When we choose between a chocolate bar and an apple, we often assume our brain is simply comparing two fixed values. However, a growing body of research suggests that the act of choosing is far more dynamic: where we look directly changes what we value. This process, known as gaze-dependent evidence accumulation, reveals that visual attention is not just a window into our preferences, but a primary driver of them.
The Mechanism of Biased Integration
In classical models of decision-making, such as the Drift Diffusion Model (DDM), evidence for different options is accumulated in parallel until one reaches a threshold. G. Tavares and Antonio Rangel proposed a critical modification in The Attentional Drift Diffusion Model of Simple Perceptual Decision-Making: the rate of evidence accumulation is not constant. Instead, it is "gated" by visual fixation.
When you look at an item, the brain "amplifies" the evidence for that item. Conversely, the evidence for the items you aren't looking at is "discounted." This means that if you happen to look at the chocolate bar longer—perhaps because of its bright packaging—the internal counter for the chocolate bar ticks upward faster than it would otherwise. This creates a "fixation bias" where the last item fixated before a decision is reached has a disproportionately high probability of being selected, simply because it received a final boost of evidence at the critical moment the threshold was crossed.
Quantifying the Gaze Discount
To move from theory to prediction, researchers developed the Gaze-weighted Linear Accumulator Model (GLAM). As detailed by Armin W. Thomas, Felix Molter, and colleagues in Gaze Bias Differences Capture Individual Choice Behavior, this model introduces a specific discounting parameter ($\gamma$).
This parameter captures how much the "unattended" options are ignored. If $\gamma$ is 1, attention has no effect. If $\gamma$ is 0, the brain completely ignores any evidence for an item unless we are looking directly at it. Interestingly, Thomas found that for some individuals, this parameter can even be negative. In these cases, the perceived value of an item doesn't just stop growing when we look away; it actively decays or "leaks," meaning we lose interest in things the moment they leave our sight.
To make these complex models accessible, the researchers released GLAMbox, a toolbox that allows scientists to use aggregate gaze fractions (the total percentage of time spent looking at an item) to predict choices, proving that even simplified summaries of gaze can accurately capture the nuances of human behavior.
The Feedback Loop of Value and Threat
Is gaze just a random search, or does value tell the eyes where to go? Toby Wise, Peter Dayan, and Raymond J. Dolan argue in A computational account of threat-related attentional bias that there is a bidirectional feedback loop.
In their study of threat perception, they found that we naturally pay more attention to stimuli that might cause harm. However, this increased attention then acts as a "gain" on the learning process, causing us to overestimate the actual probability of threat. This creates a cycle: we look at what we fear, and looking at it makes us fear it more. This mechanism provides a potential explanation for the persistent attentional biases observed in anxiety disorders, where the act of monitoring for threats actually reinforces the perception of danger.
Why the Brain Biases Itself
From an evolutionary perspective, why would the brain allow gaze to "distort" evidence? Anthony I. Jang and Ravi Sharma suggest a normative explanation in Optimal policy for attention-modulated decisions explains human fixation behavior. They argue that attention doesn't just boost the magnitude of a signal; it increases the reliability of the information.
By focusing on one item, the brain reduces the noise in that specific stream of evidence. Under this view, the "fixation bias" is an emergent property of an optimal strategy: the brain spends more time looking at items that are likely to be the best, and because it has more reliable information about those items, it is more likely to choose them.
This hierarchy of control is managed by specific neural circuits. Amitai Shenhav and Sebastian Musslick found that while the ventromedial prefrontal cortex (vmPFC) tracks the reward value of evidence, the dorsal anterior cingulate cortex (dACC) manages the allocation of attention itself, deciding which information is worth the effort of focus (Shenhav et al., 2018).
From Simple Snacks to Risky Lotteries
The implications of this research extend far beyond choosing between snacks. In Gaze-dependent evidence accumulation predicts multi-alternative risky choice behaviour, Armin W. Thomas and Hauke R. Heekeren demonstrated that these same gaze-driven mechanisms explain how we choose between complex, risky lotteries.
Contextual effects—like the "decoy effect," where adding a third, inferior option changes your preference between the original two—can be explained by how the new option shifts your gaze patterns. By changing how we sample information, the environment can systematically bias our decisions without ever changing the objective values of the choices themselves.