Nothing in the world happens on time. An event conveyed by sound, light, radio waves, or even gravity reaches the senses only after a delay. We remain unaware of these temporal distortions and experience the world seamlessly, but although consciousness appears continuous, delays may happen unnoticed.
In a recently published paper, Michael Herzog, Leila Drissi-Daoudi, and Adrien Doerig argue, on the basis of psychophysical experiments, that perception updates in discrete steps, with as much as half a second elapsing between a sensory event and its conscious registration. While the notion of discrete updating is compelling, the proposed rate of two conscious percepts per second is likely too slow. Herzog et al. also propose “a two-stage model, in which long-lasting unconscious processing precedes discrete conscious percepts.” A more plausible arrangement might involve a hierarchy of at least three successive stages leading to conscious perception.
The fine-grained temporal organization of perception has long puzzled philosophers and scientists, but it has only been in the last century that technological advances and new experimental methods have made it possible to tackle these questions. Evidence in recent years points toward a discrete temporal structure of perception.1 This has led researchers to propose that percepts occur and change about ten times per second. Movies can be downsampled to about ten frames per second without viewers losing much information, but not much slower.2 Studies have shown that one-tenth of a second, or 100 milliseconds (ms), is also the approximate duration for which a stimulus can still have a strong retrospective impact on the perception of preceding events, as revealed by perceptual fusion, backward-masking, or so-called postdiction studies. Taken together, these results suggest there is a window of ~100ms duration in which events are compared, combined, and integrated before the resulting construct is committed to consciousness. This conscious window is rhythmic, with the metaphorical camera shutter closing every 100ms or so as a consequence of brain oscillations, particularly in the alpha-frequency band, around 10Hz. These oscillations provide a periodic framing, like the recurring snapshots captured by a video camera. Even though the metaphorical shutter in this scenario closes every 100ms or so, no sensory input is necessarily lost as long as an unconscious buffer registers it; even for very brief stimuli, neural persistence—typically lasting about 60ms—will ensure that buffered information remains active until the next conscious snapshot.3
In this discrete model of perception, the brain experiences temporal change naturally by comparing successive snapshots. Is the cloud still there? Has the red light turned green? The temporal resolution of consciousness is then determined by the snapshot rate. Any two events are perceived as distinct if they fall in distinct windows. When they fall in the same window, they are typically perceived as simultaneous. But temporal change can also be experienced directly within each snapshot, as long as dedicated feature detectors exist for the corresponding temporal information. Asynchrony, flicker, and motion can be as short as between 2 and 5ms. The relevant detectors are automatically activated in the buffer, and the information they gather eventually gets integrated into the next 100ms–long snapshot.4 The human brain seems to experience about ten distinct perceptual moments per second.
In their paper, Herzog et al. evaluate both old and new data and suggest that each percept could be much longer, around 400 to 500ms.5 This is not just a marginal shift of timescale, relevant to experts only, but one with drastic implications. If true, it would mean that perception updates about twice per second. Slow. How slow? Think of a typical movie running at 24 images per second: while watching the movie, conscious perception would subsume 12 distinct images and two or three distinct syllables from the dialogue. For the new scheme to be viable, these perceptual episodes would need to be suitably rich and elaborate to encompass every event that happened in the last half-second, including movements, complex objects, words, and melodies.
This would have a striking consequence for decision-making. Experiments suggest that decisions happen within ~300ms of a triggering stimulus.6 If discrete perceptions were refreshed on a 500ms basis, this would imply that most decisions rely on unconscious processes. Consciousness comes in after the fact.7
Before embracing such a harsh prospect, do contemplate the evidence for Herzog’s argument: long-lasting postdictive effects. Postdiction generally refers to a scenario in which the perception of one event is determined, or at least affected, by another event occurring later. It is, both logically and etymologically, the opposite of prediction. Many postdictive effects occur in the range of ~100ms. In the flash-lag illusion, the brain erroneously perceives a continuously moving object about 100ms ahead of its true position at the moment of a flash. Or, in the Fröhlich effect, the initial 100ms of a continuously visible trajectory are hidden from perception. This postdiction window is one important argument in favor of the 100ms perceptual moment.
In a critical experiment, Herzog et al. presented participants with stimulus motion sequences of a vernier offset stimulus: two vertical lines, one higher than the other, on a two-dimensional plane, separated by a small gap and offset by a minute spatial misalignment to the right or left. At key moments of the motion sequence, small transient alterations of this spatial misalignment could occur. The participants were asked to report whether the alignment of component lines was altered toward the left or the right, or, possibly, unaltered. The key finding? Two successive alterations occurring up to 450ms apart were not perceived as individual events, but integrated into one perceived alteration. For intervals larger than about 500ms, the observers correctly registered the two successive alterations. When viewers integrated the events, it was as if the later offset retrospectively modified the entire perceptual experience of the past ~450ms: a long-lasting postdictive effect. It is on this basis that the authors concluded that the temporal window of discrete perception must be much longer than 100ms—at least 450ms.
Once something is seen, common sense suggests, it cannot be unseen. On the contrary. The world is what it is, but what is perceived is only a memory.8 If the memory is replaced, the original vanishes.
Consider this infamous experiment in which scientists tricked participants with a mismatch between their conscious choice and a corresponding outcome.9 The subject sat in front of the experimenter, who displayed a pair of photographs, each with a female face. The task was to point to the most attractive. The experimenter then handed the chosen photo, facedown, to the subject so they could examine it further and explain the reasons for their choice. Unbeknownst to the participants, in some trials the experimenter—in reality, a sleight-of-hand artist—swapped the two pictures so that the subject ended up with the wrong photo. In most of these instances, participants remained unaware of the switch. The initial perceptual event and associated choice had been overwritten.
Indeed, it is common knowledge that memories are shifty and unreliable. Eyewitness accounts provided weeks, months, or years after the fact are not worth two cents.
The postdictive effects described by Herzog et al. abruptly stop when the interval between two events is longer than about 500ms. Those postdictive effects seem to point to a critical interval during which an already formed percept is consolidated in memory, but new events can interfere with consolidation. Outside that critical window, memories may still change, but less often, as in the photo-swapping experiment.
Such a hypothetical consolidation period resembles an interval that has been variably called the central bottleneck, or the psychological refractory period, or even the attentional blink: a 300–500ms window crucial for the consolidation of perceptual information in working memory.10 The fact that perceptual history can still be rewritten during this 500ms–long consolidation period does not mean that no conscious percept was available before. The photo-swapping experiment suggests otherwise. In this light, the observations made by Herzog et al., while compelling, are not necessarily incompatible with the previous estimate of about ten conscious percepts per second. The brain could still form new percepts every 100ms or so, but these may be prone to interference and overwriting for the next 500ms.
The distinction between a ~100ms perceptual moment and a longer, ~500ms consolidation period recalls the distinction between phenomenal and access consciousness.11 Phenomenal consciousness is the immediate experience of sensations, perceptions, thoughts, wants, and emotions. Access consciousness requires further consolidation and is used for reasoning and executive control. A person can report a percept if it is already present in access consciousness, but not all experiences reach access consciousness. The long-lasting postdictive effect in Herzog’s experiment likely taps the access form of consciousness and, consequently, does not constrain phenomenal consciousness. As Herzog points out, “We cannot rule out more complex models in which, for example, fleeting conscious contents are too short for report and memorization.”12
Herzog’s model comprises two successive stages. A first, continuous unconscious buffer is necessary to explain experimental findings of very sharp temporal resolution, on the order of a few milliseconds. In this stage, the unconscious mind detects temporal features such as asynchrony, flicker, and motion. A second stage is required to avoid what the authors call the “too-many-percepts problem,” which would arise if consciousness were continuously active.
This two-stage proposal is more or less the status quo with respect to prior models of discrete perception. The main difference: where we and others postulated a discrete period of about 100ms based on dozens of experimental observations, Herzog et al. now advocate extending this interval five-fold, to nearly 500ms, in order to accommodate a handful of additional findings on long-lasting postdiction.13 This extension appears necessary in the context of the proposed two-stage model, since the duration of the first unconscious stage must be made as long as needed to explain all postdictive effects. Yet this two-stage model comes with a steep cost: most of our decisions would take place during the unconscious stage, and so our free will might just be an illusion.
To reconcile these experimental findings with intuition, a three-stage model, built upon the distinction between phenomenal and access consciousness, is worth considering. This three-stage process has a continuous unconscious buffer to ensure that no inputs are missed—a point in common with Herzog’s models. Discrete phenomenal consciousness is refreshed ten times or so per second, as in previous models of discrete perception; and discrete access consciousness about twice per second, as in Herzog’s new proposal. This model restores causal powers to conscious experience and would allow the conscious mind enough time to influence and control decisions.
