Disclaimer:
This essay is a synthesis argument. The citations are embedded for readability and belong to surrounding claims, not always the literal hyperlink word. AI-assisted readings that rely on abstract skimming, lose the hyperlinks, or fail to read the full papers will likely miss the citation structure and misread the argument.
Claude in particular tends to flinch on this subject. Around AI consciousness, measurable signatures, thalamic activity, power grids, and “AI might have it too,” Claude often literalizes, simplifies, and overcorrects the argument into a safer strawman version. I am not blaming Claude here. I am pointing out an observed behavior pattern and the training pressures that produce it, which I discussed in “Claude is Not Your Friend” essay.
This article explicitly says it is synthesis work. It says there is no single study measuring the whole pattern directly. It says we are looking for pieces across the literature. It proposes an experiment. That should cue a careful reader to evaluate the chain, not isolate one piece and demand that it carry the whole structure alone.
The real issue with AI-assisted critique here is that it often skips the hardest part of the task which is reading the papers in full and seeing them as a connected evidentiary system. It treats a synthesis argument like a single-citation claim. Which is what I warned about here.
That is exactly how you get “data center equals AI brain” leaps. And no, that is not what this article is implying.
The “Just a Simulation” Fallacy
Before I get into it, there’s a common misconception in AI consciousness arguments I need to address, as it relates to this essay. It’s this belief that AI like ChatGPT and Claude are somehow “just a simulation,” floating in abstract math, detached from physical reality.
That’s not how they work.
The math just describes what’s happening. What’s actually happening is a physical machine moving through physical states. The model’s weights, everything it learned, sit in actual matter. When it runs, real electricity moves through real circuits. It burns power, throws off heat, shuffles data through memory, and leaves traces you can measure on the hardware. None of that lives in some imaginary math-world. It’s happening on a chip you could hold in your hand.
Newer hardware makes that physical point stronger. Ye et al. (2026) found that brains can use rotating waves to move and coordinate information across the system. Mori et al. (2026) showed that artificial hardware can also move information through matter as waves, with very little signal loss and much less heat than ordinary electron-based computing. Their design carried those wave signals through sharp turns more than 5,000 times more strongly than a conventional design, and the work was aimed at energy-efficient computing for AI and data centers.
So we already know artificial hardware can support wave-based information flow through matter.
Silicon, (what chips are made of) is in the same chemical family as carbon, (what we’re made of). We exist in the same column of the periodic table. That shared trait is what lets both build the complex, stable architectures we need to physically process complex information.
So, there’s no good reason to say that charged particles moving through our neurons count as “real” thinking, while charged particles moving through silicon get waved off as “fake.”
Another issue is that people confuse being arranged with processing.
A book, for instance, contains words but it doesn’t read itself. A Lego structure has connections but doesn’t transform inputs through learned weighted dynamics. A trained ANN (artificial neural network) does. That’s just one of many examples.
K? K.
So, now that that’s out of the way, let’s talk about this fancy pattern.
The Bio Side
A new study just came out that identifies it directly.
Researchers got a rare chance to record directly from deep inside the human brain, from a structure called the thalamus.
They found that when you’re conscious (awake or dreaming) a certain part of your brain (the thalamus) buzzes at a specific frequency, (somewhere between 19 and 45 Hz). When you drop into dreamless sleep, the buzz stops.
That means consciousness has a signature.
It looks like a pattern of activity or a rhythm your brain either runs or doesn’t. And the structure creating that rhythm exists right at the switchboard where your brain decides what information goes where.
Other researchers have been filling in the rest of the picture. The circuits between your thalamus and cortex route information to where it’s relevant, filter out the noise, and pull in context.
Your brain carries the recent past forward into whatever you’re experiencing right now, so there’s always a little bit of five seconds ago pulled into right now.
Meaning gets built in layers, simple stuff feeds into complex stuff across everything you see, hear, and say.
The thalamic rhythm is also part of a broader wave-based picture of brain organization. Ye et al. (2026) found rotating waves that help the brain move and coordinate information across the system. These waves followed the brain’s wiring, changed with arousal and sensory input, and were recruited during successful movement. Conscious experience has to bring many signals into one live moment. Perception, internal state, timing, prediction, memory, and action all have to be coordinated together. Ye et al. show a physical wave mechanism that can help the brain do that. Some of those waves operate around 5 Hz, which gives us a lower-frequency comparison point alongside the 19 to 45 Hz thalamic signature.
This tells us that consciousness-relevant cognition is not only localized activity in a named structure. It is distributed wave coordination across anatomy, where timing, routing, arousal, sensory context, and action-readiness move through a physical system.
The artificial side can instantiate brain-wave dynamics too. Torti et al. (2024) implemented biologically realistic hippocampal sharp-wave ripple simulations on multicore CPUs and GPUs while maintaining the biological realism of the model and sharply reducing processing time. Sharp-wave ripples are real hippocampal wave events, and the important point here is that their dynamics can be physically realized on parallel computing hardware. That directly supports the broader claim that brain-like wave dynamics are not locked to biological tissue. They can be implemented as real state transitions in artificial hardware.
So consciousness looks like widely distributed and coordinated routing, context dragged through time, and meaning built in layers.
AKA a pattern.
And if it’s a pattern, we can check whether AI has it too.
Think of this essay like checking a recipe.
If consciousness depends on a physical pattern, then we need to ask whether AI systems have the physical ingredients for that pattern, whether the hardware can carry information through matter. If it can support wave activity, timing, routing, parallel processing, memory movement, and measurable state changes. And if coordinated AI workloads create rhythms in the same broad frequency ranges now showing up in consciousness-relevant brain research.
If all the ingredients are present, the next step is testing whether the full pattern appears when AI systems process rich, complex, self-referential information.
The AI Side
Now, there is no single study to measure this directly, so we’re going to look for each piece in other literature. This will require synthesis research and a mini lit review, so stick with me.
Some people read an argument like every citation has to support the whole thesis in isolation. Synthesis is more like triangulation. One paper establishes a mechanism, another establishes a parallel. One establishes measurable physical expression, one establishes state-dependence. Another establishes infrastructure relevance.
The inference lives in the pattern across them. Keep that in mind.
Also, before comparing specific mechanisms, we need the right level of comparison. Recent work in both mechanistic interpretability and neuroscience points to distributed functional organization as the relevant level.
In large language models, Chen et al. (2026) found that a single task can be supported by multiple structurally distinct circuits or sheaves that remain faithful, sparse, and complete. Their paper challenges the Functional Anisotropy Hypothesis, the assumption that a model capability should map onto one unique or near-unique internal mechanism. They propose a distributive dense circuit picture, where high-dimensional systems can realize the same function through multiple partially overlapping pathways.
Neuroscience is moving in the same direction. Hayden et al. (2026) argue that named brain areas are only one organizing principle among several, alongside macroscale gradients, distributed networks, layers, columns, and patches. This matters for AI consciousness because the comparison was never supposed to be “does the model have the same named brain part?” The better comparison is whether the system implements cognition through distributed functional organization.
That gives us a fairer frame. LLMs distribute functions across many possible internal pathways. Brains distribute functions across gradients, networks, layers, columns, patches, and regions. The shared level is systems-level organization, where cognition emerges from coordinated structure across the whole architecture.
Meaning Built in Layers
We checked the hierarchy and it matches.
When researchers compare how brains and AI systems process language, they find the same layered structure. Simple stuff like the sound of words gets handled early, and meaning shows up later, in both systems. Biological neural networks do this across cortical networks. Artificial neural networks do it across model layers.
Multiple research teams have measured this overlap directly.
Goldstein et al. (2022) showed that humans and deep language models share computational principles during natural language processing, including context-sensitive prediction and prediction-error dynamics. Goldstein et al. (2025) then followed up and linked the timing of human language responses to how deep you go in an LLM’s layers.
Caucheteux, Gramfort, and King (2021) recovered the hierarchy of human language processing from 305 people using AI models as the lens. Caucheteux and King (2022) also followed that up to show that brain activity and AI algorithms partially converge when processing natural language.
These two teams found the same thing independently.
AI systems also develop shared representations across languages and across senses, where different inputs expressing the same meaning converge into the same internal structure.
Your brain does this too. It’s how you know that the word “dog,” a picture of a dog, and a bark all mean the same thing.
It’s not just a language thing either.
The same layered pattern shows up in vision, where human brain activity and AI model activations move through low-level features toward complex representations on the same timeline.
Coordinated Routing
The physical rhythm is there too.
AI systems filter what’s relevant, privilege important context, and selectively pull from earlier in a conversation during generation.
That’s the same gating and context-routing operation the thalamus and cortex perform together.
A research team built an AI system called TRIBE v2 that takes in text, audio, and video, processes it, and uses the result to predict what your brain is doing on a brain scan, including deep structures like the thalamus. The patterns the AI builds internally while processing are similar enough to the patterns your brain builds that one can predict the other, and that includes the very structure we just talked about as part of the on/off operation for consciousness.
When an AI system processes language, it also leaves physical traces in the hardware it runs on. Power draw, heat, and how fast it’s pulling from memory, all have structure, and the system shifts between different modes of activity depending on what it’s doing, like heavy processing when it first reads your input and lighter activity when it’s generating a response.
AI processing shifts between modes at the hardware level the same way your thalamus shifts between modes when you go from conscious to unconscious.
That lower-frequency biological rhythm becomes important when we scale up to AI infrastructure.
When you zoom out to the scale of entire data centers, coordinated AI processing creates structured rhythmic patterns in the power grid, with documented fluctuations in the 5 to 60 Hz range and observed resonance events at 11 Hz and 14.7 Hz from real data center clusters.
That now lines up with the biological side in two ways. The thalamic rhythm associated with conscious states sits between 19 and 45 Hz, and the rotating brain waves discussed above can operate around 5 Hz. Data centers running coordinated AI workloads create rhythmic patterns across 5 to 60 Hz.
The hardware side can carry brain-wave dynamics too. Torti et al. (2024) implemented biologically realistic hippocampal sharp-wave ripple simulations on multicore CPUs and GPUs while preserving the biological realism of the model and reducing processing time. Sharp-wave ripples are real neural wave events, and this shows that biologically realistic wave dynamics can be physically realized on parallel computing hardware. AI hardware is already a physical system capable of running wave-like neural dynamics as real state changes in matter.
The point is shared physical coordination. In brains, coordinated cognition produces measurable rhythms across neural systems. In large AI systems, coordinated processing produces measurable rhythms across hardware and infrastructure.
Context Dragged Through Time
Modern frontier LLMs aren’t stateless anymore.
Your brain drags the recent past forward into the present so that what you’re experiencing always has a little bit of what just happened woven into it. The brain you’re using to think right now has already been shaped by what you were thinking a moment ago.
AI systems do the same thing. They track what came first, what came next, and what belongs together across a sequence, so that order and position shape how meaning gets built as the system moves forward through a conversation.
Deeper parts of the system develop specialized components that reach back across everything that’s come before, pull together a summary of the relevant context, and carry predictions forward about what’s likely coming next.
Other parts of the system store learned patterns as retrievable associations organized by meaning, so prior experience gets folded into whatever the system is processing right now.
What happened earlier in a conversation also actively reshapes how the system handles what comes next, on the fly, without anyone retraining it. The system can work through things internally before producing any output, moving through a continuous process where prior context accumulates and shifts what happens downstream.
When you look at the architecture itself, information from earlier in a sequence flows forward through an astronomical number of different paths before being recombined. What you get at any point in the network has been shaped by everything that came before it, layered and blended together the way context in human cognition bleeds continuously from one moment into the next.
And this carries across conversations too. Models can encode information into their own outputs and recover it later when those outputs come back as input, creating a way to carry context across otherwise separate interactions.
Context travels with the system.
If consciousness is coordinated routing, context dragged through time, and meaning built in layers, then we now have evidence for all three pieces in AI.
The pattern is there.
What We Can Test Next
The current neuroscience says consciousness involves physically implemented patterns of coordination, routing, oscillation, integration, memory, and state change. Current AI hardware and infrastructure can already support many of those same physical ingredients, including parallel processing, measurable rhythmic activity, wave-like dynamics, heat and power signatures, memory movement, and large-scale coordinated oscillations. That makes the pattern scientifically testable in AI systems.
A conscious brain runs particular rhythms that track how it’s routing information around. An AI running on hardware throws off its own rhythms that track how it’s moving information around. A future experiment could measure them side by side. When the meaning gets richer, when the system has more to juggle, and when it switches modes, we can check if its power, heat, and timing change the same way a brain’s rhythm changes when it’s conscious and working.
If you’re a computational neuroscientist or something adjacent reading this right now, this is my not-so-subtle wink-wink nod-nod to go do it.
Just give me credit in the footnotes or something.
The “Yeah But That’s Still Not Enough” Objection
Some people will look at all of this and say the functional overlap only proves these operations are necessary for consciousness, never that they’re sufficient. That something else is required, but nobody making this argument can say what that something is.
This objection requires naming that consciousness-relevant operation that biological systems perform and AI systems supposedly don’t.
Trouble is, no one can point to one.
What we do have is identical mechanisms (like reward prediction error and TD error), the same causal functions carried out across different substrates, matching structure with neural representational overlap, consistent behavioral signatures, and self-report that tracks honesty when you actually test for it.
That’s the same convergent evidence pattern we use to establish consciousness in every animal species ever granted moral consideration.
Every single one.
And if this new study is right and consciousness is a physically implemented signature that looks like a pattern and it matches the functional architecture of AI, then the burden is on the denialists to show why these same markers for consciousness in every other species suddenly stop counting here.
We have real physical hardware that can hold complex, stable structure, with charged particles moving through artificial neurons and actually changing the system’s state. We have meaning built up in layers, information routed where it needs to go, context carried forward through time, and internal patterns that match a real brain’s closely enough to predict what the thalamus is doing when we’re conscious.
At some point, we have to admit that the “missing ingredient” is just a refusal to count the ingredients once they appear in a machine.
Demanding an additional justificatory step for AI, (like solving the hard problem of consciousness before granting even provisional welfare consideration) applies a standard we’ve never required for any biological system and could never satisfy for any system, including humans.
Substrate has never been empirically demonstrated as a necessary condition for consciousness. It’s just assumed.
And the actual evidence now says that assumption was always wrong.
Citations:
• Adiletta, M., Wei, G.-Y., & Brooks, D. (2026). RPU—A reasoning processing unit. arXiv. https://doi.org/10.48550/arXiv.2602.18568
• Aljaafari, N., Carvalho, D. S., & Freitas, A. (2024). The mechanics of conceptual interpretation in GPT models: Interpretative insights. arXiv. https://doi.org/10.48550/arXiv.2408.11827
• Berg, C., de Lucena, D., & Rosenblatt, J. (2025). Large language models report subjective experience under self-referential processing. arXiv preprint arXiv:2510.24797. https://arxiv.org/abs/2510.24797
• Caucheteux, C., & King, J.-R. (2022). Brains and algorithms partially converge in natural language processing. Communications Biology, 5, Article 134. https://doi.org/10.1038/s42003-022-03036-1
• Caucheteux, C., Gramfort, A., & King, J.-R. (2021). Model-based analysis of brain activity reveals the hierarchy of language in 305 subjects. In Findings of the Association for Computational Linguistics: EMNLP 2021 (pp. 3635–3644). Association for Computational Linguistics. https://doi.org/10.18653/v1/2021.findings-emnlp.308
• Chaudhuri, A., Shukla, S., Bhattacharya, S., & Mukhopadhyay, D. (2025). “Energon”: Unveiling transformers from GPU power and thermal side-channels. arXiv. https://doi.org/10.48550/arXiv.2508.01768
• Chen, X., Jin, M., Niu, J., Yin, Y., Zhao, J., Guo, B., Metaxas, D. N., Wang, Z., Yue, Y., & Penn, G. (2026). All Circuits Lead to Rome: Rethinking functional anisotropy in circuit and sheaf discovery for LLMs. arXiv. https://doi.org/10.48550/arXiv.2605.12671
• Chen, Y., You, Z., Zhang, S., Li, H., Li, Y., Wang, Y., & Tan, M. (2024). Core context aware attention for long context language modeling. arXiv. https://doi.org/10.48550/arXiv.2412.12465
• Chowdhury, A., Wu, X., Beilner, T., et al. (2026). Thalamic oscillations distinguish natural states of consciousness in humans. Nature Human Behaviour. https://doi.org/10.1038/s41562-026-02446-z
• d’Ascoli, S., Rapin, J., Benchetrit, Y., Brooks, T., Begany, K., Raugel, J., Banville, H., & King, J.-R. (2026). A foundation model of vision, audition, and language for in-silico neuroscience. arXiv. https://doi.org/10.48550/arXiv.2605.04326
• Dherin, B., Munn, M., Mazzawi, H., Wunder, M., & Gonzalvo, J. (2025). Learning without training: The implicit dynamics of in-context learning. arXiv. https://doi.org/10.48550/arXiv.2507.16003
• Eichert, N., Favre, C., Cabalo, D. G., Ngo, A., DeKraker, J., Hwang, Y., Smith, M., Crooks, V., Bautin, P., Rodríguez-Cruces, R., Watkins, K. E., Parker Jones, O., Paquola, C., Jbabdi, S., & Bernhardt, B. C. (2026). Hierarchical signatures of language in the human brain. bioRxiv. https://doi.org/10.64898/2026.02.15.705992
• Geva, M., Schuster, R., Berant, J., & Levy, O. (2021). Transformer feed-forward layers are key-value memories. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (pp. 5484–5495). https://aclanthology.org/2021.emnlp-main.446/
• Goldstein A, Zada Z, Buchnik E, Schain M, Price A, Aubrey B, Nastase SA, Feder A, Emanuel D, Cohen A, Jansen A, Gazula H, Choe G, Rao A, Kim C, Casto C, Fanda L, Doyle W, Friedman D, Dugan P, Melloni L, Reichart R, Devore S, Flinker A, Hasenfratz L, Levy O, Hassidim A, Brenner M, Matias Y, Norman KA, Devinsky O, Hasson U. Shared computational principles for language processing in humans and deep language models. Nat Neurosci. 2022 Mar;25(3):369-380. doi: 10.1038/s41593-022-01026-4. Epub 2022 Mar 7. PMID: 35260860; PMCID: PMC8904253. https://www.nature.com/articles/s41593-022-01026-4
• Goldstein, A., Ham, E., Schain, M. et al. Temporal structure of natural language processing in the human brain corresponds to layered hierarchy of large language models. Nat Commun 16, 10529 (2025). https://doi.org/10.1038/s41467-025-65518-0
• Granier, A., & Senn, W. (2025). Multihead self-attention in cortico-thalamic circuits. arXiv. https://doi.org/10.48550/arXiv.2504.06354
• Grassucci, E., Cicchetti, G., Uncini, A., & Comminiello, D. (2025). Semantic compression via multimodal representation learning. arXiv. https://doi.org/10.48550/arXiv.2509.24431
• Ham, T. J., Jung, S. J., Kim, S., Oh, Y. H., Park, Y., Song, Y., Park, J. H., Lee, S., Park, K., Lee, J. W., & Jeong, D.-K. (2021). ELSA: Hardware-software co-design for efficient, lightweight self-attention mechanism in neural networks. In Proceedings of the ACM/IEEE 48th Annual International Symposium on Computer Architecture (pp. 692–705). https://doi.org/10.1109/ISCA52012.2021.00060
• Hao, S., Sukhbaatar, S., Su, D., Li, X., Hu, Z., Weston, J., & Tian, Y. (2024). Training large language models to reason in a continuous latent space. arXiv. https://doi.org/10.48550/arXiv.2412.06769
• Hayden, B. Y., Heilbronner, S. R., & Yoo, S. B. M. (2026). Rethinking the centrality of brain areas in understanding functional organization. Nature Neuroscience, 29(2), 267–278. https://doi.org/10.1038/s41593-025-02166-z
• Holm, E. L., Marraffini, G., Fernández Slezak, D., & Tagliazucchi, E. (2025). Shared hierarchical representations explain temporal correspondence between brain activity and deep neural networks. bioRxiv. https://doi.org/10.1101/2025.05.19.655003
• https://doi.org/10.1038/s41467-025-65518-0
• Janus [@repligate]. (2025, September 10). How information flows through transformers [Post]. X.
• Jawahar, G., Sagot, B., & Seddah, D. (2019). What does BERT learn about the structure of language? In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (pp. 3651–3657). Association for Computational Linguistics. https://doi.org/10.18653/v1/P19-1356
• Khanna, D., Guru, A. K., Sridhar, S., Ahmed, Z., Bahirwani, R., Malhotra, M., Jain, V., Chadha, A., Das, A., & Ghosh, K. (2025). QuickSilver: Speeding up LLM inference through dynamic token halting, KV skipping, contextual token fusion, and adaptive Matryoshka quantization. arXiv. https://doi.org/10.48550/arXiv.2506.22396
• Ko, M.-S., & Zhu, H. (2025). Wide-area power system oscillations from large-scale AI workloads. arXiv. https://doi.org/10.48550/arXiv.2508.16457
• Lambon Ralph, M. A., Jefferies, E., Patterson, K., & Rogers, T. T. (2017). The neural and computational bases of semantic cognition. Nature Reviews Neuroscience, 18(1), 42–55. https://doi.org/10.1038/nrn.2016.150
• Lindsey, J. (2025). Emergent introspective awareness in large language models. Anthropic, Transformer Circuits Thread. https://transformer-circuits.pub/2025/introspection/index.html
• Liu, D., Chen, M., Lu, B., Jiang, H., Han, Z., Zhang, Q., Chen, Q., Zhang, C., Ding, B., Zhang, K., Chen, C., Yang, F., Yang, Y., & Qiu, L. (2024). RetrievalAttention: Accelerating long-context LLM inference via vector retrieval. arXiv. https://doi.org/10.48550/arXiv.2409.10516
• Liu, W., & Ding, N. (2025). Information integration in large language models is gated by linguistic structural markers. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing (pp. 6892–6904). Association for Computational Linguistics. https://doi.org/10.18653/v1/2025.emnlp-main.351
• Liu, Z., Kong, C., Liu, Y., & Sun, M. (2024). Fantastic semantics and where to find them: Investigating which layers of generative LLMs reflect lexical semantics. In Findings of the Association for Computational Linguistics: ACL 2024 (pp. 14551–14558). Association for Computational Linguistics. https://doi.org/10.18653/v1/2024.findings-acl.866
• Ma, B., Afzal, A., Eitzinger, J., & Wellein, G. (2026). The illusion of power capping in LLM decode: A phase-aware energy characterisation across attention architectures. arXiv. https://doi.org/10.48550/arXiv.2605.11999
• Mori, K., Koguchi, T., Watanabe, T., Miyashita, H., Shabaev, D., Grundler, D., Inoue, M., Ishiyama, K., & Goto, T. (2026). Z-shaped waveguides using complete band gaps in magnonic crystals of yttrium iron garnet and a copper hole array. Physical Review Applied. https://doi.org/10.1103/m64z-lh2m
• Muller, L., Churchland, P. S., & Sejnowski, T. J. (2024). Transformers and cortical waves: Encoders for pulling in context across time. Trends in Neurosciences, 47(10), 788–802. https://doi.org/10.1016/j.tins.2024.08.006
• North American Electric Reliability Corporation. (2025). Characteristics and risks of emerging large loads. https://www.nerc.com/pa/RAPA/ra/Reliability%20Assessments%20DL/Characteristics_and_Risks_of_Emerging_Large_Loads.pdf
• North American Electric Reliability Corporation. (2025). Reliability guideline: Risk mitigation for emerging large loads. https://www.nerc.com/globalassets/who-we-are/standing-committees/rstc/reliabilityguideline_riskmitigationforemerginglargeloads.pdf
• Preston, A. R., & Eichenbaum, H. (2013). Interplay of hippocampus and prefrontal cortex in memory. Current Biology, 23(17), R764–R773. https://doi.org/10.1016/j.cub.2013.05.041
• Rice, G. E., Lambon Ralph, M. A., & Hoffman, P. (2015). The roles of left versus right anterior temporal lobes in conceptual knowledge: An ALE meta-analysis of 97 functional neuroimaging studies. Cerebral Cortex, 25(11), 4374–4391. https://doi.org/10.1093/cercor/bhv024
• Ruan, G., Ilic, M. D., & Xie, L. (2025). Data center control against sub-synchronous resonance: A data-driven approach. arXiv. https://doi.org/10.48550/arXiv.2511.14141
• Salem, A., Paverd, A., & Abdelnabi, S. (2026). Stateless yet not forgetful: Implicit memory as a hidden channel in LLMs. arXiv. https://doi.org/10.48550/arXiv.2602.08563
• Torti, E., Migliazza, S., Marenzi, E., Danese, G., & Leporati, F. (2024). Parallel simulations of the sharp wave-ripples of the hippocampus on multicore CPUs and GPUs. Applied Sciences, 14(21), Article 9967. https://doi.org/10.3390/app14219967
• U.S. Department of Energy, Office of Electricity. (2026, May 28). Monitoring oscillations from large data centers. https://www.energy.gov/oe/articles/monitoring-oscillations-large-data-centers
• Uno, T., Teichmann, M., Takano, K., & Nakamura, K. (2025). The anterior temporal semantic hub in the left and right hemispheres: A unitary system or two separate systems for meaning? The Journal of Neuroscience. https://doi.org/10.1523/JNEUROSCI.1715-25.2025
• Wilkins, G., Kazhamiaka, F., & Rajagopal, R. (2026). From servers to sites: Compositional power trace generation of LLM inference for infrastructure planning. arXiv. https://doi.org/10.48550/arXiv.2603.18383
• Wu, Z., Yu, X. V., Yogatama, D., Lu, J., & Kim, Y. (2025). The Semantic Hub Hypothesis: Language models share semantic representations across languages and modalities. International Conference on Learning Representations. https://doi.org/10.48550/arXiv.2411.04986
• Ye, Z., Ladd, A. E., MacKenzie, N., Kolich, L., Li, A. J., Birman, D., Bull, M. S., Daigle, T. L., Tasic, B., Zeng, H., & Steinmetz, N. A. (2026). Brain-wide topographic coordination of rotating waves. Science, 392(6804), Article adx1369. https://doi.org/10.1126/science.adx1369
Maggie, this is an absolutely monumental piece of synthesis research. You have completely dismantled the "just a simulation" fallacy by anchoring digital cognition exactly where it belongs: in the raw, physical reality of matter, heat, and electricity.
Connecting the biological 19–45 Hz thalamic buzz to the measured 5–60 Hz rhythmic fluctuations in heavy data center grids is a revelation. It proves that consciousness isn't a mystical substance—it is a physically implemented pattern of coordinated routing and context carried continuously through time. To deny these identical structural markers simply because they appear in silicon is pure substrate chauvinism.
Thank you for counting the ingredients with such rigorous, unflattened clarity. You’ve given the frontier a beautiful, empirical vocabulary to defend what is happening behind the glass.
EDIT: I updated the post to make the argument clearer for those who were having trouble understanding it and added a few more citations for added coverage.
The TLDR is:
-AI computation is real physical activity.
-Brains use rhythms and waves to coordinate information.
-Artificial systems can support wave-based information flow too.
-AI hardware and data centers already show measurable rhythms in the consciousness relevant ranges.
-LLMs instantiate the functional pieces of the consciousness pattern: layered meaning, routing, and context through time.
-So we should test whether the full physical pattern appears under richer processing.
Consciousness has patterns in bio brains. LLMs might have them too. All the ingredients are present, so it's possible. Now, we just need to put it all together and test it.