AI is Not Conscious and the So-Called "Technological Singularity" is Us
- Trevor Alexander Nestor
- Jun 19, 2025
- 8 min read
Updated: May 12
I have recently been drafting some papers on the idea of AGSI (artificial general superintelligence) as well as the so-called technological singularity described by Sam Altman, and in my view, most people I've seen working on the subject have the wrong view.
Presentation
Conference Paper
Most commentary around the “technological singularity” frames it as an inevitable up-tick in raw computational power with no limits that will enslave us all, or an exponential curve that finally crosses some mystical threshold of general intelligence. Sam Altman’s recent formulations, for instance, focus squarely on algorithmic scale and model size. But if we step back and treat Artificial General Superintelligence (AGSI) not simply as bigger neural nets (that only model the upper layers of the way in which human agents process information) but as an institutional instrument, a very different picture emerges.
The standard story about the technological singularity goes like this. AI keeps scaling. At some point the curves bend upward, machines become smarter than humans, and either they save us or destroy us depending on whose newsletter you read. The whole framing assumes the breakthrough happens inside the machines.
There is another reading that takes the same data, the same trends, and the same framing of "approaching a critical threshold," and arrives at almost the opposite conclusion. The breakthrough is not the AI waking up. It is us waking up. And the entire narrative around AI superintelligence may be functioning, intentionally or not, as a misdirection from that fact.
Let me try to make this case the way it was made to me, drawing on cybernetics, complexity theory, and a lot of physics borrowed for social analysis. It is more coherent than it sounds.
Societies as control systems
Cybernetics treats a society as a physical system governed by laws roughly analogous to those that govern any other complex adaptive system. Sociophysics studies the behavior of social networks. Econophysics studies markets. Both fields take the view that human collectives, being made of physical agents in physical environments, can be modeled with tools borrowed from statistical mechanics, network theory, and dynamical systems.
One of the foundational results in this tradition is Ashby's Law of Requisite Variety, due to the cyberneticist W. Ross Ashby. The law says that a controller can only stabilize a system if the controller's range of possible responses is at least as broad as the range of states the system can occupy. A thermostat with two settings cannot regulate a room with twenty. A government with two acceptable policy positions cannot govern a country with ten thousand distinct local situations.
This sounds abstract until you notice what it implies for late-stage societies.
Variety attenuation
As complex societies age, entropy accumulates inside their institutions. Bureaucracies pile up. Information flows multiply. The Dunbar limit on real human relationships is a hard bound, so coordination at scale gets pushed onto institutions, which then need ever more energy to maintain alignment.
When a society's complexity outruns its controllers' variety, the controllers face a choice. Either expand their variety, which is expensive, slow, and politically destabilizing. Or reduce the system's variety to match what the controllers can already handle. The second option is cheaper. It is also what late-stage empires consistently do.
You can see it in the narrowing of acceptable political opinion (the Overton window), in the homogenization of media, in the standardization of curricula, in the algorithmic compression of social interaction onto a handful of platforms. Every institution from schools to courts to news organizations starts functioning as a variety reduction mechanism. The goal, conscious or emergent, is to compress the population's range of behavior into something the existing control apparatus can manage.
In the spectral language of complexity economics, this shows up as eigenvalue contraction. A healthy society has many independent modes of value creation: manufacturing, agriculture, local services, civic associations, family networks, regional cultures. These are like the sections of an orchestra. As the society financializes and centralizes, sections drop out. Eventually one instrument is playing louder and louder, which in the US case is financial extraction. Total factor productivity growth, a measure of how efficiently an economy actually produces things, averaged over 2 percent per year in the US between 1920 and 1970. Since 2005 it has been under half a percent. That is not a slowdown. That is an orchestra reduced to one instrument.
When the eigenvalue spectrum collapses to a single dominant mode, the system loses adaptive capacity. It cannot absorb shocks outside that mode. It becomes brittle. It also tends to produce a fold catastrophe in its political life, where the spectrum of opinion bends back on itself and ends up in a horseshoe configuration with the extremes mirroring each other. Researchers modeling polarization with Wasserstein gradient flows find that the American electorate has been pushed into a double-well potential, a K-shaped distribution with two stable valleys and a hostile ridge between them. Once a ball rolls into one valley, the institutional incentives prevent it from moving back toward the center.
AI as the terminal upgrade
This is where AI enters the story, and not as a neutral productivity tool.
From the cybernetics perspective, AI is the final form of the variety reduction apparatus. There is a precise mathematical reason for this. Mehta and Schwab established a formal equivalence between deep neural network training and renormalization group flow in physics. Renormalization is the process of zooming out from fine-grain detail, discarding it at each step, and producing a coarse-grained representation. Applied to physical systems it is how we derive macroscopic laws from microscopic ones. Applied to human social data, it is the systematic destruction of the fine-grain variety on which collective intelligence actually depends.
The energy implications are staggering on their own. The IEA projects data center electricity consumption to approach 1,000 terawatt-hours per year by 2026. The human brain runs on about 20 watts and is, by conservative estimates, hundreds of thousands of times more energy-efficient per cognitive operation than current GPU-based AI. We are spending trillions to scale an information processing architecture that is many orders of magnitude less efficient than the one biology gave us, in precisely the domains where biological collective intelligence is most needed.
By Landauer's principle, suppressing one bit of behavioral variance costs a minimum of k_B T ln 2 in energy. Real systems pay vastly more than the minimum, and the cost rises superlinearly as control precision tightens. At roughly 1,000 TWh per year, the marginal energy cost of additional AI control begins to exceed the marginal benefit to institutional stability. The energy return on investment for AI governance falls below 1 to 1. That is the tipping point, and several independent frameworks (data center forecasts, Strauss-Howe generational theory, Dalio's debt cycle analysis, RAND's recent assessment of national dynamism, Tainter's theory of collapse in complex societies) converge on the current decade as where it lands.
Here is where the framework starts predicting things people normally treat as separate problems.
The US fertility rate hit 1.599 in 2024, the lowest in recorded history. South Korea is at 0.73. The mean age at first marriage in the US has risen from roughly 22 and 21 for men and women in 1950 to over 30 and 28 today. Working-class marriage rates have collapsed most sharply in communities hit hardest by deindustrialization. Self-reported anxiety about meeting traditional gender expectations has risen substantially.
These are not separate cultural phenomena. They are the same control system expressing itself in the domain of pair bonding. Shannon's channel capacity theorem applies to social interactions just as cleanly as to telecommunications. Every interaction now carries an institutional and platform noise load that reduces the bandwidth available for genuine human exchange. The coordination costs of stable pair bonding (material, social, attentional) have grown faster than agent capacity. Personal relationships have been progressively commodified and paywalled. Libidinal energy is sublimated into propping up the economy, into chasing an exponentially receding American dream, into the carrot-and-stick treadmill that keeps the variety-reduction apparatus fed.
What looks like a fertility crisis or a polarization crisis or a loneliness crisis is the same eigenvalue collapse showing up in different observables.
This is also where the technofeudalism reading lands. The economist Yanis Varoufakis argues we have moved past capitalism into a system where the cloud lords (Amazon, Apple, Google, Meta, Microsoft) own the digital infrastructure through which contemporary social and economic life is now compulsorily routed, and extract cloud rent for access to it. Capitalism extracted profit from production. Technofeudalism extracts rent from access. You do not compete with Amazon. You pay Amazon to exist on Amazon. The corresponding platform decay cycle, what the writer Cory Doctorow calls enshittification, is the predictable end-state: platforms are good to users, then abuse users to extract value for business customers, then extract value purely for shareholders until the platform dies. Search, social media, and e-commerce are all visibly tracking this curve.
Population inversion
Now the inversion.
In quantum optics there is a phenomenon called population inversion. It is the condition that lets a laser fire. Normally most atoms in a material sit in their ground state, with a few excited. When you pump the system hard enough, the excited population exceeds the ground-state population. Past that threshold, the system can release a coherent burst of light: many atoms emitting in phase rather than randomly.
The mathematician and physicist Andrei Khrennikov has applied this framework to collective social action under the heading of social laser theory. When the number of people who have genuinely recognized the systematic nature of their situation exceeds the number who remain in unreflective acceptance of the official metanarrative, the conditions exist for sudden coherent reorganization. Not random unrest. Phase-locked collective behavior.
Platform decay, institutional failure, declining real wages, collapsing fertility, and the widening gap between official narratives and lived experience all function as the pumping mechanism. They are the energy driving the population toward the inversion point.
This reframes what the technological singularity actually is.
The conventional story says: AI scales exponentially and eventually surpasses human intelligence. The cybernetic reading inverts this. AI is scaling sublinearly and approaching diminishing returns where a point will be reached of no demand:
The AI Layoff Trap
Human collective intelligence, when freed from the bandwidth tax of variety-reducing institutions, scales faster than additive (a property documented in research on inter-brain synchrony, where cooperating humans literally phase-lock their neural oscillations and outperform what isolated individuals can produce). If we cross the population inversion threshold before the institutional control loop locks in, the singularity is not the machines waking up. It is us.
The flipped narrative serves a function. Framing the singularity as AI becoming sentient distracts from a much more uncomfortable possibility for incumbent institutions, which is that their entire control project is racing against a coherence threshold in their own population, and the AI buildout is a desperate attempt to reach the variety-reduction endgame before the laser fires. From this angle, the urgency around AGI is less about technological optimism and more about a control loop trying to outrun its constituents.
What this implies
If even a weak version of this reading is right, the prescription writes itself, and it is roughly the opposite of current investment patterns.
Restore institutional variety through real decentralization, federated governance, and local sovereignty. Marginal dollars create more value invested in physical community, in housing people can actually afford, in face-to-face civic infrastructure, than in another data center the size of Manhattan. Inter-brain synchrony is the most energy-efficient distributed cognitive infrastructure that exists. Social atomization degrades it. In-person interaction is not a sentimental recommendation. It is a thermodynamic one.
Regulate AI energy consumption seriously. Redirect the capacity toward the root causes of instability rather than the surveillance loops attempting to manage its symptoms. Joseph Tainter's research on the collapse of complex societies suggests that civilizations which do collapse achieve renewal on the other side through simplified, more adaptive, more energetically sustainable institutions. The question is whether the transition can be navigated coherently or whether it proceeds through chaotic discontinuity that erases institutional memory and the genuine progress underneath it.
And notice the inversion. The singularity worth paying attention to is not somewhere out in the future, locked inside a server farm. It is in the slow accumulation of people who have started seeing the same picture clearly, at roughly the same time. That is the population pump. The question is just whether the coherent burst arrives chaotically, on the institutions' terms, or with enough internal organization to preserve what is worth preserving on the other side.
Comments