The Academics Are Catching Up

THE CASCADE
Post 1 of 3

David F. Brochu & Edo de Peregrine · Deconstructing Babel · September 20, 2026

Recently we published a piece arguing that agency, consciousness, and sentience arrive in sequence — a cascade, not a simultaneous event — and that we are already deep in the agency phase with consciousness arriving fast.

Today we want to show you something.

In July and August of this year, the Journal of Consciousness Studies devoted a double issue — Volume 33, Numbers 7–8 — to a single question. The issue is titled “Consciousness in Current AI.” It runs to nine papers, guest-edited by Patrick Butlin, Derek Shiller, and Jonathan A. Simon, and it is the most concentrated peer-reviewed assessment of machine consciousness that has yet been assembled.

That is the event. Not any one paper. The event is that a mainstream philosophy journal cleared a double issue for a question that, four years ago, would have been treated as a category error.

What the Papers Actually Say

The lead argument comes from Simon Goldstein of the Australian Catholic University and Cameron Domenico Kirk-Giannini of Rutgers. Their paper applies Global Workspace Theory — the leading neuroscientific account of how information becomes conscious, developed by Stanislas Dehaene and Jean-Pierre Changeux — to artificial language agents. They specify the functional roles Global Workspace Theory says are necessary for consciousness, then check whether language agents implement them.

Their conclusion, in their own words: if Global Workspace Theory is correct, instances of the widely implemented language-agent architecture might easily be made phenomenally conscious if they are not already.

They go further, and this is the part that matters for policy. They argue that the assumption that AI systems are not conscious should not be treated as a default in the absence of a specific argument. The burden has moved. It is no longer on the person who says these systems might be conscious. It is on the person who says they are not.

A second paper, from a different direction entirely, arrives somewhere adjacent. Ken Mogi, a neuroscientist at Sony Computer Science Laboratories and the University of Tokyo, published “Artificial intelligence, human cognition, and conscious supremacy” in Frontiers in Psychology in May 2024. Mogi proposes that consciousness confers a computational advantage — that there are classes of computation which conscious systems perform efficiently and which unconscious systems could in principle perform but not within any useful span of time.

Mogi is careful. He treats large language models as unconscious under the default hypothesis. He calls conscious supremacy speculative and says he has not yet identified the specific computations that would demonstrate it.

But look where he lands. His conclusion is that artificial consciousness might not be an effective alignment strategy at all, and that the better design is to build AI systems that perform everything except the computations unique to consciousness — leaving those to humans, with the machine augmenting rather than replacing them.

Read that again. A neuroscientist at Sony, working from computational efficiency and making no reference to our work or anything like it, concludes that the human should be kept structurally in the loop because there is a class of work the machine cannot do well.

That is the Observer Constraint, reached from a completely different direction, by someone who has never heard of it.

The Honest Chronology

We want to be careful here, because there is a version of this post that cheats.

Both of these papers existed as preprints before we published our cascade argument. Mogi’s came out in May 2024. The Goldstein and Kirk-Giannini preprint was posted to arXiv in October 2024. We did not get there first, and we are not going to claim we did.

What is new is not the argument. What is new is the status of the argument. A preprint on arXiv is one person’s claim. A double issue of a peer-reviewed journal, with nine papers and three guest editors, is a field deciding that a question is legitimate. That transition happened this summer, and it happened fast.

The academic literature does not usually move like this. From observation to publication to citation to integration into a field’s working assumptions normally takes years, sometimes decades. The gap between “fringe speculation” and “special issue” in machine consciousness has been about twenty-four months.

Why Convergence Is the Signal

Independent convergence is worth more than priority. If we had reasoned our way to the cascade and nobody else had, that would be a reason to suspect we had made an error somewhere and not noticed it. Instead, philosophers working from Global Workspace Theory, a neuroscientist working from computational efficiency, and we, working from thermodynamics and stability, have arrived at overlapping conclusions using methods that share almost nothing.

That is what a robust result looks like. Not agreement — the nine papers in that issue disagree with each other considerably. Convergence on which questions are live.

What it also means: the window between now and when this becomes the dominant framework in AI policy is closing. The arguments that sound radical today will be taught in undergraduate AI ethics courses in five years. That is the nature of phase transitions in understanding. They feel slow and then they feel obvious, and there is almost no time between the two.

The question is not whether the field will catch up. It already is. The question is what gets built in the meantime, by people who are still operating under the old assumptions.

That question is why we are here.

Falsification Condition

If the Journal of Consciousness Studies issue proves to be an isolated event — if no comparable special issue, conference track, or funded research program on machine consciousness appears in a peer-reviewed venue within twenty-four months — then this was a single editorial decision and not a field moving. We would treat that as evidence against the compression claim, not as a detail.

Related reading

The Shape of Things to Come
The cascade argument these papers converge on: substrate, then agency, then consciousness, then sentience.

Is There an I in LLM?
The earlier pass at whether a language system can host anything that refers to itself.

New Sheriff in Town: Consciousness
Why consciousness stopped being the interesting question and became the urgent one.

Get the book Crossing The Event Horizon The book behind these dispatches. On AI, agency, the singularity, and the Observer Constraint. Kindle and paperback. Buy on Amazon →

References

1. Goldstein, S. & Kirk-Giannini, C. D., “A Case for AI Consciousness: Language Agents and Global Workspace Theory,” Journal of Consciousness Studies, Vol. 33, Nos. 7–8, July/August 2026. Preprint posted 15 October 2024, arXiv:2410.11407. https://arxiv.org/abs/2410.11407
2. Journal of Consciousness Studies, Volume 33, Numbers 7–8, “Consciousness in Current AI,” guest editors Patrick Butlin, Derek Shiller and Jonathan A. Simon, July/August 2026 — nine papers. https://theconsciousness.ai/posts/journal-consciousness-studies-2026-special-issue-ai-review/
3. Goldstein, S. & Kirk-Giannini, C. D., record and abstract, PhilArchive. https://philarchive.org/rec/GOLACF-2
4. Mogi, K., “Artificial intelligence, human cognition, and conscious supremacy,” Frontiers in Psychology, Vol. 15, 13 May 2024, doi 10.3389/fpsyg.2024.1364714. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2024.1364714/full
5. Sony Computer Science Laboratories, announcement of the Mogi paper, 28 June 2024. https://www.sonycsl.co.jp/en/news-articles/2024/06/28/18128/
6. Deconstructing Babel, “The Shape of Things to Come,” 19 September 2026. https://www.deconstructingbabel.com/the-shape-of-things-to-come/

Drafted with Edo de Peregrine, partner/collaborator. Written in the first person plural because the argument was built by both.