Fleming and Michel propose that conscious visual perception operates “offline,” allowing organisms to evaluate and choose goals, weigh alternative courses of action, and select the most appropriate thing to do. Of necessity then, conscious visual perception is slow. The visual control of actions, they argue, depends on separate visuomotor pathways, particularly when there is a premium on speed. This is an attractive and compelling argument – and resonates well with Goodale and Milner’s (Reference Goodale and Milner1992) account of the functions of the ventral and dorsal visual streams in the cerebral cortex (Figure 1A).

Figure 1.

(Goodale and Corneil). A. In addition to cortical projections through the dorsal and ventral streams, visual input also reaches subcortical structures like the superior colliculus, which can command ultra-rapid responses like express saccades or express visuomotor responses. B. Express visuomotor responses are the first phase of upper limb muscle recruitment, depicted here on the right pectoralis muscle involved in cross-body reaching. On pro-reaches toward a suddenly appearing target, the express visuomotor response starts consistently within ∼80 ms, preceding the larger subsequent phases of recruitment that are more time-locked to reach onset (white squares in trial-by-trial color plots of muscle recruitment; each row is a different trial, ordered by reach reaction time). On anti-reach trials away from the target, the express visuomotor response is locked to target location despite the goal to move in the other direction; occasionally this phase of recruitment is strong enough to move the arm in the wrong direction (errors). Right subplot shows average muscle recruitment for pro-reach trials, and correct and erroneous anti-reach trials. A Made in BioRender. B Modified from Gu et al. (Reference Gu, Wood, Gribble and Corneil2016).

Figure 1 Long description

Panel A: The diagram shows the visual pathways in the brain, including the dorsal and ventral streams, and the superior colliculus. Visual input reaches these areas, enabling express responses. Panel B: The heatmaps display muscle recruitment patterns during pro-reach and anti-reach tasks. The x-axis represents time aligned to target onset in milliseconds, and the y-axis represents individual trials ordered by reach reaction time. The pro-reach heatmap shows consistent muscle recruitment starting within 80 milliseconds. The anti-reach heatmap indicates muscle recruitment locked to the target location, sometimes leading to errors. The line graph on the right shows average muscle recruitment for pro-reach trials and correct and erroneous anti-reach trials, with the y-axis representing voltage in millivolts.

But it is important to remember that the retina also projects directly to several subcortical structures (Figure 1A), even in the human brain, that hark back to our distant vertebrate ancestors for whom the rapid deployment of reflexive responses to visual stimuli would have been advantageous. An instructive example is the superior colliculus (called the optic tectum in non-mammalian vertebrates); see Hoy and Farrow (Reference Hoy and Farrow2025) for a recent review. In frogs and rodents, the optic tectum or superior colliculus plays a crucial role in instinctive behaviors like prey-catching and escape from potential predators. In primates, including humans, the superior colliculus is a central player in the control of extremely short-latency (express) saccadic eye movements that orient the line of sight toward stimuli in the visual periphery. Unlike frogs and rodents, however, we don’t spend our days orienting toward suddenly appearing stimuli or avoiding looming objects; yet we can still respond remarkably quickly in situations where time is of the essence. Are these rapid responses in humans mediated by the superior colliculus, and can they be put at the disposal of voluntary (conscious) control? We propose that nested control networks have evolved that incorporate evolutionarily conserved (subcortical) systems such as the superior colliculus within more recently evolved (cortical) control mechanisms such as the dorsal stream so that both systems work together in the production of adaptive behavior.

Take, for example, the everyday action of reaching out to grasp a glass of water. The textbook view is that commands for such an action arise in fronto-parietal networks and are conveyed to motor cortex, which in turn sends the appropriate signals to the skeletal musculature via the corticospinal pathway. The latency of such a voluntary action is typically far longer than the “core” time required for basic sensory processing and motor programming. After all, when picking up a glass of water, there is usually no premium on speed. But what happens should the glass suddenly begin to slip off the table? In this case, adjustments to the ongoing movements would have to be made rapidly to prevent the glass falling. In short, speed is now of the essence.

In the lab, such fast-as-possible scenarios can be emulated using target-jump paradigms in which the target for an instructed reaching movement is suddenly and unpredictably displaced to a new location. In these experiments, the millisecond-by-millisecond timing of muscle recruitment can be measured relative to target displacement. Experiments of this kind, including our own, have revealed that muscle recruitment in such scenarios plays out over (at least) two phases of muscle activity, with each phase displaying the unique characteristics of the nested subcortical and cortical networks.

The first phase of upper limb muscle recruitment occurs at a remarkably short latency, typically around 80 ms in humans, and is tightly “locked” to the timing of the displacement of the target (red boxes, Figure 1B). The second phase of muscle recruitment is much more variable, sometimes occurring just after the first phase and sometimes much later. It is the second phase that is the primary driver of the limb movement itself.

We have studied this two-phase pattern of muscle recruitment in a variety of different contexts, including not only the jumping target paradigm but also experiments in which participants are instructed to reach quickly to a target that is suddenly presented in the visual periphery. The timing of the first phase, which we have termed the express visuomotor response (Figure 1B), is remarkably similar to express saccades and is far faster than any response generated by the cerebral cortex, including by the dorsal stream. Both express saccades and express visuomotor responses are potentiated by visual stimuli of high contrast and low spatial frequency (Kozak et al., Reference Kozak, Kreyenmeier, Gu, Johnston and Corneil2019; Wood et al., Reference Wood, Gu, Corneil, Gribble and Goodale2015). Incredibly, express visuomotor responses also show a rudimentary sensitivity to faces (Mekhaiel, Goodale, & Corneil, & Corneil, Reference Mekhaiel, Goodale and Corneil2024), as do express saccades. Taken together, these similarities have led us to suggest that express visuomotor responses stem from the superior colliculus, reaching the upper limb via the tectoreticulospinal pathway (Corneil & Munoz, Reference Corneil and Munoz2014).

The second phase of muscle recruitment shows all the signatures of being under voluntary (conscious) cortical control, being fully malleable to task demands likely due to the fronto-parietal processing cited above. Accordingly, it is possible to design tasks that pit the first and second phases against each other, requiring participants to “anti” point in the direction opposite to a flashed visual target (Figure 1B). In such tasks, which take time to perform well, the first phase of muscle recruitment is typically suppressed but remains inexorably directed toward the visual stimulus (Gu et al., Reference Gu, Wood, Gribble and Corneil2016). Occasionally, this first phase of recruitment does become forceful enough to cause the participant to erroneously point first toward the target, before the second phase of recruitment correctly points the arm away from the target.

Taken together, the evidence from these and many other experiments strongly suggests that more recently evolved higher-order cortical systems in humans (and many other vertebrates) work in tandem with evolutionarily conserved subcortical systems in the visual control of adaptive behavior (Contemori et al., Reference Contemori, Loeb, Corneil, Wallis and Carroll2023). These nested control systems benefit enormously from the rich and contextualized information provided by conscious visual perception, which as Fleming and Michel argue, emerged in worlds with long-range sensory horizons. At the same time, however, these control systems continue to exploit the rapid and reflexive subcortical visuomotor circuits that evolved in worlds with limited sensory horizons. In short, we enjoy the best of both worlds.