When choosing between a candy bar and an apple, standard sequential sampling models—like the classic Drift Diffusion Model (DDM)—assume your brain accumulates evidence for all options simultaneously and uniformly. In this view, evidence for each alternative climbs steadily toward a decision threshold, and eye movements are merely a passive reflection of whichever option currently has a higher value.
However, human decision-making is highly dynamic. We do not stare blankly at all options at once; instead, we shift our gaze from one item to another. A growing body of research demonstrates that visual attention does not just reflect our preferences—it actively shapes them by modulating the very process of evidence accumulation.
The Mechanism of Asymmetric Accumulation
How exactly does looking at an item alter a decision? In The Attentional Drift Diffusion Model of Simple Perceptual Decision-Making, G. Tavares et al. formalize this using the Attentional Drift Diffusion Model (aDDM). Instead of integrating evidence at a constant rate across all options, the brain dynamically shifts its focus based on real-time gaze tracking. When an item is fixated, its evidence accumulation rate is fully weighted. Crucially, evidence for the non-fixated items is discounted by a substantial margin, often exceeding 60%.
This asymmetrical weighting explains a well-documented behavioral phenomenon: the last-fixation choice bias. Because the item currently being looked at receives a higher relative weight in the accumulation process, it experiences a rapid surge in evidence at that critical moment. Consequently, the final stimulus fixated before a decision is reached has a disproportionately high probability of crossing the decision threshold and being selected.
To make these models computationally tractable for complex choices, Armin W Thomas et al. introduced the Gaze-weighted Linear Accumulator Model (GLAM) in GLAMbox: A Python toolbox for investigating the association between gaze allocation and decision behaviour. Rather than tracking every millisecond-level fixation sequence, GLAM simplifies the process by weighting the evidence signal based on the aggregate proportion of time an item was fixated. By applying a specific discounting parameter to non-fixated items, the authors demonstrate that one can capture individual differences in how strongly gaze allocation influences final choices while maintaining high predictive accuracy and computational efficiency.
Value Boosting versus Information Fidelity
While early models treated attention as an external, ad hoc "value boost" added to the fixated option, newer normative frameworks explain why this bias occurs from an information-theoretic perspective.
In Optimal policy for attention-modulated decisions explains human fixation behavior, Anthony I. Jang et al. argue that attention does not artificially inflate an item's intrinsic value. Instead, looking at an item enhances the reliability or signal-to-noise ratio of the information gathered from it. Similarly, Mingyu Song et al. in Proactive Information Sampling in Value-Based Decision-Making: Deciding When and Where to Saccade model this as fixation-dependent noise scaling. The attended source provides high-fidelity, low-variance samples, while unattended peripheral sources are subject to increased variance and noise.
This distinction between value boosting and information fidelity yields unique behavioral signatures. If attention merely adds a constant positive boost, looking longer at an option should always make you more likely to choose it. However, in Considering What We Know and What We Don’t Know: Expectations and Confidence Guide Value Integration in Value-Based Decision-Making, Romy Frömer et al. demonstrate that attention interacts with prior expectations: increased attention boosts choices for options expected to be better-than-average, but actually hinders choices for options known to be worse-than-average. Attention clarifies the true value of an item rather than blindly making it more appealing.
Proactive and Optimal Information Sampling
If gaze modulates evidence accumulation, what guides gaze? The relationship is fundamentally bidirectional: value guides gaze, and gaze shapes value.
Instead of moving randomly, our eyes follow an optimal policy designed to maximize information gain under cognitive and physical constraints. Frederick Callaway et al. show in Fixation patterns in simple choice reflect optimal information sampling that decision-makers dynamically allocate attention to options that are both highly uncertain and estimated to have high value. This active sampling strategy ensures that cognitive resources are spent resolving uncertainty where it matters most.
This cost-benefit trade-off can be modeled using decision-theoretic frameworks. As demonstrated by Erdan Gu et al. in Generating Sequence of Eye Fixations Using Decision-theoretic Attention Model, sequential attention shifts balance the cumulative reward of processing new information against the physical and computational costs of shifting focus. At a neural level, Amitai Shenhav et al. reveal in Dissociable neural mechanisms track evidence accumulation for selection of attention versus action that the brain simultaneously manages this dual problem: the ventromedial prefrontal cortex (vmPFC) tracks the reward value of evidence, while the dorsal anterior cingulate cortex (dACC) manages uncertainty to guide high-level attentional strategies.
Neural Evidence for Serial Value Processing
The computational assumption that the brain prioritizes the fixated option is strongly supported by neurophysiological evidence. Traditional multi-alternative models often assume that the brain computes and averages the values of all visible options simultaneously. However, single-cell recordings tell a different story.
In Covert shift of attention modulates the value encoding in the orbitofrontal cortex, Yang Xie et al. show that neurons in the orbitofrontal cortex (OFC) dynamically shift their encoding to represent the value of only the currently attended item. When attention shifts, the neural representation of value shifts with it in a serial, "winner-take-all" fashion. This confirms that the brain does not process all options in parallel; instead, attention acts as a gatekeeper, dictating which option's value is actively fed into the evidence accumulator at any given moment.
Scaling to Multi-Alternative and Risky Choices
While early attention-modulated models focused on simple binary choices, recent work has extended these principles to complex, multi-alternative, and risky environments.
When choosing among many options, Felix Molter et al. find in Uncovering the computational mechanisms underlying many-alternative choice that models where gaze actively enhances or modulates value consistently outperform passive models where gaze merely filters which items are considered.
Furthermore, in Gaze-dependent evidence accumulation predicts multi-alternative risky choice behaviour, Armin W Thomas et al. demonstrate that gaze-dependent evidence accumulation can even predict classic "irrational" economic biases, such as decoy or compromise effects. Contextual changes in a choice set systematically alter visual sampling behavior. By increasing dwell time on a specific option, the context amplifies that option's evidence accumulation rate, driving predictable shifts in choice.
By moving away from the assumption of uniform, parallel information processing, gaze-dependent evidence accumulation models provide a deeply intuitive explanation for how we make decisions. Attention is not a passive spectator; it is an active, optimizing mediator that dynamically filters noise, shapes value representation in the brain, and ultimately steers the course of our choices.
Key Foundational Work
The Attentional Drift Diffusion Model of Simple Perceptual Decision-Making (2017) — contributed 8 concept elements to this direction.
Generating Sequence of Eye Fixations Using Decision-theoretic Attention Model (2005) — contributed 2 concept elements to this direction.