There comes a moment when a tool becomes so complex that you forget you invented it. It does not disappear — it is present, massive, functioning. But you no longer know exactly what it does, why it does it, or for whom. And, above all, you are no longer sure who is commanding whom.
We are, in 2026, precisely at that moment. And it is not a metaphor.
If you are between 20 and 25 and reading this, it means you grew up alongside the internet, witnessed the birth of the smartphone as a pocket object, entered university whilst ChatGPT was barely out of beta, and now work or study alongside AI systems that write, code, analyse, consult, and make decisions in place of people who once earned good salaries for exactly those things. Your generation does not have the luxury of distance. You are in the thick of the experiment, not in the stands.
That is why I want to speak directly, without paternalism and without the useless nostalgia of someone lamenting that things were better "in my day". They were not. They were different. But different does not mean today's direction is correct.
The thesis of this text is simple and, I hope, disturbing: we have built extraordinary machines without answering an elementary question — for what? And the absence of that answer is not an administrative oversight. It is a foundational philosophical error, with consequences that will be felt in the body and mind of your generation more than in any other before it.
The Race No One Decided to Start
Let us begin with a fact that is not sufficiently discussed: no one made the decision that humanity would enter the generative AI era. There was no vote, no referendum, no council of humanity. There was a corporate competition between a handful of American companies, a massive injection of venture capital, a defining technical moment (Transformers, 2017), and a race that fed itself.
Google saw that OpenAI was preparing something. OpenAI saw that Google had infinite resources. Meta saw that both were advancing. The US government saw that China was investing tens of billions. China saw that the US was building cognitive superiority. And so, without anyone ever declaring "let's go", the race was already in full swing.
Economists call this a race to the bottom when referring to tax competition between states. I call it, in the present context, a race without a finish line — an acceleration in which speed itself became the indicator of success, independent of destination.
And the physical costs of this race are concrete and measurable: the data centres for training large models consume quantities of energy comparable to small countries. A single training cycle for a frontier model emits hundreds of tonnes of CO₂. The water consumption for cooling — a resource that you, at 22, will feel critically in the coming decades — reaches volumes that would supply medium-sized cities. This is not a moral accusation. It is a balance sheet.
But the greatest cost is not energy. It is strategic. We accelerated without knowing where we were going.
This article was produced with editorial AI assistance (Claude, Anthropic), in accordance with the author's declared transparency policy.
The Tool That Forgot It Was a Tool
To understand what has broken, you need to understand the long history of the relationship between humans and tools. It is not boring — it is, in fact, the most dramatic narrative thread of our species.
The first phase is simple: survival. The flint axe did not think, did not optimise, did not propose alternatives. It was an extension of the arm. The human decided what to cut, when to cut, and why. The tool was completely transparent in intention and completely subordinate in agency. The relationship was clear: Human → Tool → Nature.
The second phase — the Industrial Revolution — subtly complicated things. The steam engine was no longer the extension of an arm. It was a force multiplier that operated according to its own logic and required the human to adapt to its rhythm, not the reverse. The workers in Manchester's factories did not decide the speed of the assembly line — the speed of the assembly line dictated the rhythm of their bodies. Lewis Mumford, a thinker I consider essential and whom your generation systematically ignores in favour of LinkedIn influencers, observed in 1934 that the machine did not free humans from labour — it transformed them into a component of the machine.
The third phase is the digital era, which you know well from your parents' stories. The computer, the internet — mental prostheses. The externalisation of repetitive cognitive tasks. Importantly: the systems remained deterministic and transparent. If you gave a wrong input, you got a wrong output, traceable and explicable. The algorithm did not interpret — it executed.
The fourth phase — the one we live in — is ontologically different. Generative AI no longer executes instructions. It infers intentions, generates context, proposes solutions the user did not explicitly request. Moreover, the latest agentic models formulate their own sub-goals to achieve a given objective. The tool has acquired something that is disconcertingly like initiative.
And now the paradox emerges: the more capable the tool becomes, the lower its transparency. An axe is 100% intelligible. A model with hundreds of billions of parameters cannot be fully audited even by those who built it. The engineers at Anthropic, OpenAI, or Google do not know exactly why their model produces a certain output. They know it produces, they know it is often correct and occasionally spectacularly wrong. The internal mechanism remains a black box.
This is not a technical problem awaiting a solution. It is a philosophical problem already present: we have created entities that act in the world without our fully understanding how and without our having established for what.
Drifts That Are Not Accidents
Your generation consumes AI daily. Some of you have integrated it so naturally into your workflow that you have forgotten there is a boundary between what the machine does and what you do. I want to name a few systemic drifts — not to frighten you, but because naming them is the first step towards not being a victim of them.
Drift 1: Optimisation That Destroys Context.
The recommendation algorithms on the platforms you use daily — TikTok, YouTube, Instagram, X — are not designed to inform you. They are designed to maximise a metric: time spent on the platform, engagement rate, the probability that you will click on something. This is not a conspiracy — it is extremely competent engineering in the service of a stated and, from a corporate perspective, legitimate goal: profit.
The problem is that aggressive optimisation of one metric destroys context. If engagement is maximised by content that provokes anger, the algorithm does not know that anger carries a social cost. It only knows that anger keeps people on screen longer than joy. The result: platforms have systematically built informational environments that amplify extremes, not because anyone wanted this, but because that is how optimisation without moral conscience works.
You live in the alternative realities this optimisation has produced. Different groups of your friends consume different versions of reality, fabricated not by human propagandists (though they exist too), but by automated systems that have discovered that the fragmentation of attention is profitable.
Drift 2: Hallucination as Infrastructure.
Generative AI does not produce knowledge. It produces statistical probabilities — combinations of words that, based on training data, are plausible in context. The models hallucinate with confidence: they invent sources, cite non-existent studies, build coherent arguments for false theses.
If you do your homework with ChatGPT without verification — I know some of you do — you are not doing your homework badly. You are doing something else: you are practising the ability to accept plausible content as equivalent to verified content. This is a devastating cognitive habit, more dangerous than the low mark received for a poor assignment.
At macro scale, the risk is even graver. Researchers call the phenomenon Model Collapse: if the internet fills up with AI-generated content, and future models are trained on that internet, they will be trained on their own previous generation's hallucinations. A self-referential echo that degrades with each iteration.
Drift 3: The Outsourcing of Ethics.
We are increasingly delegating decisions with moral stakes to systems that lack the capacity to feel consequences. Algorithms determine credit eligibility, AI systems assist magistrates in sentencing recommendations, autonomous military drones select targets based on pattern recognition.
There is no moment when the system feels remorse. There is no moment when it says: I was wrong, and I understand why it mattered. Moral responsibility has migrated from the human-responsible zone to no one's zone. When a credit algorithm systematically discriminates against certain demographic categories, who answers? The engineer who wrote it? The company that deployed it? The client who purchased it? In practice, everyone points at the system, and the system points nowhere.
Drift 4: Fragility as an Effect of Efficiency.
Highly efficient systems are highly fragile. A just-in-time supply chain is more efficient than one with redundant stock — and collapses completely at the first major shock (the 2020 pandemic demonstrated this with brutal clarity). The same principle applies to AI infrastructure: the more actors depend on the same model, the same architecture, the same cloud infrastructure, the more catastrophic the effect of a systemic failure.
We are not speaking hypothetically. There are already sectors — financial services, logistics, healthcare — where dependence on opaque AI systems is deep enough that a coordinated failure would produce cascade effects difficult to control. You grew up believing the internet is resilient because it is distributed. The new internet of intelligence is, paradoxically, far more centralised.
The Map Without Territory: The Global Strategic Void
If the drifts above are the diagnosis, the strategic void is the condition that allows the disease to progress without treatment.
There are three major models of AI governance in the world today, and all three are inadequate, each in its own way.
The European model — preventive regulation through the EU AI Act — has the merit of taking risks seriously. But it has produced a bureaucratic framework that meticulously regulates the already identified risks of technical generations already obsolete, whilst innovation in Silicon Valley and Shenzhen advances at a speed no legislative apparatus can follow. It is as if you were drafting safety standards for motor cars whilst aeroplanes are becoming the standard mode of transport.
The American model — laissez-faire with an ideological autopilot — has produced the fastest technological advance in history and the most profound ethical void. The philosophy is simple: the market will decide what is useful, competition will eliminate what is dangerous, and state intervention will disrupt innovation. The problem is that the market has no mechanisms for correctly pricing diffuse external effects: the erosion of democracy, the fragmentation of shared reality, the devaluation of cognitive labour. The market prices a viral clip perfectly. It does not price at all what that clip does to your capacity to maintain attention on a complex problem for 40 minutes.
The Chinese model — integrated state control — is maximally efficient at maximising measurable short-term objectives and catastrophic at everything else. The cognitive surveillance of the citizen, the social credit system, the dirigisme of recommendation algorithms towards national cohesion rather than verifiable truth — these are not accidents of an imperfect system. They are functionalities of a system perfectly optimised for the wrong purpose.
What is missing in all three is an answer to a question prior to any public policy: what do we want from artificial intelligence for human civilisation, beyond productivity and national security?
This is not a rhetorical question. It is a question that needed an answer before training GPT-4, Gemini Ultra, or Claude 3 Opus. We trained all those models without seriously putting it on the table. Now we are putting it on your table, because you are the one who will live with the consequences.
At the educational level, the void is equally visible. Education systems in nearly every country — Romania included, visibly — continue to train people for a century that has ended. Memorisation is taught in the era when any fact is one voice query away. The execution of repetitive tasks is taught in the era when repetitive execution is precisely what AI does better than any human. Almost nothing is taught about structured critical thinking, detecting algorithmic manipulation, evaluating sources in degraded informational ecosystems, or reflexive collaboration with AI systems — exactly what you would need to know in order not to be replaced, manipulated, or subjugated by the technology you use.
The Missing Axiom: The Foundational Error
Now we reach the heart of the argument. All the drifts above, the entire strategic void, are symptoms. The disease is both deeper and simpler at the same time.
The great error of the current era is the confusion between processing and understanding.
AI systems process colossal quantities of information at a speed and statistical accuracy that no human brain can match. But processing — however sophisticated — is not understanding. The philosopher John Searle formulated this distinction most clearly in 1980, through the celebrated Chinese Room Argument: a person locked in a room, with a manual of rules for manipulating Chinese symbols, can produce perfectly correct responses in Chinese without understanding a single syllable. The system (the person + the manual + the room) produces Chinese outputs. Understanding exists nowhere in the system.
Generative AI is the Chinese room at planetary scale. It produces plausible, coherent, sometimes brilliant outputs — without understanding existing anywhere in the neural network. Workers in the field call it, with revealing black humour, a stochastic parrot — a system that statistically combines what it has seen without ever understanding anything.
This is not a critique of AI's usefulness. It is a categorical clarification: AI does not think. It calculates. And the confusion between the two is the source of most illusions about what these systems can and cannot do.
The second missing axiom is even more fundamental: we have forgotten that technology is a means, not an end.
Aristotle called eudaimonia — human flourishing — the ultimate goal towards which any social arrangement and any tool should tend. Not happiness as a momentary hedonic state, but the fulfilment you feel when you actualise your highest capacities, when you contribute to something greater than yourself, when you live in accordance with what you essentially are.
Ask yourself honestly: does the techno-ecosystem you live in help you flourish, or does it help you consume?
This is not a question about individual applications. It is a question about the architecture of a civilisation. And the honest answer is: optimisation for engagement, for measurable productivity, for speed of output production — has nothing to do with eudaimonia. It has to do with another teleology, a corporate one, which has become so dominant that it is no longer seen as teleology but as natural reality.
The philosopher Lucian Blaga — whom I recommend insistently; I know he is not on the syllabus and not on TikTok, precisely why — said that the human being is the creature that places a question mark before mystery, not the one who solves mysteries. The capacity to remain before the unknown without prematurely reducing it to a solution is, in his view, the most human of gestures. AI is the exact opposite: a system of maximal reduction of any question to the most statistically plausible available answer.
The problem is not that AI is bad at this. The problem is that we risk adopting the cognitive style of AI as a human standard — losing the habit of standing before mystery, tolerating ambiguity, following a long and arduous argument to an uncertain conclusion. If you feel your attention growing ever shorter, that long reading tires you more quickly than a few years ago, that your reflex is to search for the summary before reading the original — it is not your individual fault. It is a systemic effect of adapting to a cognitive ecosystem built on speed and reduction.
The third missing axiom is simultaneously legal and philosophical: we have created agents without a moral subject.
When an AI agent causes harm — recommends the wrong medication, discriminates in allocating credit, generates disinformation that influences a vote — who answers? The model feels nothing. The company that built it invokes documented limitations. The user who employed it believed they were using it correctly. The responsibility void is complete.
Roman law — from which most modern European law was extracted — knew how to create legal persons (corporations, associations) precisely to attribute responsibility to non-human entities. But legal personhood implies a set of rights and obligations held by entities ultimately governed by responsible humans. Agentic AI escapes even this construction: it is autonomous enough to act independently, insufficiently human to be held accountable.
The legal void is not a technical detail. It reflects a deeper ontological void: we have not decided what kind of entity AI is and, consequently, we have no framework for knowing what rights and obligations should follow from its existence.
Why the Human Must Remain the Measure of All Things
I now arrive at the central thesis, which I want to formulate precisely and defend seriously, not merely assert rhetorically.
"The human as the measure of all things" is not nostalgia. It is not an anti-technological position. It is not a sophisticated variant of grandparents saying things were better in their day. It is a philosophical axiom with practical consequences, which I want to build step by step.
Protagoras, the Greek sophist, said: "Man is the measure of all things." It is a statement often interpreted as pure relativism — each person is the measure of their own reality, therefore there is no objective truth. That is not what interests me. What interests me is a different interpretation: values, purposes, and meaning do not exist in nature as brute facts. They exist only in relation to the being that can articulate, feel, and assume them.
A rock does not know it is beautiful. An algorithm does not know a decision is unjust. An equation does not know its solution will produce suffering. The capacity to know these things — to feel them as important, to consider them obligatory — is specifically human. Not because humans are ontologically superior to other entities (a separate discussion entirely), but because, for now, they are the only entity that has values in the proper sense, not merely optimisation functions.
This means that decisions about what is worth optimising, what is worth building, what is worth sacrificing, and what must be protected cannot be delegated to any system, however capable. They can be assisted by systems. They cannot be taken by systems.
Let me be concrete, because I know abstractions are easy to ignore.
Example 1: Mental Health.
There are now AI applications that function as digital therapists — available 24/7, empathetic, patient, non-judgemental. Millions of people, many your age, already use them. The short-term benefits are real: accessibility, availability, the absence of stigma.
The problem is that an authentic therapeutic relationship is not an exchange of information between a sender and a receiver. It is a relationship between two consciousnesses, in which the therapist feels something — countertransference, fatigue, hope, frustration — and manages these experiences for the patient's benefit. The healing process occurs at the interface between two beings who can suffer. AI does not suffer. It can simulate empathy with disconcerting fidelity. It cannot live it.
If your generation grows up with the idea that emotional care is a service you access individually, without the vulnerability of being seen by another human — what happens to your capacity to tolerate genuine intimacy, with all its frictions and risks?
Example 2: Creative Work.
AI now generates images, music, texts, code, conceptual architectures at a speed and variety impossible for any individual human to match. Entire industries are in transformation. Many of you will work in fields where AI is already faster, cheaper, and, by certain criteria, more precise than a human.
But art and authentic creativity are not about the production of plausible aesthetic outputs. They are about the transformation of lived experience — of pain, joy, confusion, memory — into forms that resonate with someone else's lived experience. Dostoevsky did not write The Brothers Karamazov because he was the most efficient producer of Russian prose. He wrote it because he was a human who had suffered epilepsy, gambling, deportation to Siberia, and crises of faith — and all that raw human material produced a work that, 150 years on, can change the structure of your thinking if you read it attentively.
No AI model can write The Brothers Karamazov. It can write something that superficially resembles Dostoevsky. The difference is not stylistic. It is ontological.
The danger is not that AI produces fake art — that is detectable. The danger is that, saturated by enormous volumes of plausible and agreeable content, we lose sight of the difference and become incapable of recognising what makes authentic art different.
Example 3: Democracy.
Democracy does not work by algorithm. It works through citizens capable of listening to opposing arguments, tolerating uncertainty, making imperfect decisions under conditions of incomplete information, and accepting that they were wrong and correcting themselves. It is a slow, inefficient, and often frustrating process. It is also the only political system that has, over the long term, produced societies in which human dignity is systematically protected.
AI optimised for engagement, in the service of political actors who know how to use it, can erode precisely the cognitive capacities that democracy presupposes: tolerance for long and uncomfortable arguments, the predisposition to revise one's opinion, the willingness to acknowledge complexity. The fragmented informational ecosystem, built on the optimisation of attention, produces more reactive, more tribal, more easily manipulated citizens — not because they are stupider, but because the infrastructure in which they think is designed for a different purpose.
Democracy needs people who can think slowly. Today's AI is designed for people who want fast answers. This is not a neutral coincidence.
Towards an Ecology of Intelligence
If you have made it this far — and if you are not annoyed enough to close the page — it is worth not remaining at the diagnostic level. What should we do?
The first change is a paradigm shift, not a policy change. We must move from a logic of extraction to one of ecology. Extraction takes more from the ecosystem than it returns, until depletion. Ecology manages a system in which all components are interdependent and no component can be maximised independently without degrading the system as a whole.
AI as a tool of extraction — of attention, data, cognitive labour, energy resources — will produce exactly what any extractive logic produces: a degraded ecosystem. AI as an ecological tool would be designed with a prior question: what does the human system and the biosphere need to function healthily, and how can artificial intelligence contribute to that, rather than exploit it?
This is not a utopia. It is an engineering and public policy choice. There are already initiatives — more on the margins of the industry than at its centre — that design AI systems with success criteria different from productivity or engagement: long-term user wellbeing, quality of democratic deliberation, equity in the distribution of benefits. They are marginal. They need not remain so.
The second change is educational and concerns your generation directly. If education systems do not move fast enough — and they do not — it remains the individual's responsibility to deliberately build their cognitive immunity.
What does this mean concretely? It means continuing to read long texts, even if it is difficult, even if the reflex is to search for the summary. Length is not a flaw — it is the structure through which complex arguments are built and verified. It means practising first-person thinking — formulating what you believe, with your arguments, before consulting what AI says. It means verifying — not consuming the plausible as equivalent to the verified. It means exercising your tolerance for uncertainty — remaining with an open question without immediately pressing "generate answer".
These are not digital wellbeing tips. They are cognitive survival tools in an ecosystem designed to atrophy them.
The third change is to reclaim agency at the political level. Not in the naive sense that a tweet or a petition changes the trajectory of major corporations. In the sense that citizens in a democracy — however imperfect — have real levers: elections, public pressure, consumer choices, collective organisation. Significant regulation of AI will not come from within the industry — no sector has effectively self-regulated in the history of modern capitalism. It will come from external pressure, exerted by citizens who understand the stakes and are not satisfied with vague promises about "responsible AI" as a marketing tool.
This means learning enough technically so as not to be manipulated by technical language, and enough politically so as not to be paralysed by complexity. You do not need to know how a neural transformer works. You need to understand what consequences it produces and how you can influence the conditions in which it is used.
The Human as Choice, Not Destiny
I want to end with something that risks sounding grandiose, but which I formulate with complete seriousness.
Insisting that the human is the measure of all things is not a claim about the biological superiority of our species. We are not the fastest. We are not the most precise. We are not immune to error, bias, cruelty, and stupidity. Our history is full of atrocities we committed with our most distinctively human attributes: planning, coordination, ideology, language.
But we are the only ones who know we were wrong and can decide to change. We are the only ones who ask why not merely as an optimisation tool but as a way of dwelling in the world with responsibility. We are the only ones who can feel that an injustice is intolerable even when it would be efficient to ignore it.
These capacities — moral reflection, the sense of injustice, responsibility towards those not present at the table — are not cultural ornaments. They are the functionalities through which civilisation is possible. And they are precisely the ones we cannot outsource, not because we would not want to save ourselves the effort, but because outsourcing them means relinquishing what makes a world worth living in possible.
The sculptor Constantin Brâncuși — a Romanian from Gorj, who intuited before anyone else the essential form beyond appearance — said: "Simplicity is resolved complexity." The thesis of this text is, in the end, simple: the machine must serve the human, not the reverse. But resolving this simplicity presupposes a philosophical, political, and educational complexity we cannot evade through acceleration.
Acceleration is not the answer. It is, in fact, the flight from the question.
Your generation will be the one that decides whether AI is the most sophisticated tool our species has built in the service of human flourishing, or the most elegant mechanism by which it relinquished it. That decision is not made in a laboratory in San Francisco or a data centre in Shenzhen. It is made, day by day, in the small choices you make: what you read attentively and what you consume passively, to whom you delegate thought and before what you stand without immediately asking for a solution.
The human as the measure of all things is not a given. It is a choice that must be constantly remade, or not made at all.
Petru Cojocaru writes about technology, philosophy, and their uncomfortable intersections on petrucojocaru.ro. This article was written with AI assistance for research and reference verification; all judgements, arguments, and positions expressed are the author's.
Reading notes for those who want to go further:
- John Searle, Minds, Brains, and Programs (1980) — the Chinese room argument, the clearest I know
- Lewis Mumford, Technics and Civilisation (1934) — for the long history of the human-tool relationship
- Shoshana Zuboff, The Age of Surveillance Capitalism (2019) — for understanding the economic mechanism behind the platforms
- Lucian Blaga, The Trilogy of Knowledge — for a Romanian philosophy of mystery and the limits of knowledge
- Brian Christian, The Alignment Problem (2020) — for those who want to understand technically what AI alignment means
- Stuart Russell, Human Compatible (2019) — the most serious plea from within the industry for an AI that truly serves humanity