Of AI, Mind and Mimic

Ask a room full of professionals what AI is, and the conversation stops.

Not because they don't know. They all use it, every day, and more of them each week. The conversation stops because they have never had to agree on the word, and the word has outrun them.

Try it. What is AI? The engineer says a model that predicts the next token. The marketing professional says a thing that writes my emails in bulk. The doctor says something that reads my scans faster than I do. The policy-writer says a risk. All of them are right, and none of them agree. That silence is not a gap in their expertise. It is the absence of a shared definition. And it is, quietly, the most important thing about AI that nobody is discussing.

Unfortunately, we are having this discussion at the wrong speed, at a time when the AI adoption curve has gone vertical.

McKinsey counted, in 2025, 88 per cent of organisations using AI in at least one business function. A year earlier the same survey found 78 per cent. The year before that, 55. Microsoft and LinkedIn, polling 31,000 knowledge workers across 31 countries, found three-quarters already using generative AI at work — and, more telling, 46 per cent of those users had started within the previous six months. In the United States, the St. Louis Federal Reserve put generative-AI use at 54.6 per cent of all adults by August 2025, up from 44.6 per cent a year earlier.

India leads the world. The best available survey about the adoption of AI by the Indian population in general is a research paper by Kiran Garimella, Rutgers University School of Communication and Information, published in April 2026.

The survey — an online panel checked against a face-to-face rural sample — puts overall generative-AI use at roughly two-thirds of working-age adults, among the highest in the world.

But the headline hides the real story, which is the line inside India itself. Outside work, 75.2 per cent of urban Indians use it, against 61 per cent in semi-urban districts and 54 per cent in the villages. Among graduates it runs at 77 per cent; among those who never finished a degree, 32.1.

These figures are astonishing for a technology most of us did not know existed four years ago.

A technology adopted this fast, by this many people, on the strength of a word that does not have a singular, clear definition, is a technology that will be feared in the dark and misused in the light, because you cannot govern a thing you cannot name.

This is not an argument against AI. It is an argument for an old, unfashionable habit: define your terms. Before we ask what "artificial" intelligence is, we have to ask what intelligence is. The second question was settled centuries ago, in dictionaries and in philosophy. The industry simply chose not to consult either.

The dictionaries agree more than the technologists do.

The Oxford English Dictionary says intelligence is "the faculty of understanding; intellect."

The Oxford Learner's Dictionary defines intelligence as "the ability to learn, understand and think in a logical way about things."

Merriam-Webster says "the ability to learn or understand or to deal with new or trying situations," and, more pointedly, "the ability to apply knowledge to manipulate one's environment or to think abstractly."

Under all three definitions sits the same Latin root: intellegere, to understand — from inter, "between," and legere, "to choose."

To be intelligent, in the oldest sense of the word, is to choose between. To see the world not as one given thing but as a field of possibilities, and to select. Not to react. Not to predict the next word. To understand, and then to choose.

That last word — choose — is the one the whole argument turns on. A machine that predicts does not choose, in this old sense. It selects the most probable next token. The distinction sounds like pedantry. It is the entire difference between a mind and a mimic.

The philosophers sharpened what the dictionaries stated.

Aristotle called the human being the animal that possesses "logos" (it means reason, and also speech), and made that faculty the thing that sets us apart from everything else that lives.

Two thousand years later, Descartes built a whole philosophy on one verb: "I think, therefore I am." For him intelligence was not one property of the mind among many. It was the mind's essence. Kant went further and tied it to freedom: to be intelligent is to act from principles rather than merely react to causes.

All three put intelligence on the inside — in understanding, in judgement, in consciousness — never in what an outside observer happens to see. That inwardness is the load-bearing wall of the classical idea. It is also the exact wall the AI industry would later knock out.

The moderns, at least the honest ones, kept circling the same place. Douglas Hofstadter defined intelligence as the ability to reach beyond the immediate and find the deep, fluid analogy hiding inside a situation — a leap no lookup table makes. And David Chalmers drew the line even harder: however well a system behaves, there remains a stubborn question about whether it "feels" anything, the "hard problem" that no amount of output can answer. None of this is settled. That is the point. It is genuinely contested, and it has been contested for centuries. The AI industry behaves as though it were not.

In 1950 Alan Turing did the quiet, world-changing thing. He proposed to consider the question, "Can machines think?" Then he described this question as meaningless. Next he proposed a test. If a machine can converse well enough that a human judge cannot tell it from a person, call it intelligent.

In one stroke, the definition of intelligence moved from inside the mind to outside it. From understanding to performance.

The industry has run on Turing's shortcut ever since. The working definition of AI became: a system that does things which, if a human did them, it would be called intelligent. It is a genuinely useful definition. It lets you build without first resolving three thousand years of philosophy. Its merit is real: by measuring behaviour instead of essences, it turned a debating society into an engineering discipline.

But look at what the shortcut threw away. The dictionaries and the philosophers defined intelligence as understanding — as grasping meaning, as choosing. The new definition kept none of that. It kept only the surface: a thing counts as intelligent if it looks, to a judge, like a person.

In 1980, the American philosopher John Searle argued that a system can produce all the right answers, can pass the test, without understanding a single word it says. Imagine a man who knows no Chinese, locked in a room with a book of rules. Someone slides a slip of paper under the door. It carries Chinese characters — a question. The man looks up the marks in his book. The book tells him what to write back. He writes it and slides the paper out. To the person outside, the answer is perfect — it reads like a native speaker wrote it. But the man understood nothing. He only matched one set of marks to another. A machine running a program, Searle said, does exactly this. It takes in the marks, follows the rules, and gives back the right marks. It never understands a word. Producing the right output is not the same as understanding. Syntax is not semantics. Performance is not comprehension.

So the AI industry has been living with a quiet contradiction at its centre. It calls its machines intelligent by a definition of intelligence that no dictionary recognises and no philosopher before Turing would have accepted. Meanwhile it behaves, and markets, as though the classical meaning were intact. "AI" now names two different things at once: the old idea of a mind, and the new reality of a very good prediction engine. Most people, most of the time, are using the second while believing the first.

There is a joke in the field, credited to the computer scientist Larry Tesler: "intelligence is whatever machines haven't done yet." Chess was intelligence until a machine won. Writing a scholarly paper was proof of intelligence until it wasn't. Each time the machine crosses a line, the line is redrawn. The joke is really a confession. The industry has never agreed what the word means, because the word has always been defined by whatever is still safely human.

The failure to define the word is exactly why the two bad outcomes — fear and misuse — have room to grow.

Fear feeds on the unknown. What can't be named can't be sized. When "intelligence" floats free of its definition, it summons the image of an unknown creature that understands us, judges us, might one day turn on us.

Clarity, on the other hand, cuts that image down to size. What we are actually dealing with is a system that learns statistical patterns and produces output that resembles understanding. It is astonishing, world-changing, and worth taking seriously. But it is not a mind. No one is afraid of a calculator, because everyone knows what a calculator is. Look at this machine for what it actually is, and you stop being afraid of it too.

Misuse comes from the same confusion, from the other side. Once "behaves like a person" is allowed to mean "understands like a person," anyone can sell you behaviour as understanding — the chatbot that sounds as though it cares, the model that writes with confidence about things it does not know, the company that harvests your words while warmly inviting you to talk. Left unexamined, the word "intelligence" is a licence to mislead. Understood, it is a defence: you can tell the difference between a tool and something that claims your trust.

Go back to the room full of professionals. This time, before you ask what AI is, put the dictionaries on the table. Understanding. Judgement. The choosing-between. That is intelligence. The machine, for now, does something else, and what it does is remarkable enough to deserve its own honest name. Agree on that first, and the room will stop going silent. It will start having the argument that actually matters: not what the machine is, but what we will let it do, and what we will not.