Fleming and Michel argue that conscious vision unfolds too slowly, requiring up to 400 ms, to support real-time action. This conclusion rests on two assumptions – that unconscious vision alone can support relatively fast behavioral responses, and that long-latency postdiction effects reflect the minimal time required for conscious perception – both of which we find questionable.

Fleming and Michel’s target article builds on a functionalistic basis; what is conscious vision good for? While we agree with them that neural correlates of conscious vision alone do not solve the latency of conscious vision, the research on correlates serves as a good starting point, and an acceptable theory of conscious vision should align with the correlative evidence. We maintain that their argument is at odds with what this research tells us. According to a large body of electroecephalographic and magnetoecephalographic research using various paradigms, conscious vision has been observed to correlate with activity around 200 ms after stimulus onset, known as the Visual Awareness Negativity (VAN). In contrast, the later well-known correlate (the P3) has, in many studies, been dissociated from conscious vision and may serve higher-order cognitive functions (for a review, see Förster, Koivisto, & Revonsuo, Reference Förster, Koivisto and Revonsuo2020).

Moving from correlates to function, Fleming and Michel’s case rests on the premise that visually guided behaviors with latencies under ∼400 ms are typically guided by unconscious processes. This is a strong assumption, given that reaction times in many visual tasks center near 300–400 ms. If Fleming and Michel’s case is correct, unconscious perceptual processes should enable humans to respond to visual stimuli they report not seeing. Unconscious visual information can influence behavioral responses to a consciously seen stimulus (e.g., unconscious priming, Vorberg et al., Reference Vorberg, Mattler, Heinecke, Schmidt and Schwarzbach2003), but the evidence that completely unconscious visual information could reliably trigger rapid behavioral responses is lacking. For example, because unconscious stimuli alone typically cannot trigger behavioral responses, unconscious priming is measured by having observers respond to a consciously visible mask. Even though the fastest responses in such paradigms may escape masking, this only applies to a small fraction of the trials (roughly 5% based on van Rullen & Koch, Reference VanRullen and Koch2003). Furthermore, previous studies may have overestimated the influence of unconscious processes by measuring conscious perception using overly conservative methods, as argued by Phillips (Reference Phillips2020) in the case of blindsight. Moreover, blindsight is typically measured using indirect methods such as forced-choice responding or priming because the blindsight patients cannot spontaneously and reliably respond to visual stimuli in the blind visual field. Other evidence suggests that blindsight capacity may be mediated by neuroplastic changes in patients, rather than revealing anatomical pathways that support unconscious visually guided behavior (Railo & Hurme, Reference Railo and Hurme2021). These examples suggest that the capacity of unconscious vision to guide real-time behavior is far more limited than Fleming and Michel argue, whereas conscious perception is tightly linked with real-time visually guided behavior.

Further evidence suggests that, in addition to marking the timing of visual experience, VAN may reflect processes that help initiate visually guided behavior: Participants can reliably report simple contents of conscious vision soon after VAN and before the onset of the later correlate in the P3 time window (Railo, Revonsuo, & Koivisto, Reference Railo, Revonsuo and Koivisto2015). Because the visually guided responses occur before the P3, the later correlate cannot mediate the guidance of behavior based on conscious vision, whereas VAN is in a position to do so. Note that we do not claim that complex visual decisions or volitional behaviors require no further processing time; rather, the evidence suggests that simple responses may be based on conscious representations that emerge around 200 ms after the stimulus.

Next, we turn to Fleming and Michel’s claim that conscious vision is “delayed” up to 400 ms. Here, they appeal to rare and atypically long-lasting postdiction effects, in which perception of a stimulus is influenced by another, presented 350–450 ms later. They discuss two sets of effects.

The first set concerns sequential metacontrast masking experiments. However, these are not a reliable indicator of the typical timing of conscious vision. Because metacontrast masking delays conscious vision (VAN) relative to unmasked stimuli (Railo & Koivisto, Reference Railo and Koivisto2009), the same delay likely occurs in these studies. Under more typical, unmasked conditions, the delay is likely to be shorter than that observed in the metacontrast masking experiments. Moreover, since sequential metacontrast masking is absent in most naturalistic situations for which we are evolutionarily adapted – those discussed by Fleming and Michel – it is unwarranted to generalize from such an artificial paradigm to the general temporal dynamics of conscious vision.

The second set of long-lasting postdiction effects discussed by Fleming and Michel involves attention’s influence on conscious vision. While this topic would merit fuller discussion, two brief points suffice. First, insofar as attention and consciousness are distinct, and attention relates to reportability, it is unclear what the cited studies show about consciousness itself. Second, some of the studies they cite contradict their claim. For example, Ono et al. (Reference Ono, Yamada, Takahashi, Sasaki and Ariga2023) report the peak postdiction effects at 16 and 50 ms stimulus-onset asynchronies (SOAs), with no effect in the later time window emphasized by Fleming and Michel.

In light of the above, the long-lasting postdiction effects do not unequivocally support Fleming and Michel’s position. Then again, broader evidence from postdiction paradigms suggests much shorter integration windows (Arstila, Reference Arstila2015). In both metacontrast masking and the flash-lag effect, postdictive interference disappears once the SOA exceeds 50–80 ms (e.g., Breitmeyer & Öğmen, Reference Breitmeyer and Öğmen2006; Eagleman & Sejnowski, Reference Eagleman and Sejnowski2007). The temporal thresholds differ for different forms of apparent motion; however, even for long-range visual apparent motion, which permits the longest intervals, the optimal SOA is around 100 ms, and perception of smooth motion is largely absent once the SOA approaches 200 ms (e.g., Kolers & Pomerantz, Reference Kolers and Pomerantz1971; Strybel & Vatakis, Reference Strybel and Vatakis2004). When these temporal windows are combined with the ∼120 ms of feedforward processing noted in the target article, the resulting delay aligns well with VAN. The emerging overall profile is, hence, far more consistent with VAN-based estimates of conscious perception than with the prolonged delay Fleming and Michel propose.