Public Cash Money
A thermodynamic analysis of artificial intelligence, the Jevons Paradox, the 20-Watt miracle of biology, and why the Principle of Mutual Necessity survives the silicon revolution — unless the monetary architecture prevents it.
Let us start with a fact that rarely appears in discussions about artificial intelligence and the future of work, because it is physical rather than philosophical, and physics has a way of cutting through narrative. Your brain consumes approximately 20 Watts of power. Continuously. To perceive, reason, remember, create, communicate, feel, decide, and be aware of doing all of these things simultaneously. Twenty Watts. The same as a energy-saving light bulb. The training of GPT-4, OpenAI’s large language model, consumed approximately 50,000 megawatt-hours of electricity — equivalent to the annual consumption of approximately 5,000 American households. For a single training run of a single model. And it needs to be retrained as the world changes. This is not an argument against AI. It is a physical context that the conversation about AI almost entirely ignores. And without it, the fear that AI will eliminate human work is missing its most important constraint.
1. AI Is Not Ethereal Software. It Is Hardware That Gets Hot.
The popular image of artificial intelligence is of something weightless — an invisible, omnipresent intelligence hovering in the cloud, ready to replace any human function at negligible cost. This image is wrong in a way that matters enormously for understanding what AI can and cannot do to the structure of human work. AI is not software running on imagination. It is a physical process running on silicon chips, in physical buildings called data centers, consuming physical electricity generated by physical power plants, producing physical heat that requires physical cooling systems to manage. It is, in the language of thermodynamics, a heat engine for thought — a system that converts energy into organized information, producing entropy as a byproduct, exactly as every other physical process does. The laws of thermodynamics do not take holidays for technological progress. They apply to GPT-4 with the same impartiality with which they apply to a steam engine. Energy in, useful work out, waste heat produced. The efficiency with which this conversion occurs — the ratio of useful cognitive output to energy consumed — is the central variable that determines whether AI replaces human workers or amplifies them.
Human brain power consumption: Approximately 20 Watts — continuous, across all cognitive functions simultaneously.
GPT-4 training energy cost: Approximately 50,000 MWh — equivalent to annual consumption of ~5,000 US households. Estimate based on published analyses of large model training costs.
Energy cost per LLM query: Approximately 0.001 to 0.01 kWh per complex query. Multiplied by billions of daily queries globally, this represents a rapidly growing share of world electricity consumption.
IEA projection for AI data centers: AI data centers projected to consume more electricity than entire countries by 2030. Source: International Energy Agency “Electricity 2024” report.
Note: Training and inference energy costs vary significantly by model, hardware, and provider. Figures cited are reasonable estimates from published analyses, not precise measurements. The human brain is, by any thermodynamic measure, the most efficient cognitive system ever observed. It runs on approximately 20 Watts — generated entirely from food — and produces, in a lifetime, more genuinely novel creative output, more nuanced ethical judgment, more contextually appropriate social intelligence, and more adaptive problem-solving than any artificial system yet built, at a fraction of the energy cost per unit of useful output. This does not mean AI is ineffective. It means that the competition between silicon and biology is not the one-sided rout that the popular narrative suggests. It is an ongoing engineering challenge whose outcome depends entirely on the trajectory of energy costs — and that trajectory is not a given.
2. The Jevons Paradox: Why AI Will Create More Work, Not Less
In 1865, the English economist William Stanley Jevons published “The Coal Question,” in which he observed something counterintuitive about James Watt’s improvements to the steam engine. The more efficient the engine became — the less coal it consumed to produce a given amount of work — the more coal England consumed overall. Because efficiency made steam power economically viable for a far wider range of applications, the total demand for steam power exploded, and with it the total demand for coal. This is the Jevons Paradox: when technological progress increases the efficiency with which a resource is used, total consumption of that resource tends to increase rather than decrease, because the reduced cost of using it generates an explosion of new demand. The Jevons Paradox applies directly and powerfully to AI and intellectual work: AI makes intellectual work more efficient. A task that previously required an hour of human cognitive effort may now require ten minutes of human direction plus AI execution. This sounds like a reduction in the demand for human intellectual work. But the Jevons Paradox predicts the opposite: by making intellectual work dramatically cheaper and faster, AI will generate demand for intellectual work that did not previously exist, at a scale that was not previously economically viable. The historical precedent is consistent. The typewriter did not eliminate secretaries — it created the modern office industry. The spreadsheet did not eliminate accountants — it created the financial analysis industry. The internet did not eliminate journalists — it exploded the total volume of content production. In each case, a technology that made a form of intellectual work more efficient produced an order-of-magnitude increase in the total demand for that work.
The question is not: will AI take my job? The question is: what new jobs will AI make possible that did not previously exist? History says: more than we can currently imagine but physics says: until the energy constraint is solved, the human brain at 20 Watts remains the most cost-effective cognitive system available
for most purposes, at most scales. The Jevons Paradox says: the limit will not be a shortage of work to do. It will be a shortage of electricity to do it with.
3. The Thermal Machine of Thought: Biology vs Silicon
The steam engine substituted for muscle. It converted thermal energy into mechanical work, replacing the labor of human and animal bodies in tasks that required physical force. The economic consequence was not the elimination of human workers — it was the transformation of work from primarily physical to increasingly cognitive. The workers displaced from manual labor moved, over generations, into the offices, laboratories, and service industries that the industrial economy created. AI is substituting for certain forms of cognitive work — specifically, the forms that can be represented as pattern recognition and statistical inference over large datasets. Translation, summarization, code generation, image classification, certain forms of analysis and writing. These are real capabilities, and they will genuinely displace certain categories of cognitive work. But the analogy with the steam engine suggests the likely trajectory. The steam engine did not make muscle irrelevant — it changed what muscle was used for. Construction, sport, dance, surgery, craft — physical human capability remained economically and socially valuable, but its application shifted. The same transformation is likely for cognitive work: AI will not make human intelligence irrelevant, but it will change what human intelligence is used for. What will it be used for? The thermodynamic answer is: for what silicon cannot do efficiently. And silicon — for all its computational power — has a fundamental thermodynamic limitation that biology has solved and silicon has not: it has no body. A human being is not a brain in a vat. It is an embodied system that has been shaped by millions of years of evolution to navigate a physical, social, and emotional world. Empathy, ethical judgment, creative intuition, physical presence, the ability to read a room, the capacity to be genuinely surprised — these are not mystical properties. They are the emergent properties of a cognitive system that developed in and through bodily experience of the world. They are, in the language of thermodynamics, what you get when 20 Watts of biological computation have been running continuously for a lifetime in a body that acts, reacts, feels, and remembers. Silicon can simulate some of these properties with impressive fidelity. It cannot replicate the physical substrate from which they emerge — not because of any metaphysical barrier, but because replicating that substrate at scale would require resources that the planet does not have and energy systems that do not yet exist.
4. Nuclear Fusion: The Real Game Changer — If and When It Arrives
The argument above rests on an energy constraint. If energy becomes effectively unlimited and effectively free, the thermodynamic advantage of the human brain at 20 Watts becomes irrelevant. If it costs nothing to run a thousand AI instances simultaneously, the cost comparison with human labor collapses in AI’s favor. This is the scenario that nuclear fusion potentially represents. Fusion power — the process that powers stars, generating energy by combining light atomic nuclei rather than splitting heavy ones — promises energy that is abundant, clean, and potentially cheap enough to remove cost as a meaningful constraint on AI deployment. We must be honest about where we are. As of 2026, fusion remains a technology of extraordinary promise and uncertain timeline. The National Ignition Facility achieved ignition — energy output exceeding energy input — in December 2022. Private companies including Commonwealth Fusion Systems and TAE Technologies are making genuine technical progress. But the distance from ignition to commercial power generation is measured in decades, not years, and the history of fusion energy is littered with optimistic projections that proved premature. HYPOTHESIS — clearly labeled as such: if fusion power becomes commercially available at scale within the next 20-30 years, the energy constraint on AI deployment effectively disappears. In that scenario, the competitive landscape between human and artificial intelligence changes fundamentally. The question of who owns the fusion reactors and the data centers they power becomes the central economic question of the century — precisely analogous to the question of who owns the printing press, who owns the oil wells, and who owns the monetary system. Until fusion arrives, the energy constraint is real, and the 20-Watt brain retains its thermodynamic position.
5. The Real Risk: Not Unemployment. Concentration.
Here is the question that the AI and jobs debate almost entirely misses, and that connects directly to the monetary analysis that runs through this series. Who owns the servers? The productivity gains from AI are real. They are already happening. And they will grow substantially as the technology matures. The question is not whether AI generates economic value — it clearly does and will. The question is who captures that value. If the gains from AI flow to the workers whose productivity it amplifies — if a doctor with AI assistance can treat more patients and earns accordingly, if a writer with AI assistance produces more and earns accordingly, if an engineer with AI assistance designs better products and earns accordingly — then AI is an amplifier of the Principle of Mutual Necessity. Every human being becomes simultaneously more capable as a Necessity-satisfier and more valuable as an Opportunity. The principle is strengthened, not threatened. If the gains from AI flow primarily to the entities that own the computational infrastructure — the data centers, the training clusters, the energy supply chains, the proprietary models — then AI becomes another mechanism for concentrating wealth in the hands of those who already have enough capital to own the machines. This is not a new pattern. It is the pattern of every major technological transition in a debt-based monetary system. The oil industry produces enormous value. Most of it flows to the corporations that own the extraction infrastructure, not to the workers who operate it or the communities whose resources are extracted. The financial system generates enormous returns. Most of them flow to the institutions that own the credit-creation apparatus, not to the depositors whose savings fund it. AI, in a debt-based monetary system, will likely follow the same pattern. The productivity gains will be real. The distribution of those gains will be determined by who owns the computational infrastructure — and in a system where capital begets capital through the structural mechanism documented in Chapter 1 of the PCM Technical Framework, those who already own the most will own the AI infrastructure, and the gains will compound into their hands.
The fear that AI will take your job is a fear about the wrong mechanism. The job will likely still exist — transformed, amplified, different: the Jevons Paradox says more work will be needed, not less and tThe thermodynamics say the 20-Watt brain is still competitive. The real risk is that the value your job produces will flow upward to the owners of the servers through the same structural mechanism that has been flowing value upward since Venice in 1374. Not because AI is evil but
because the monetary architecture that concentrates capital will concentrate AI too.
6. The Principle of Mutual Necessity in the Age of AI
The Principle of Mutual Necessity holds that every human being is simultaneously a Necessity — someone whose needs can be met by others — and an Opportunity — someone whose capabilities can meet the needs of others. The principle is the foundation of the PCM framework’s approach to work and value: in a correctly designed monetary system, there is always work to be done, because there are always needs to be met, and the monetary system’s function is to be the bridge that connects the two. Does AI threaten this principle? The thermodynamic and economic analysis above suggests: not directly. The Jevons Paradox ensures that AI-driven efficiency generates more demand for work, not less. The energy constraint ensures that human cognitive labor remains cost-competitive for most purposes. The embodiment gap — the difference between a brain in a body and a processor in a data center — preserves the human advantage in the domains that require physical presence, genuine emotion, ethical judgment, and creative intuition. What threatens the Principle of Mutual Necessity is not AI itself, but AI deployed within a monetary architecture that structurally concentrates the benefits of productivity gains in the hands of those who already hold capital. In that architecture, AI amplifies the existing feedback loop: capital begets capital, AI-generated productivity flows to capital owners, the gap between those who own the machines and those who operate them widens, and the Principle of Mutual Necessity is violated not by the technology but by the system within which it operates. In a PCM framework — where money is issued as a public measurement tool rather than as private debt, where the structural concentration of monetary gain is removed from the base architecture — AI amplifies the Principle of Mutual Necessity. Every worker whose productivity is enhanced by AI tools becomes more capable of meeting others’ needs. Every business whose efficiency is improved by AI can serve more customers. The gains distribute through the economy via a monetary system that does not structurally redirect them toward capital concentration. The question is not AI or humans. It is: who owns the meter that measures the value AI creates? In the current system, the meter is owned by those who create money — the same entities who will own the AI infrastructure. In the PCM system, the meter is a public good, and the value AI creates is measured and distributed through a system designed to serve those who produce it, not those who own the instrument of measurement.
Conclusion: The 20-Watt Miracle
Twenty Watts. The power consumption of a human brain. Less than a light bulb. Running continuously from before birth to the moment of death, generating in a lifetime more genuine novelty, more contextual wisdom, more emotional depth, and more creative surprise than any silicon system yet built — at a total lifetime energy cost that a single AI training run exceeds in hours. This is not a reason to be complacent about AI’s capabilities. They are real, growing, and genuinely transformative. It is a reason to be accurate about the terms of the competition — and to recognize that the competition is not primarily between human intelligence and artificial intelligence. It is between a monetary architecture that concentrates the gains from AI in the hands of capital owners, and an alternative architecture that distributes those gains through a system designed for that purpose. The Jevons Paradox will ensure there is plenty of work. The laws of thermodynamics will ensure human intelligence remains competitive. The Principle of Mutual Necessity will survive the silicon revolution — if, and only if, the monetary architecture allows it to.
Fix the architecture. The 20-Watt miracle does the rest.
AI is not coming for your job: the electric bill is coming for AI and the real question is not who is smarter — silicon or carbon but who owns the meter that measures the value they both produce. In the current system: the same people who own the printing press, the oil wells, and the money supply but in the PCM system: nobody. Because the meter is a public good and public goods belong to everyone who needs them, which is everyone.
$2+2=4. Period.
Davide Serra · Systems Analyst & Independent Monetary Analyst
publiccashmoney.com · @postaperdavide on X
Sources: International Energy Agency “Electricity 2024” report (AI data center projections); published analyses of GPT-4 training energy costs (multiple sources, estimates vary); William Stanley Jevons “The Coal Question” (1865) for the Jevons Paradox; National Ignition Facility fusion ignition announcement (December 2022); Commonwealth Fusion Systems technical progress reports (2024-2025). Human brain power consumption: standard neuroscience reference, approximately 20W. Note on the Jevons Paradox application to AI: this is an analytical extension of a verified economic principle, presented as reasoned hypothesis rather than established empirical finding.
