being wrong

Predictive models say perception runs on loops. Higher brain areas send expectations downward. Incoming sensory data send errors upward when those expectations flop. Somewhere in the middle, attention helps decide which signals deserve the VIP treatment.

That helps explain why gaze changes with context. If you are making a sandwich, your eyes skip around the kitchen differently than if you are hunting for your car keys. Same eyeballs. Different internal model. Different mission.

Recent reviews fit this bigger shift. Peelen, Berlot, and de Lange describe perception as a joint inference process in which scene context and object expectations shape one another (doi, PMCID). Hanning, Deubel, and Szinte show that preparing an eye movement comes with its own special kind of attention, not just the covert version wearing a fake mustache (doi). And Anderson’s review makes the broader point that attention is adaptive, shaped by goals, learning, and tradeoffs rather than pure reflex (doi).

Why this matters outside the lab, where nobody says “scanpath entropy”

If this framework holds up, it changes how we think about attention problems in development and psychiatry. Treviño notes that predictive gaze strategies seem to emerge early and may be disrupted in neuropsychiatric conditions. That matters because eye tracking is relatively noninvasive and surprisingly revealing. Watching where people look, when they hesitate, and how they update could become a practical way to probe hidden cognitive states.

It also matters for AI and human-computer systems. If human visual selection is predictive and uncertainty-driven, then better machine vision may come from systems that do not merely rank pixels by contrast but actively choose where to sample next.

The annoying part: the debate is not over

Treviño is not claiming every gaze shift is a triumph of elegant Bayesian wizardry. There are still real arguments about what counts as prediction, what counts as attention, and whether some reported "prediction" effects are really confounds from surprise, repetition, or task demands. Feuerriegel, Vogels, and Kovács make exactly that critique in their review of expectation suppression in vision (doi).

So the current picture is not "mystery solved." It is more like "the suspect list got better." The field is moving from a flat, stimulus-driven account toward a model where attention is flexible, inferential, and tied to action. Your eyes are not passive windows. They are part of the experiment your brain is constantly running on the world.

Which, frankly, explains a lot about why you can stare straight at the ketchup in the fridge and still fail to see it. The bottle was right there. Your prediction model, meanwhile, was off filing a false report.

References

Treviño M. Adaptive visual selection: Predictive control of visual attention. Neurosci Biobehav Rev. 2026;181:106523. DOI

Peelen MV, Berlot E, de Lange FP. Predictive processing of scenes and objects. Nat Rev Psychol. 2024;3:13-26. DOI PMCID

Hanning NM, Deubel H, Szinte M. To look or not to look: dissociating presaccadic and covert spatial attention. Trends Cogn Sci. 2021;44(8):669-686. DOI

Anderson BA. An adaptive view of attentional control. Am Psychol. 2021;76(9):1418-1432. DOI

Feuerriegel D, Vogels R, Kovács G. Evaluating the evidence for expectation suppression in the visual system. Neurosci Biobehav Rev. 2021;126:368-381. DOI

Denison RN. Visual temporal attention from perception to computation. Nat Rev Psychol. 2024;3:261-274. DOI

Disclaimer: The image accompanying this article is for illustrative purposes only and does not depict actual experimental results, data, or biological mechanisms.