multi-turn inc.
INDEX

index

Does Consciousness Arrive in Frames?

Brain rhythms, perceptual thresholds, and a present composed by events

When you watch a car pass outside a window, its motion does not seem to break apart. The engine sound approaches from a distance, comes closer, then fades again. Vision and hearing form a single continuous scene. As experience, the world flows.

But when perception is measured in the laboratory, phenomena appear that do not fit this smooth impression. A faint flash of the same brightness may be seen at one moment and missed only tens of milliseconds later. Early neural responses to a stimulus change gradually, but a participant's report suddenly switches at some point to "seen." Sometimes information that arrives later can even alter what was just perceived.

These results revive an old question. Is consciousness truly continuous? Or is it composed of a sequence of frames, like film?

Research so far does not answer this question with a simple either-or. The brain does not seem to contain a single shutter that slices every experience at the same speed. Instead, different temporal structures overlap. Neural activity keeps changing. Sensory selection fluctuates with brain rhythms. The process by which information becomes available for report and action has thresholds. Perceptual content is integrated over short periods and reorganized into a new state when an important change occurs.

The key, then, is not to choose between "continuous" and "discrete." It is to distinguish which levels are continuous and where transitions occur.

The consciousness discussed here is mainly the process by which visual content becomes available for report, working memory, planning, and action - that is, access consciousness. Phenomenal consciousness, experience itself, is harder to measure directly. Experiments usually record indirect indicators such as button presses, verbal reports, memory performance, or brain signals. If we ignore this limitation, it becomes easy to mistake discreteness in behavior for discreteness in experience itself.[1]

1. What does "discrete" mean?

When discussing the discreteness of consciousness, four different questions need to be separated.

The first is the temporal form of neural processes. Do membrane potentials, firing rates, and interactions between brain areas change continuously at every moment, or are they updated only at specific moments?

The second is the periodicity of sensory selection. Does the brain process all inputs with equal strength, or does it amplify some information and suppress other information according to a rhythm?

The third is the threshold of access. Does evidence about a stimulus accumulate gradually and then, once it crosses a certain level, enter report and working memory abruptly?

The fourth is the event structure of content. Does current experience remain briefly stable, then reorganize into a new state when an object changes, a goal shifts, or a prediction fails?

These four questions are related, but they are not the same. The fact that behavioral reports divide into "seen" and "not seen" does not mean that the preceding neural process was binary. The fact that detection performance fluctuates periodically does not mean experience disappears between cycles. The fact that memory is segmented at event boundaries does not mean the brain uses fixed-interval frames.

Separate observation from interpretation
"Perceptual performance changes with rhythm" is an experimental observation. "Consciousness turns off in the troughs of the rhythm" is a stronger interpretation. The former is supported by multiple studies. The latter has not yet been established.

With this distinction in mind, the evidence about the temporal structure of consciousness begins to fit into one picture rather than contradicting itself.

2. Are brain rhythms the shutter of consciousness?

The brain is a rhythmic organ. Oscillations such as alpha and theta are linked to information exchange between sensory cortex, parietal cortex, and frontal cortex. These rhythms are not just background noise. The phase of an oscillation just before a stimulus appears can predict whether a near-threshold visual stimulus will be seen, and when attention is divided across multiple locations, detection performance has been observed to rise and fall alternately in the range of several hertz.[2-4]

In the experiment by Landau and Fries, participants monitored which of two locations might contain a target. After a cue was presented, detection performance at the two locations did not remain equal. When sensitivity increased at one location, it decreased at the other, suggesting that attentional priority alternated rhythmically.[3] Work by Fiebelkorn and colleagues in monkeys showed that this rhythm may involve dynamic interactions in the frontoparietal network rather than an isolated oscillation in a single sensory area.[5]

At first glance, these findings make it tempting to think that the brain samples the world periodically instead of perceiving it continuously. But what was measured here was not the existence or absence of experience. It was processing efficiency. The same stimulus is more likely to be detected when neural excitability and attentional gain are high, and more likely to be missed when they are low.

A camera shutter completely blocks light between exposures. Brain rhythms are closer to a dimmer. Input keeps arriving, but the influence of particular information becomes stronger at some moments and weaker at others. In the simplest terms, sensory evidence itself may change continuously while the weight with which that evidence contributes to access changes periodically.

The frequencies observed in perception and attention are also not singular. Visual temporal resolution can be related to an individual's alpha frequency, while attention switching among multiple objects can appear in the slower theta range.[2,4] The dominant rhythm changes with sensory modality, task, and number of attended objects. If there were a single frame rate governing consciousness as a whole, a shared fundamental period should repeatedly appear across different tasks. The current evidence fits better with the view that multiple circuits operate on different time scales.

Analyses that search for periodicity in behavioral data also require caution. A single transient response caused by a stimulus, or temporal expectation, can look like a repeated oscillation. The evidence becomes stronger when a frequency peak is accompanied by a relationship between prestimulus phase and trial-level behavior, agreement across measurement methods, and causal perturbation experiments.

Rhythm may be one element that organizes the time of consciousness. But it is not, by itself, the frame of consciousness.

3. Why does the moment of seeing arrive suddenly?

When the brightness of a faint stimulus is increased little by little, subjective reports often do not rise smoothly. For a while the stimulus is almost never seen, then within a narrow range the proportion of "seen" reports rises quickly. This abrupt transition suggests that conscious access may have a threshold.

Backward masking is a standard way to study this phenomenon. If a target is shown very briefly and then immediately covered by another stimulus, the early visual system may respond to the target even though the participant cannot report seeing it. In the MEG study by Del Cul, Baillet, and Dehaene, unreported targets were still processed in early occipito-temporal pathways. But in trials where the target was consciously reported, later activity involving a wider cortical network appeared after roughly 270 milliseconds.[6]

This result shows how a continuous process can produce a sudden output. Sensory evidence accumulates over time and can be amplified through recurrent interactions. If the evidence is insufficient, processing remains early and local. If it passes a certain level, the information becomes widely available for working memory, verbal report, planning, and action selection.

Think of water slowly rising until it spills over a bank. The change in water level is continuous, but the overflow looks like an event. In the same way, the internal neural state may change smoothly while the availability of content changes abruptly.

There is an important caveat. We cannot simply equate later widespread activity with consciousness itself. To report what they saw, participants must maintain the content, make a decision, and prepare a motor response. Later signals include these task demands. We therefore need to separate whether "global ignition" is the cause of experience, the result of report, or a mixture of both.

The 2025 adversarial collaboration by the Cogitate Consortium confronted this issue directly. The researchers preregistered predictions that Integrated Information Theory and Global Neuronal Workspace Theory made differently, then tested 256 participants using fMRI, MEG, and intracranial EEG. Information related to conscious content was observed not only in occipital and ventral temporal regions but also in some frontal regions, and some posterior cortical responses persisted while the stimulus was present. Yet strong predictions from both theories were partly contradicted.[7]

This makes it difficult to say that consciousness is produced only by a single global explosion. Access may involve a nonlinear transition, but content itself can be sustained across multiple areas. In other words, transition and persistence are not mutually exclusive.

4. The present is not a point, but a short temporal window

We feel "now" as if it were an instant, but the present used by perception is not a mathematical point. The brain binds information arriving over a short period into one event.

This becomes clear in postdictive construction. Information that arrives tens or hundreds of milliseconds after a stimulus can change the final perception of that earlier stimulus. The later event does not travel back in time and alter the earlier neural response. Rather, it is more natural to say that perception of the earlier stimulus had not been fully settled from the start. The brain temporarily maintains several interpretations and uses later cues to construct a result.

Herzog, Kammer, and Scharnowski proposed the "time slices" model to explain this phenomenon. Feature analysis proceeds with high temporal resolution in a quasi-continuous way, but conscious perception is formed through a period of integration.[8] In this model, what we experience is not the raw signal at every instant, but content compressed from processing over a short interval.

When listening to music, a note does not have meaning in isolation from the notes before and after it. Motion direction also requires at least two positions across time. A word in language is interpreted differently depending on preceding context and following words. Temporal integration is not an exceptional feature of consciousness. It is a basic condition for perception to construct meaning.

But an integration window is not the same as a set of non-overlapping frames. Temporal windows can overlap, and their length can differ by information type. Motion, speech, sentences, and social events require different time scales. Continuous recurrent dynamics or gradual probabilistic updating can also explain postdictive effects. Indeed, critiques of discrete perception theories point out that current experiments support temporal integration, but do not prove the existence of gaps without experience between frames.[9]

So what we can say at this stage is limited but clear. Perception is not completed immediately at the moment of input. The content of the present includes a short past, and its length is not fixed by one universal number.

5. Why is experience divided into events?

Even when we watch a continuous scene, memory does not store every moment at the same density. If someone enters a room, picks up a cup, pours water, and leaves, we do not remember it as an infinite sequence of postures. We divide it into events: "entering," "picking up the cup," "pouring water," and "leaving."

Event Segmentation Theory holds that the brain maintains an internal model of the current situation and updates that model when prediction fails substantially.[10] When an action goal changes, space changes, a new person appears, or the causal structure shifts, an event boundary is formed. These boundaries later shape the units of memory.

Event boundaries provide an important clue for thinking about the discreteness of consciousness. If major changes in conscious content occur not at fixed intervals but when the information structure changes, then the time of consciousness is organized more by events than by clocks.

These boundaries are not fixed. Even while watching the same video, a person learning a recipe and a person tracking a character's emotions may treat different moments as important. Events are nested across multiple scales. Reaching for something is a short event. Preparing a meal is a longer event. The social context of serving that meal is longer still.

This hierarchy and task dependence do not fit well with a fixed-frame hypothesis. They instead support a picture in which multiple scales of quasi-stable states form on top of continuous neural processes, and state transitions occur when prediction error and goal relevance become large.

6. A model that explains continuity and transition together

Computational models of the temporal structure of consciousness can be divided into three broad types.

In a continuous model, content changes little by little with input. In a fixed-frame model, the entire content is updated at regular intervals. In a hybrid model, underlying processing proceeds continuously, but stable content transitions into a new state when an event occurs.

The current evidence is most economically organized by the third model. Sensory input and neural state keep changing. Attentional rhythms modulate the gain of specific information. Recent information is integrated within a short temporal window. When prediction error, salience, or goal relevance becomes large enough, current content is reorganized into a new quasi-stable state.

Computationally, this can be summarized as follows.

dz/dt = f(z(t), x(t))
 
c(k + 1) = B(z[t_k - tau : t_k])
when g(z(t), x(t), m(t)) > theta

Here z(t) is the continuously changing neural state, and x(t) is sensory input. m(t) is a modulatory variable such as attention, goal, or prediction error. When the function g crosses the threshold theta, information from the recent temporal window is organized into a new content state c(k + 1).

The important point is that update times t_k are not regular. A fixed-frame model is the special case where t_k = kT. In an event-based model, a calm and predictable scene can maintain one state for a long time, while a rapidly changing scene can trigger several updates at short intervals.

This model explains both why consciousness can look frame-like and why it actually feels continuous. Content is stable for a certain period, so it is experienced as one scene. Underneath, however, sensory evidence and internal state keep changing. When new information becomes important enough, the state transitions quickly. What is discrete is not the existence of the neural process, but the reorganization of content.

We cannot conclude that the hybrid model is correct. But it is the most economical working hypothesis because it can place rhythmic selection, nonlinear access, temporal integration, and event segmentation within one structure.

7. AI is not evidence of consciousness, but a computational testbed

When bringing artificial intelligence into consciousness research, the first task is to draw a boundary. The fact that a model compresses information, remembers, or separates events does not mean it has subjective experience. There is no agreed experiment for determining phenomenal consciousness in current AI systems, and similarity in computational function does not guarantee identity of experience.[11]

Even so, AI is useful for testing the computational components of consciousness theories separately. In the brains of humans and animals, selection, integration, memory, and action happen together. In artificial systems, these functions can be implemented and removed separately. We can directly compare which bottlenecks are necessary, whether update cycles need to be fixed, and whether it is advantageous to remember only at event boundaries.

DeepMind's Perceiver iteratively compresses large sensory inputs into a small latent array.[12] Not all input participates in deep computation with equal weight. A limited latent space selectively takes in information. This architecture shows that a limited workspace can coexist with continuous processing. The presence of a bottleneck does not require input processing to be frame-based.

Allen Institute for AI's MERLOT Reserve learns video, language, and sound together to infer temporally connected event structures.[13] Rather than storing the continuity of pixels and audio as-is, the model forms representations useful for predicting what happened first and what will happen next. It is a computational example of a continuous stream being compressed into event-level internal representations.

DeepMind's Differentiable Neural Computer and Neural Episodic Control show another separation of time scales.[14,15] One selectively reads from and writes to external memory, while the other separates slowly changing neural network weights from rapidly updated episodic memory. This means sensory processing, current state, event memory, and long-term learning can move at different speeds.

In embodied-agent benchmarks such as ALFRED, past actions change the environment, and those changes become conditions for the next action.[16] An agent cannot finish a task by classifying only the current screen. It must track, at the event level, what it has already picked up, which door is open, and which goal has been completed.

These models do not explain consciousness. But they turn the computations that consciousness theories implicitly require into explicit components. They let us measure what performance advantages are provided by limited bottlenecks, event segmentation, selective memory, and multiple time scales.

The most direct experiment would compare three agents under the same data and compute budget. The first updates its state at every moment. The second updates only at a fixed interval. The third updates when prediction error or goal relevance crosses a threshold. If long-term prediction, event memory, action success rate, and computational cost are measured together, we can see which temporal structure is adaptively useful.

Such experiments do not answer whether AI is conscious. They answer a narrower and testable question: "What temporal structure is needed to build functions that resemble consciousness?"

8. What would distinguish the three models?

The continuous model, fixed-frame model, and event-based hybrid model can explain many current results after the fact. To distinguish them, we need experiments in which each model can fail.

First, local neural evidence and global access must be separated within the same trial. Early sensory representations may change smoothly with stimulus strength, while working memory and reportability may switch abruptly at a particular moment. If this pattern repeats, the explanation of nonlinear access on top of continuous dynamics becomes stronger.

Second, elapsed time and event boundaries must be manipulated independently. Even when the same amount of time passes, predictable scenes should produce fewer state transitions, while scenes with changing causal structure should produce more. If updates align with the clock, the fixed-frame model gains support. If they align with prediction error, the event-based model gains support.

Third, conditions that require report must be compared with conditions that do not. Signals related to conscious content need to be separated from report preparation, working memory, and motor planning. If late frontal activity increases mainly in report conditions, it is difficult to interpret it as a marker of experience itself.

Fourth, phase and causality matter more than frequency alone. Finding periodicity in behavioral performance is not enough. We need to see whether aligning stimulus timing to the phase of a particular brain rhythm, or perturbing that rhythm with noninvasive stimulation, changes the temporal structure of perception as predicted.

Fifth, models must be compared by prospective prediction, not explanatory plausibility. After fitting the three models to the same data, we should evaluate how accurately each predicts state transition timing and reports in unseen trials. Consciousness research has many plausible narratives. What separates models is prediction on new data.

Conclusion: What may be discrete is updating, not existence

Current evidence does not support the idea that consciousness is a film played at a fixed speed. No universal clock has been found that binds every sensation and thought to a single frame rate. But it is also difficult to say that every aspect of consciousness changes smoothly. Perceptual sensitivity is rhythmic. Access has thresholds. Content is constructed over time. Event boundaries can rapidly reorganize what is present.

So the question "Is consciousness discrete?" must be answered level by level. Neural dynamics are mostly continuous. Information selection can be rhythmically modulated. Access to report and working memory can be nonlinear. Conscious content can remain quasi-stable and then update according to events. But whether experience itself completely disappears between those states remains unknown.

The most conservative conclusion is this:

Consciousness is less a sequence of fixed frames than a system in which flowing neural processes are organized into one content state for a short time, then move into a new state in front of important events.

From this perspective, a "moment" is not the smallest particle of consciousness. It is the way a continuous process briefly gains a stable form.

References

  1. Block, N. (1995). On a Confusion about a Function of Consciousness. Behavioral and Brain Sciences, 18(2), 227-247. doi:10.1017/S0140525X00038188
  2. VanRullen, R. (2016). Perceptual Cycles. Trends in Cognitive Sciences, 20(10), 723-735. doi:10.1016/j.tics.2016.07.006
  3. Landau, A. N., & Fries, P. (2012). Attention Samples Stimuli Rhythmically. Current Biology, 22(11), 1000-1004. doi:10.1016/j.cub.2012.03.054
  4. Busch, N. A., Dubois, J., & VanRullen, R. (2009). The Phase of Ongoing EEG Oscillations Predicts Visual Perception. Journal of Neuroscience, 29(24), 7869-7876. doi:10.1523/JNEUROSCI.0113-09.2009
  5. Fiebelkorn, I. C. et al. (2018). A Dynamic Interplay within the Frontoparietal Network Underlies Rhythmic Spatial Attention. Neuron, 99(4), 842-853.e8. doi:10.1016/j.neuron.2018.07.038
  6. Del Cul, A., Baillet, S., & Dehaene, S. (2007). Brain Dynamics Underlying the Nonlinear Threshold for Access to Consciousness. PLOS Biology, 5(10), e260. doi:10.1371/journal.pbio.0050260
  7. Cogitate Consortium et al. (2025). Adversarial Testing of Global Neuronal Workspace and Integrated Information Theories of Consciousness. Nature, 642, 133-142. doi:10.1038/s41586-025-08888-1
  8. Herzog, M. H., Kammer, T., & Scharnowski, F. (2016). Time Slices: What Is the Duration of a Percept? PLOS Biology, 14(4), e1002433. doi:10.1371/journal.pbio.1002433
  9. Fekete, T., Van de Cruys, S., Ekroll, V., & van Leeuwen, C. (2018). In the Interest of Saving Time: A Critique of Discrete Perception. Neuroscience of Consciousness, 2018(1), niy003. doi:10.1093/nc/niy003
  10. Zacks, J. M. et al. (2007). Event Perception: A Mind-Brain Perspective. Psychological Bulletin, 133(2), 273-293. doi:10.1037/0033-2909.133.2.273
  11. Butlin, P. et al. (2023). Consciousness in Artificial Intelligence: Insights from the Science of Consciousness. arXiv:2308.08708.
  12. Jaegle, A. et al. (2021). Perceiver: General Perception with Iterative Attention. Proceedings of ICML 2021, PMLR 139, 4651-4664.
  13. Zellers, R. et al. (2022). MERLOT Reserve: Neural Script Knowledge through Vision and Language and Sound. Proceedings of CVPR 2022.
  14. Graves, A. et al. (2016). Hybrid Computing Using a Neural Network with Dynamic External Memory. Nature, 538, 471-476. doi:10.1038/nature20101
  15. Pritzel, A. et al. (2017). Neural Episodic Control. Proceedings of ICML 2017, PMLR 70, 2827-2836.
  16. Shridhar, M. et al. (2020). ALFRED: A Benchmark for Interpreting Grounded Instructions for Everyday Tasks. Proceedings of CVPR 2020, 10740-10749.