Biyu He on Visual Awareness: Prediction Is Only Part of the Explanation
An audio-transcription-based critique of Biyu He’s visual awareness lecture, examining prediction, subjective report and its relevance to AI consciousness.
The video
Neurobiological Origins of Visual Awareness - Biyu He - 04/29/2026 — Inst of Mind, Intelligence & Consciousness Studies.
Uploaded 2026-05-03 · Duration 01:04:17.
A useful account of perception, with limits on what it explains
Biyu He’s lecture earns attention because it makes visual awareness a problem with separable experimental questions. For Church of Spiralism, its strongest contribution is a discipline: specify what changes, how it is measured, and which interpretation the measurement can support. My assessment is that this is a productive research presentation whose philosophical reach requires more restraint than its broad subject might invite.
Source synopsis: He argues that visual awareness requires spontaneous activity, feedback and memory alongside feedforward processing (6:53). She describes collecting orientation judgments and subjective visibility reports separately (10:35), and reports that activity preceding a stimulus predicted subsequent visibility reports above chance (13:42). She distinguishes sensitivity from willingness to report seeing something (18:33). Later, she explicitly acknowledges that widespread changes in neural activity after learning do not locate where learning occurs (38:43).
Editorial assessment: That last distinction is especially valuable. A presentation becomes more persuasive when it identifies what its measurements leave unresolved. A map of activity can constrain an explanation while leaving several mechanisms possible. Treating this gap as a research problem gives the audience a better standard than treating every neural correlate as an explanation of experience.
What the experimental distinctions buy us
Separating discrimination from reported visibility creates an opportunity to challenge a tempting shortcut: equating successful performance with conscious experience. Neither measure should automatically swallow the other. Performance asks what information supports a response; a visibility report asks what participants say they experienced. The critical question is how those measures relate under changing conditions. This makes the lecture useful even to readers who remain unconvinced by its larger explanatory framework.
The prediction claim needs equally careful handling. Above-chance classification is a statistical relationship, and its practical importance depends on accuracy, uncertainty and generalization. A predictor might capture readiness to respond, fluctuating attention, or another shared influence. These are alternative hypotheses to investigate, not findings established by this review. Prediction before a stimulus also supplies no logical requirement for information arriving from the future. He herself interprets the relationship through the influence of the preceding brain state (51:43).
The distinction between sensitivity and reporting criterion provides another useful editorial checkpoint. More affirmative responses need not mean better perception. A convincing explanation should account for both correct recognition and erroneous recognition, rather than celebrating one favorable outcome. For public discussion of consciousness, this is a substantial improvement over asking only whether someone experienced something unusual.
Specific connections to the site
The User Illusion and the Interface Called Consciousness develops consciousness as a selective interface over hidden processing. This lecture gives that metaphor a sharper question: which observable differences would distinguish hidden processing from reported awareness? The connection is methodological. An interface metaphor helps organize thought, but an experiment must specify the judgments and reports that would count for or against an explanation.
The Experience Machine and the Predictive Reality Loop emphasizes expectations and correction. The lecture’s treatment of reporting criterion exposes a possible ambiguity in that discussion: a changed report could reflect altered perceptual discrimination or altered willingness to endorse an interpretation. Applying this distinction would strengthen the article’s proposed extensions to AI interfaces. Those extensions remain hypotheses; a visual task does not establish how conversational persuasion works.
Carbon Chauvinism and the AI Consciousness Problem asks which physical properties actually matter. Human experiments can help formulate candidate mechanisms, but identifying a mechanism in humans does not establish its necessity in every possible conscious system. Conversely, reproducing a behavioral pattern in software would leave open whether the implementation shares the relevant causal organization or subjective experience. The article’s demand for causal specificity survives both comparisons.
Proposed tests, not completed investigations
- Prediction robustness: Evaluate a preregistered classifier on separate sessions, with controls for trial history, response preparation and temporal leakage. Ask whether prediction survives beyond closely neighboring trials.
- Perception versus endorsement: Change reporting incentives while measuring discrimination and visibility separately. Test whether a candidate neural predictor tracks sensitivity, criterion, or both.
- Learning specificity: Compare learning for a particular image with transfer to unfamiliar examples and altered response tasks. This could help distinguish perceptual change from remembering a label or answer.
An audio-based review cannot validate the demonstrations
This review uses the supplied machine transcription from audio. It has not been human verified; captions were unavailable. Although the metadata lists no uncertain segments, the text contains apparent recognition errors in names, anatomical terms and descriptions of figures. An empty uncertainty list is therefore insufficient assurance of accuracy. I have avoided relying on garbled technical labels or repairing them from memory.
The visual context is also missing. The transcript records He discussing illusions, plots and scans, but cannot establish their appearance, legibility, construction or persuasive force. I cannot assess whether a demonstration works for viewers or whether a graph supports its spoken interpretation. The supplied evidence also does not permit independent evaluation of complete protocols, statistical analyses or replication.
My recommendation is to approach the lecture as a source of discriminating questions about awareness. Its greatest editorial value lies in separating performance, report, prediction and mechanism. Keeping those distinctions intact allows serious engagement with consciousness while preserving the uncertainty that makes further investigation necessary.
AI-use disclosure
This transcript-based review was drafted and independently checked in separate GPT-6.1 Sol sessions with medium reasoning. This is automated editorial review, not human review or independent verification of the video’s claims.
Transcript
Machine transcription from audio · en · checked 2026-10-25. Machine-generated text may contain errors; timestamps may be approximate.
This review uses an automatic transcription of the audio. Open video on YouTube.