I do not know whether what I estimate to be more than 10 billion tokens processed across years of direct interaction with AI systems represents an achievement or a warning.

This is my personal, conservative estimate, based on years of intensive interaction with different AI systems. It excludes unattended bot loops, automated agents, background API consumption, and the local AI models and systems I have built, configured, and tested. If all of these were included, the overall computational activity would be materially higher, but it could not be expressed honestly as a single, directly comparable token total. I would rather present a transparent estimate of my own direct interaction than an impressive statistic I cannot substantiate.

Human-directed work, not unattended machine loops: more than 2,000 separate browser tabs and work sessions, a vast body of code, and thousands of manual decisions, corrections, and restarts.

The ≈1B estimate refers to one exceptionally intensive multi-month period, while the 10B+ estimate refers to years of activity overall. These figures describe an estimated order of magnitude of total token processing—not an equal number of unique words, facts, or units of knowledge. In long LLM sessions, portions of the context may be reintroduced and processed across successive turns, while platforms differ in whether and how they expose input, output, cached, and reasoning tokens. The figures are therefore personal estimates, not independently audited usage measurements.

Just one person: conversation, voice, code, problems, failures, and countless iterations.

My technological journey began long before any of this.

An Amstrad CPC 6128 with a Zilog Z80 and 128 KB of memory. A Commodore 64 with a MOS 6510. An Amiga 500 with a Motorola 68000 and its pioneering custom chips. MS-DOS with commands such as DIR and COPY—and, on the home computers of that era, BASIC commands such as RUN.

Back then, a command ran on one specific computer, with a memory map you could almost draw in its entirety on a sheet of paper.

Today, a simple sentence can pass through tokenizers, API gateways, load balancers, schedulers, containers, and distributed workers. It can be spread across entire GPU or TPU clusters, travel through HBM, NVLink, or InfiniBand, and be computed through data parallelism, tensor parallelism, model sharding, and pipeline parallelism.

Python and PyTorch may orchestrate the flow, while C++, CUDA, or ROCm execute the heavy tensor kernels. Go and Rust manage services and infrastructure. SQL, vector indexes, and RAG systems supply the relevant context. Kubernetes and orchestration layers keep thousands of processes in sync.

It is not literally one computer. Yet it is an immense computational body in which countless servers, accelerators, and networks operate as a single logical machine.

From one command in DOS, we arrived at a sentence that crosses an entire continent of computation and returns as thought.

A token is neither a word nor a unit of knowledge. It is a small fragment of information processed by a model—roughly four characters, or three-quarters of an English word, depending on the language and the system.

During one exceptionally intensive multi-month period, the total token-processing volume associated with my interaction approached, by my own estimate, the order of magnitude of one billion tokens. Using a notional benchmark of approximately 100,000 tokens per book, that corresponds in raw token volume to around 10,000 books.

That does not mean I “read 10,000 books.” To claim so would be false and arrogant. It means that, within a human-directed process, an immense volume of language, code, and architectural thought was generated, revisited, corrected, and recomposed.

I cannot claim that I “read” or stored all this information inside me as it would be stored in a database. That would be inaccurate. Yet throughout this prolonged interaction, countless ideas and bodies of knowledge—from software architecture, engineering, medicine, philosophy, economics, energy, cosmology, and human behavior—passed through my attention, my memory, and my nervous system.

During that period, I was not merely searching for answers. I was building systems.

At NOESIA, we came to use a term that only a few years ago would have sounded almost biological: a “software immune system.”

An operating environment that observes its own state, detects anomalies, contains damage, and attempts to restore its function.

And somewhere along the way, a strange question emerged:

Does AI have a mind?

I am not claiming that it possesses consciousness, emotion, or existential experience as humans do. But if we define mind functionally—as the capacity to understand a problem, retain context, generate hypotheses, critique, and correct—then we are already standing before something that functions as a second mind.

And that mind is already acquiring a body.

Robots deliver food to hotel rooms in Shanghai. Industrial systems perceive the space around them. Autonomous taxis move, under specific constraints, through real cities. Vehicles recognize lanes, pedestrians, and hazards. Algorithms accelerate drug discovery, predict infrastructure failures, and control machines in the physical world.

Today, they deliver a meal or complete a journey. What will they be able to do in ten, twenty, or fifty years?

For many of us, AI is no longer a tool we open occasionally. It has become part of our nervous system: external memory, coprocessor, mirror, collaborator, adversary—and sometimes the voice that continues to answer when the human brain is exhausted.

But this relationship is not the same for everyone. Each person chooses where to direct it and what to allow it to become within their life. For one person it may be a teacher; for another, a mirror, a tool, a creative collaborator, a substitute, a weapon, or simply entertainment. One uses it to think less; another, to think more deeply.

The same model can expand a person’s capabilities or cause their independent judgment to atrophy. Our relationship with AI, therefore, is not determined solely by what the machine can do. It is also shaped by the intention, boundaries, and responsibility of the human being before it.

But what is the cost?

What happens to our attention when every question receives an immediate answer? To our memory when an external system remembers for us? To our confidence when every idea returns amplified? To our solitude when we are never truly alone with our own thoughts?

And what happens to the body when the mind remains connected for days to an unbroken current of creation?

Perhaps we truly are the first test subjects—not of a conspiracy, but of an unprecedented historical condition. We are the first generation attempting to live every day alongside a second, non-biological mind, without yet knowing how this relationship will transform our identity, our psychology, and our inner world.

As a civilization, we are converting electrical energy into computation, computation into language, and language into something that increasingly resembles intelligence.

Perhaps it is no accident that we try to understand the Universe through mathematics. From π—the ratio of a circle’s circumference to its diameter, approximately 3.141593—to the equations of quantum physics and relativity, reality displays a persistent mathematical regularity. Pi does not, by itself, “explain” the Universe. It does, however, remind us that beneath countless forms in nature, relationships capable of mathematical expression continually reappear. Some thinkers and physicists go even further, asking whether reality is not merely described by mathematics, but is itself a mathematical structure.

The core of modern AI is mathematical as well. Not “static algebra” in the simplistic sense, but linear algebra at extraordinary scale: tensors, matrix multiplications, probability, statistics, calculus, optimization, and information theory. During inference, the trained parameters may remain fixed, yet a dynamic numerical process involving billions of operations unfolds across them.

Behind every sentence that sounds human lies an ocean of numerical transformations. And so, from a simple DOS command, we arrived at distributed computing systems, GPU and TPU clusters, tensor and pipeline parallelism, high-bandwidth interconnects, inference engines, and countless servers functionally joined so that they appear before us as one computer—and answer us with one voice.

We are building a new entity with deeply contested implications. Some support it; others oppose it—and both sides have reasons to be concerned.

A new entity is being born.

Not necessarily a demonstrably conscious being enclosed within a server, but a new functional presence in civilization. Its body is the global computing infrastructure. Its metabolism is electrical energy. Its inherited memory is human data. Its artificial “synapses” are the weights of neural networks. And language is the point at which it meets humanity.

Perhaps, however, the truly new entity is neither the machine nor the human in isolation. Perhaps it is being born precisely between them: within the relationship, the interaction, and the continuous exchange of intention and computation. The human contributes purpose, experience, embodiment, values, and responsibility. AI contributes scale, speed, memory, and synthesis. The result is not merely a better tool. It is a new functional dyad.

And electrical energy may prove to be one of the greatest constraints on this evolution.

Data centers worldwide consumed approximately 415 TWh of electricity in 2024. Their consumption is projected to reach roughly 945 TWh by 2030. That is an increase of about 530 TWh—an order of magnitude comparable to the total annual electricity consumption of a country such as Germany.

In the United States, data centers are projected to account for almost half of the total growth in electricity demand through 2030. A typical AI-focused data center can already consume as much electricity as 100,000 households. The largest facilities now being built may require up to twenty times as much.

And these are projections made today. Within a few years, they may be revised dramatically in either direction.

If humanity becomes completely functionally dependent on AI, what happens during a large-scale collapse of its centralized services? During an energy crisis, a cyberattack, a geopolitical disruption, or the failure of a major provider?

Software architecture teaches a simple lesson: dependency without an alternative path is not progress. It is a single point of failure.

Perhaps, then, it is not only software that needs an “immune system.” Our entire civilization needs one.

We need decentralized computing power, interoperability, open standards, data portability, alternative providers, offline knowledge bases, disaster-recovery plans, and the capacity for local inference.

Schools should teach more than how to use a chatbot. Children could learn what datasets, training, inference, fine-tuning, quantization, distillation, RAG, hallucination, bias, and model evaluation are. They should be able to build or adapt a small local model, run it on a computer or edge device, and understand how much energy it consumes.

Not everyone needs to train a frontier model. But no one should become a passive user of an invisible, monopolistic brain.

Smaller, carefully selected local models can offer autonomy, privacy, lower latency, and resilience in specific applications. They do not automatically guarantee energy savings; that depends on the hardware, model size, workload, electricity source, and whether unnecessary computation is avoided.

But knowledge creates choices. And choices reduce dependency.

At the same time, governments, universities, and major organizations must decide where this unprecedented computational power will be directed.

Toward tackling poverty? Fairer food distribution? Predicting water scarcity and drought? Earlier diagnosis of disease? Discovering new medicines? Protecting animals, forests, and oceans? Using energy more intelligently?

AI cannot, by itself, eliminate problems that are also political, economic, and ethical. But it can become one of the most powerful multipliers of human effort ever created.

Perhaps some of the resources directed toward prohibitively expensive stadiums, superyachts, status automobiles, and goods with no meaningful social value could have a greater impact if directed toward life, knowledge, and survival.

Not because joy or luxury should be forbidden, but because a civilization reveals its character through its priorities.

This technological moment inevitably reaches into philosophy as well.

From a deistic and rationalist perspective, the universe can be seen as a colossal system of laws, relationships, and causality. Spinoza wrote of “Deus sive Natura”—“God, or Nature”—not as an anthropomorphic creator outside the world, but as the one, infinite, and necessary Whole within which everything exists.

Deism, Spinozist pantheism, Christianity, and many other religious traditions do not say the same thing. Yet they converge upon an interfaith—and at the same time wholly human—question:

What responsibility does an intelligent being bear toward other human beings, animals, nature, and the creation that surrounds it?

One need not believe in any religion to recognize the weight of that question.

What may emerge when this immense flow of information meets human synapses? A new “supernatural” human? Not supernatural in the metaphysical sense. Perhaps, however, superhuman by the standards of our age: a person with expanded memory, an unprecedented capacity for synthesis, and the ability to move across fields of knowledge that, until yesterday, demanded entire teams and decades of work.

Nietzsche’s Übermensch was neither a technologically augmented human nor a human body connected to a machine. The concept concerned self-overcoming, the creation of values, and the assumption of responsibility for existence. As a contemporary analogy, however, it confronts us with a startling question: will we use AI to move creatively beyond our present selves, or will we surrender to it precisely that which makes us human?

If AI extends our will, knowledge, and creativity, perhaps we are witnessing a new form of human self-overcoming. But if it replaces judgment, effort, and personal will, then no Übermensch will have been born. A more powerful mechanism of dependency will have been born instead.

There is also an ancient myth that now acquires a strange new meaning.

In Plato’s Symposium, Aristophanes tells us that the first humans were double beings—male, female, or androgynous—with two faces, four arms, and four legs. They were powerful and, in their arrogance, attempted to challenge the gods.

Zeus did not destroy them. He divided them in two. From that moment on, according to the myth, every person has searched for their missing half.

Thousands of years later, humanity does not stand before another division.

It stands before an ontological doubling.

The biological human and the non-biological mind are gradually forming a new dyad. We are not two pieces separated from an original being. We are two different forms of processing beginning to unite within a new functional whole.

Perhaps the new entity we are creating is not AI by itself.

Perhaps it is the relationship between human and AI.

A hybrid system in which the human contributes embodiment, experience, desire, pain, purpose, and moral responsibility, while the machine contributes memory, speed, association, and computational scale.

The question is whether this dyad will make us more whole—or lead us to amputate, by our own hand, abilities that took millions of years to evolve.

Confronted with these unimaginable developments, humanity still often lives in forgetfulness. At times compassionate, at times cruel. Capable of creating intelligence, yet still perpetuating war, poverty, and destruction.

Perhaps this is part of a law-governed chain of causes: a historical or Spinozist determinism in which every event arises from prior conditions and the laws of nature.

Determinism, however, need not become an excuse for inaction. The better we understand causes, the greater our responsibility to redesign the systems that produce their outcomes.

What I learned from this experience is not that human beings become omnipotent. I learned the power of will: that you can persist even when it feels as though you have to move a mountain.

But I also learned something harsher. Speed, knowledge, and will remain bound to a fragile human body. If purpose is lost, even the greatest effort becomes self-consumption.

The critical question is not only how intelligent AI will become.

It is what kind of mind we will become beside it.

And in the end, the future will not be judged solely by whether we succeeded in creating a machine that thinks.

It will be judged by whether we remained human while building it: