The target article provides a sweeping assessment of visual attention, and lands on the general proposal that the crucial bottleneck lies in selecting which task to perform, as opposed to selecting which early visual signals are processed. For very different reasons, we arrived at a similar proposal (Krauzlis et al., Reference Krauzlis, Bollimunta, Arcizet and Wang2014). Being electrophysiologists interested in the brain circuits for visual attention, our suggestion was prompted by some unexpected properties of neurons in those visual circuits, as well as broader considerations about the interactions across brain regions and their possible evolutionary origins. Here I briefly outline the logic for our interpretation and point out where our views may not be fully aligned with those expressed in the target article.

It is well established that visually responsive neurons in primate neocortex are modulated in several distinctive ways with changes in visual attention, and that these changes are linked to the state of visual attention (Maunsell, Reference Maunsell2015). It is also established that visually responsive neurons in the midbrain superior colliculus are modulated with attention (Herman, Katz, & Krauzlis, Reference Herman, Katz and Krauzlis2018; Ignashchenkova et al., Reference Ignashchenkova, Dicke, Haarmeier and Thier2004), and causal manipulations demonstrate that the superior colliculus is necessary for the normal control of visual attention, especially spatial attention (Cavanaugh & Wurtz, Reference Cavanaugh and Wurtz2004; Lovejoy & Krauzlis, Reference Lovejoy and Krauzlis2010). The contributions of the superior colliculus to the performance of attention tasks often include controlling the eye movements required by the task, but the scope of its contributions go beyond the motor. The superior colliculus contributes to visual attention even when eye movements are not involved – causal perturbations of the superior colliculus alter the allocation of covert visual attention (Lovejoy & Krauzlis, Reference Lovejoy and Krauzlis2010, Reference Lovejoy and Krauzlis2017). These contributions to attention require a different explanation and presumably involve ascending projections from the midbrain through the thalamus to the forebrain.

The most parsimonious version of this idea was that the superior colliculus provides feedback to neurons in early visual cortex that are modulated with attention. However, when we directly tested this idea, we found that the profound changes in attention task performance caused by manipulations of the superior colliculus did not change the visual responsiveness in cortex but instead left the attention-related modulations completely unaltered (Zenon & Krauzlis, Reference Zenon and Krauzlis2012). This dissociation between the signatures of attention in visual cortex and the ability to perform visual attention tasks was surprising and led to considerable debate and some consternation.

Here we encounter a slight disagreement. The target article interprets our findings as supporting the conclusion that attention-related modulation in visual cortex might play no meaningful role in regulating attention. We drew a more conservative interpretation that we still believe is correct – that the superior colliculus exerts its influence over visual attention through ascending projections that do not primarily target early visual cortex but instead target later stages of visual processing.

Among the potential targets downstream of the visual cortex, we pointed out that circuits through the basal ganglia were an intriguing possibility (Krauzlis et al., Reference Krauzlis, Bollimunta, Arcizet and Wang2014). Beyond regulating movements, the basal ganglia are known to be important for value-based decisions about how to act based on an animal’s belief about the current state of the world, invoking what is often referred to as a “decision policy”. We argued that the implementation of a decision policy by the basal ganglia necessarily restricts behavioral control to be guided by some sensory signals while reducing or excluding others. This has the effect of implementing a form of sensory selection, but at the point where those sensory signals are applied to the behavioral output rather than during early sensory processing itself. This perspective on basal ganglia function outlines a form of “late selection” that I think nicely aligns with the target article’s emphasis on the importance of task selection. When a subject selects a task, they adopt a belief about the current state of the world and a decision policy about how they should act – including which sensory signals should be used – given that state.

The target article goes a bit further in its conclusions, casting doubt on the importance of flexible pooling of sensory signals, a form of “early selection”. I disagree on this point and would counter that adopting different pooling strategies is a necessary component of implementing the selected task. Depending on the task being performed, the subject may be obliged to pool sensory signals very differently to reach a task-appropriate decision. How these changes in pooling strategies are implemented is a fascinating and poorly understood topic, but there is compelling evidence that task-related feedback actively changes the statistical structure of neuronal activity in visual cortex (Bondy, Haefner, & Cumming, Reference Bondy, Haefner and Cumming2018). I would suggest that the alternatives presented in the conclusion of the target article are not mutually exclusive, but instead tightly linked.

There is also an argument to be made for “selection in the middle”. In addition to interactions with the basal ganglia, ascending activity from the superior colliculus has an especially large functional effect in the middle of the visual cortical hierarchy, at a stage midway between the representation of basic visual features in early visual cortex and the representation of complex objects in late visual cortex (Bogadhi et al., Reference Bogadhi, Bollimunta, Leopold and Krauzlis2019, Reference Bogadhi, Katz, Bollimunta, Leopold and Krauzlis2021). By modulating processing at this middle point, this circuit connects the evolutionarily conserved midbrain mechanisms for fast selection of salient objects with the more recently evolved neocortical mechanisms for extracting detailed information about specific visual objects.

Extending this point about evolution, primates are not the first animals to possess something that can be identified as visual attention. We are understandably proud of our prodigious cortical forebrain, but the emergence of visual attention predates the evolution of the neocortex and involves specializations and extensions of much older forebrain and midbrain circuits (Krauzlis et al., Reference Krauzlis, Bogadhi, Herman and Bollimunta2018). We should be open to the possibility that understanding visual attention will require a fuller appreciation of how these complex circuits evolved and, even more germane to resolving disorders of visual attention, a fuller appreciation of how they develop and continue to undergo plastic changes even in the adult.