Mechanical Thought, Perpetual Motion and The Age of Consequences

 

Hype Too Cheap to Meter

Artificial intelligence promised to solve everything by granting knowledge, expertise and wisdom to all.

Those already with knowledge and expertise may abhor it, while those without embrace it with abandon. In this way, it can be seen as a great equaliser. Regardless, almost everyone ends up using it in the end.

The rhetoric surrounding AI has encompassed everything from: digital super intelligence, singularities, event horizons to the creation of God.

 

This is how the singularity goes: wonders become routine, and then table stakes … Intelligence too cheap to meter is well within grasp. —Sam Altman, The Gentle Singularity, June 10, 2025

 

It was also believed that the machines would actually create knowledge and wisdom out of nothing through recursive training and self-improvement.

 

The mechanism whereby I imagined it would happen is that we would make models that were good at coding and good at AI research, and we would use that to produce the next generation of models and speed up to create a loop that would increase the speed of model development. —Dario Amodei, Anthropic CEO, 2026

 

We must accept that those in positions of authority actually believed their own rhetoric. Since they presumably believe there will be no need for endeavour or expertise in the future, here are two prominent AI business leaders apparently saying goodbye to those who created everything for them.

 

I have so much gratitude to people who wrote extremely complex software character-by-character. It already feels difficult to remember how much effort it really took. Thank you for getting us to this point. —Sam Altman, writing on X, March 2026

 

And:

 

So imagine a non-arrogant programmer who actually does what you want and you don’t have to pay all that money to. And there’s an infinite supply of these programmers. —Eric Schmidt, Ex-Google CEO, Stanford University Q&A, 2024

 

In his recent speech at the University of Arizona, Schmidt urges graduates not to surrender their agency by being afraid, an accusation now deployed against anyone not sold on the idea. Nevertheless, under the impression his audience will be pleased, he goes on to outline how AI is going to do all the things that they have just studied for.

My own view is that all talk of a singularity will be on its way to a memory hole rather than a black hole.

 

Fifth Generation Collapse

To be fair, I was once vulnerable to this kind of hype and recall similar aspirations surrounding what was once called Fifth Generation Computers, as reported by computing magazines in Britain in the 1980s. I particularly remember a depiction of a digital hand reaching out to touch God. I have, therefore, been here before in a sense but realise many now have not.

In the past, however, such dreams were outside the domain of ordinary people—forever 10 or 20 years away—and when they crashed and burned, impacted relatively few. The difference now is that we are in the age of consequences where the delusions of those with massive technological and institutional power have catastrophic effects for everyone else—not in 20 years time, but in the now.

In 2024, a paper published in Nature coined the term “model collapse” [1]. This describes the slow degradation of large language models, followed by collapse, as increasing amounts of previous synthetically generated content is re-ingested as training data for newer models.

The process can be likened to making an audio cassette recording of an LP, and then making successive cassette copies of copies while believing that you can do this forever. In reality, you only get one or two copies before the results are no longer worth listening to.

Synthetic content was always going to outcompete that of humans which requires time, diligence, skill and effort. Hence, a synthetic flood was always going to be unavoidable.

Degradation and collapse, however, were not anticipated.

Rather, it was assumed that synthetic data could be used for recursive endless self-improvement, or at least would not be harmful.

 

From here on, the tools we have already built will help us find further scientific insights and aid us in creating better AI systems. Of course this isn’t the same thing as an AI system completely autonomously updating its own code, but nevertheless this is a larval version of recursive self-improvement. —Sam Altman, The Gentle Singularity, June 10, 2025

 

But where would the underlying information actually originate?

I am not suggesting that AI cannot be used, or is not being used, in some fields to discover such things as security vulnerabilities. But here, the operable word is “discover,” not invent. The information must already exist.

The idea that AI can be used to amplify human thought and agency has been a common trope on social media. But it also amplifies something else too, as we will discuss.

In any case, the actual solution to model collapse proposed by the paper could only be that of the technical one—that synthetic content must be filtered from human and what it calls “trustworthy” sources.

But what happened to the recursive self-improvement? What’s wrong with the synthetic? And just what is trustworthy and what is not?

The paper acknowledges the difficulty and suggests a focus on data from “education domains”. However, a separate study also from 2025 suggested that, in the computer sciences at least, some 22% of papers already showed signs of synthetic generation [2]. At the same time, physicist Sabine Hossenfelder argued that AI slop is spreading in scientific literature and may overtake it.

The paper published in Nature is framed in terms of data errors when compared against reality. This is reductionist thinking, namely that things can be analysed in isolation. But just what is reality when there is now a feedback loop to the models, such that the models shape reality just as reality shapes the models?

It must be stated, however, that the quality of AI output has nevertheless improved somewhat in the last year or so. This is certainly my own experience. Enthusiasts may well claim that the problem has been solved.

It is interesting to note that the paper also states:

 

…the value of data collected about genuine human interactions with systems will be increasingly valuable in the presence of LLM-generated content in data crawled from the Internet.

 

So human beings are still needed after all?

I do not agree that the problem is solved. Rather, I think there is an increasing focus on draining people of yet more information without acknowledgment of its value, the effect of which is necessarily to degrade their experience and lives. I do not believe this is widely understood.

If you do not see how it is draining people, then consider the writer or artist whose works remain legally theirs but become worthless.

I do not agree that there is a technical “fix” for this.

 

Perpetual Motion of the Second Kind

Having had some past exposure to artificial intelligence, but being an outsider to modern developments, I no longer share the mindset that dominates the technical and professional world. I have moved on from the thinking I once had.

Back in 2022, I was shocked by AI as much as anyone else. However, it was immediately apparent to me that the models would be used to flood the Internet, and the fruits of that would feed back to the models. I saw this in terms of a trend toward informational thermodynamic equilibrium, and I was basing my thinking on what I knew of Information Theory. As a programmer with a degree in physics, I was aware of such things, but do not claim “expertise.”

I saw the promises of AI to be those of perpetual motion—not in some metaphorical sense as in how the term “singularity” has been deployed—but quite literally. I was surprised that no one else seemed to pick up on this.

Mathematician, Claude Shannon, laid the foundations of what is sometimes referred to as “classical information theory” in 1947 in his influential work, A Mathematical Theory of Communication. His theory derived from an understanding of entropy and the physics of thermodynamics. His insight was to define information as negative entropy.

Simply put, if we consider entropy to be a measure of disorder, we may therefore consider negative entropy to be that which describes order.

Earlier, Erwin Schrödinger, had proposed the idea that entropy had a wider significance beyond thermodynamic utility:

 

How would we express in terms of the statistical theory the marvellous faculty of a living organism, by which it delays the decay into thermodynamical equilibrium (death)? We said before: ‘It feeds upon negative entropy’, attracting, as it were, a stream of negative entropy upon itself, to compensate the entropy increase it produces by living and thus to maintain itself on a stationary and fairly low entropy level. — Erwin Schrödinger, What is Life, 1944

 

Shannon’s theory is classical in the sense that it is incomplete in the same way Newtonian mechanics is incomplete when compared with relativity. It is also reductionist in the sense that it assumes that information, as represented by a sequence of probabilities, merely exists in and of itself and thus may be considered in isolation from the outside world. For this reason, Shannon information is sometimes referred to as self-information.

Shannon theory is an engineering theory and has utility that works within limitation, just as Newtonian mechanics works within limitation. It can be used, for example, to determine the maximum capacity of an electronic communication channel and explains how you can zip a computer file to get smaller one, but cannot keep doing this to arrive at a file of zero size without losing all the information.

What Shannon theory does not consider, however, is that information itself behaves as a conserved quantity (with a caveat [3]) in the same way as energy behaves as a conserved quantity.

Prior to both Shannon and Schrödinger, Leo Szilard had demonstrated that to create order in one place, you must dump disorder in another. Szilard fled the Nazi regime in 1933, but had earlier published an important work: On the reduction of entropy in a thermodynamic system by the intervention of intelligent beings. [4]

In the introduction, he writes:

 

The objective of the investigation is to find the conditions which apparently allow the construction of a perpetual-motion machine of the second kind, if one permits an intelligent being to intervene in a thermo-dynamic system. When such beings make measurements, they make the system behave in a manner distinctly different from the way a mechanical system behaves when left to itself. —Leo Szilard, 1929

 

With the result being:

 

We show that it is a sort of a memory faculty, manifested by a system where measurements occur, that might cause a permanent decrease of entropy and thus a violation of the Second Law of Thermodynamics, were it not for the fact that the measurements themselves are necessarily accompanied by a production of entropy.

 

The implication is that you cannot create information (negative entropy) from nothing just as you cannot create energy from nothing—hence the connection to perpetual motion.

 

The Domain of the Unscientific

Despite the revelations of the early 20th century, the dominant thinking of our culture remains grounded in materialistic and rational working assumptions. Scientist and philosopher, Stephen Meyer, argues that modern science has become so entrenched in materialism that ideas not rooted in its assumptions are not perceived to be scientific at all.

While it may be considered valid to apply information theory to material stuff and to the transmission of encoded signals, the same is no longer accepted for such things as knowledge and consciousness because these are considered to be philosophy, and therefore outside the scientific domain.

You will note that Szilard’s thinking from 1929 involved an “intelligent being” that was capable of intervening in mechanical processes. It seems, however, that immediately after the Second World War, science became afraid of the implications of where things were heading.

Consciousness cannot be accounted for in physical terms. For consciousness is absolutely fundamental. It cannot be accounted for in terms of anything else. —Erwin Schrödinger, The Observer, 11 January, 1931

When modern information physicists speak of “information”, they are invariably speaking of “physical information.” This is information referring to the structure of physical stuff, since when we look closely, the sense of material solidity evaporates leaving behind only informational structures. However, it excludes any concept of a mind that would either create or perceive the information.

Although science has progressed since the 1800s, we are left with the working assumption that things must remain causal and rational. Since materialism assumes that all existence is formed from atoms in motion obeying deterministic laws that simply exists, physical information must therefore be likewise.

In materialism all things are assumed to be causal and, in principle, measurable and knowable. This is considered good because it informs rational thinking and rational thinking is considered the best kind of thinking. Therefore, everything must be measured so that it can be thought about and best managed.

Where things are not knowable in practice because it is impossible, difficult or merely inconvenient, then statistics, models and key performance indicators (KPIs) serve as substitutes. In this way, the sense of rationality is preserved.

Whether or not such proxies meaningfully reflect reality is another question.

 

Vegetative Electron Microscopy

It may be instructive, if not disturbing, to consider what thermodynamic equilibrium in human beings might look like:

Hot Skull is a Turkish TV drama from 2022. It presents a dystopian future in which the majority of the population have succumbed to a condition in which they lose all ability to think for themselves, but exist in a vegetative state while vocalising meaningless but grammatically correct speech known as “jabber.”

This condition is infectious with the main character, a linguist called Murat, being the only one to be immune (others must wear ear protectors).

What’s interesting, if not bizarre, is that things are portrayed as though there has been a virus outbreak, something easily relatable to viewers nowadays. However, there is no actual mention of any biological connection in the show whatsoever.

The condition merely manifests on exposure to jabber, which looks like:

 

I hope you don’t sacrifice the instinct for friction to your mass of matter.
In my neighbourhood making the coffee of champions for today.
Tumble like a trout behind the sacrificial slide. If the trout is blue bronze, I’m as bright as a knob. —Jabber, Hot Skull (Sıcak Kafa, 2022, Netflix)

 

I’m not warning that things are about to play out as portrayed in some fictional TV drama. Nevertheless, it is interesting to note that Hossenfelder, in her video linked above, references a particular example of “vegetative electron microscopy” —a meaningless term that has suddenly appeared in over 20 scientific papers.

Rather, I am suggesting that, at a societal level, the system degrades and then things begin to break down. Events may be nuanced and complex. It is a common observation that with online connectivity, for example, comes polarisation and bubbles rather than the harmony that was envisaged in the early days of the World Wide Web.

 

The Limits of Nihilism

Since information technology began to dominate all human existence, everything became  about “scaling” —the ability to grow bigger with minimal effort. The lessons stemming from when the Internet was rolled out to everyone on the planet were learned in the form of IPv6, which provides enough internet addresses for every atom in the sun with more than enough left over for the planets just in case.

Such scaling has only ever benefited those already powerful for whom no limits are assumed—just the next even bigger thing or, as Altman puts it, “Wonders become routine, and then table stakes.”

However, I suggest that artificial intelligence represents the limit of such nihilism.

It is not a technical limit with a technical fix, but a deeper fundamental one—one that perhaps could not be readily perceived using mechanical thinking and by those lacking a sense of humility.

It is this: Automation cannot reduce entropy, only keep it the same or increase it.

For the rest of us, perhaps will we be better served by revisiting older ways of understanding things.

For myself, I like to think of The Tower of Babel as being when, believing we have the knowledge of God, we build faster than we can manage the entropy the construction process creates.

 

And the whole earth is of one pronunciation, and of the same words… 

and Jehovah saith, `Lo, the people [is] one, and one pronunciation [is] to them all, and this it hath dreamed of doing; and now, nothing is restrained from them of that which they have purposed to do. 

Give help, let us go down, and mingle there their pronunciation, so that a man doth not understand the pronunciation of his companion.’ —Genesis 11 (YLT)

 

I’m not suggesting no one should ever build anything, only that we build with care for others rather than in arrogance. Arrogance is merely the dumping of disorder onto others in the form of liability and consequences, whereas care is its conscious minimisation.

Nor do I suggest that the understanding that came out of the Enlightenment should be abandoned. I do not think this. Rather, I am suggesting that mechanical rationality alone has reached the limit of where it is functional, and therefore science requires a Reformation.

 

_____________________________
[1] Shumailov, I., Shumaylov, Z., Zhao, Y. et al. AI models collapse when trained on recursively generated data. Nature 631, 755–759 (2024). https://doi.org/10.1038/s41586-024-07566-y

[2] Liang, W., Zhang, Y., Wu, Z. et al. Quantifying large language model usage in scientific papers. Nat Hum Behav 9, 2599–2609 (2025). https://doi.org/10.1038/s41562-025-02273-8

[3] So where does information actually come from? Contrary to expectations of heat death, cosmologist David Layzer proposed that the total amount of information in the universe is increasing, rather than decreasing. This does not violate the Second Law and is possible through the expansion of space itself. Ref: David Layzer, Cosmogenesis: The Growth of Order in the Universe (New York: Oxford University Press, 1991).

[4] Szilard, Leo. 1929. “Über die Entropieverminderung in einem thermodynamischen System bei Eingriffen intelligenter Wesen.” Zeitschrift für Physik 53: 840–856. English translation: “On the Decrease of Entropy in a Thermodynamic System by the Intervention of Intelligent Beings,” trans. Anatol Rapoport and Mechthilde Knoller, Behavioral Science 9, no. 4 (1964): 301–310.

Table of Contents