Who Declared the Algorithm Intelligent? And When Did Humanity Decide to Believe It?

Who, when, and on what scientific grounds decided that an algorithm was intelligent? Did mathematical calculations suddenly acquire reason while the rest of us somehow missed this historic moment? Or did someone simply come up with an impressively sophisticated name for a technology now being marketed as the crowning achievement of human civilisation? I must admit, the marketing strategy is remarkable. Call a computational system intelligent, teach it to generate convincing responses, and before long, millions of people begin conversing with it as though it were a thinking being. Even more fascinating is how readily those same people sometimes interpret a generated answer as evidence that the machine actually understands the question. How convenient! Apparently, we no longer need to demonstrate the existence of intelligence. We merely need to create a sufficiently convincing impression of it.

History, however, is considerably more interesting than advertising slogans. In 1955, John McCarthy and his colleagues proposed a research project using the term Artificial Intelligence. The following year, the Dartmouth workshop helped establish a scientific field dedicated to investigating whether certain aspects of learning and intelligent behaviour could be described precisely enough to be reproduced by machines. Notice the distinction. We are talking about a scientific hypothesis and a research programme, not a certificate confirming that algorithms possess human understanding, consciousness, or the capacity for independent reasoning. Yet somewhere between university laboratories, technology corporations, and advertising campaigns, that distinction began to fade rather suspiciously. Science asked whether machines could exhibit intelligent behaviour. Marketing rushed to announce that they were already intelligent.

And this is where the great adventure begins, along with the genuinely interesting questions. As a linguist and programmer, I understand the difference between processing linguistic structures, calculating probabilities, and the human interpretation of meaning. A language model can produce a grammatically flawless sentence, imitate argumentation, and suggest solutions to complex problems. But does that mean it possesses understanding in the human sense? Does convincing linguistic output demonstrate the presence of thought, or does it reveal how easily we attribute cognitive abilities to systems that display particular patterns of behaviour? More importantly, who benefits from this terminological confusion? When a technology receives a name that implies intelligence, it acquires not only a scientific identity but also a considerable commercial advantage. And I cannot help wondering which is more impressive: the machine's ability to generate answers or humanity's willingness to believe that a thinking mind must necessarily exist behind them.

Who Gave the Algorithm Its Title of Intelligence?

Science asks questions, investigates hypotheses, tests results, and acknowledges limitations. Marketing, apparently, has decided to skip a few of those tedious stages. All a system needs to do is generate convincing text, acquire an intelligent-sounding name, and present itself as the next great revolution. The easiest part comes afterwards: informing humanity that the future has already arrived, although full access requires a subscription. And what if the results turn out to be inaccurate? Well, revolutions never promised technical support for human expectations.

But before we begin celebrating this magnificent new machine intelligence, I would like to return to a rather elementary question. Who exactly decided that the algorithm was intelligent? At what point did mathematical operations, logical rules, and computational procedures acquire a quality that humanity has associated for centuries with knowledge, reasoning, and understanding? More importantly, who established that successfully performing a particular task was sufficient grounds for attributing intelligence to the system performing it? If we follow that logic without making the necessary distinctions, we may soon have to organise an international scientific conference in honour of the calculator. After all, it has been performing mathematical operations faster than most humans for decades, yet nobody has invited it to deliver a lecture on the philosophy of numbers.

The story takes us back to 1955, when John McCarthy introduced the term Artificial Intelligence in a scientific research proposal. The original idea was to investigate whether certain abilities associated with human intelligence could be described mathematically and simulated by machines. Notice that we are discussing a scientific hypothesis, not evidence that an algorithm thinks, understands, or possesses consciousness. Apparently, however, somewhere between the research laboratory and the marketing department, the hypothesis received a rather spectacular promotion. From simulating intelligence to becoming an intelligent machine. What an extraordinary career advancement for an algorithm!

And I cannot help admiring the remarkable career of the word intelligent itself. Once upon a time, the concept demanded serious discussions about knowledge, logic, memory, learning, abstraction, and problem-solving. Today, we encounter it attached to phones, watches, household appliances, cars, corporate platforms, and practically every new digital service in need of a more impressive market position. Of course, in technical language, intelligent may refer to automated adaptation, complex information processing, or particular functional capabilities. The problem begins when the term is used in a way that encourages consumers to assume the presence of understanding, independent judgement, or almost human abilities without clearly explaining what the system can and cannot actually do. How remarkable! One word, sufficiently flexible to describe an entire scientific discipline and sufficiently marketable to enhance the appeal of almost anything to which it is attached.

I would not, however, blame the Dartmouth researchers for the linguistic extravagance that followed. They formulated research questions within an emerging scientific field. Responsibility for how these concepts are used today cannot simply be transferred to the people who introduced them. Quite the opposite. Their original formulation gives us every reason to treat grand claims with greater caution. A scientific hypothesis does not become an established fact merely because it becomes popular, and simulating a particular behaviour does not automatically resolve every question about the nature of that behaviour. A technology can be extraordinarily useful without our attributing abilities to it for which sufficient evidence has yet to be established. Unfortunately, a modest technical description does not sell the future quite as effectively as the promise of a machine that thinks on our behalf.

And now we arrive at a contradiction that, in my view, deserves far more serious attention. If a machine produces a result resembling the outcome of human intellectual activity, what exactly have we demonstrated? That the machine thinks? That it successfully reproduces certain characteristics of thinking? Or that we ourselves are not always particularly good at distinguishing the outward appearance of intelligence from the mechanisms that produce it? This is not some trivial dispute over terminology. The answer influences how we evaluate these systems, what decisions we allow them to support or execute, and how much trust we place in them.

Because a machine's ability to imitate certain manifestations of intelligence is one thing. Humanity's willingness to accept imitation as proof is quite another. And the truly fascinating part begins when someone discovers how profitably that distinction can be sold as a technological miracle.

Before awarding the machine a diploma in intelligence, however, we should clarify one essential detail. An algorithm is a sequence of formally defined operations, not a miniature electronic brain contemplating the meaning of life between calculations. Modern machine-learning systems are vastly more complex than classical algorithms, and language models can process enormous quantities of information, identify patterns, generate text, and solve certain tasks with impressive accuracy. But at what point does computational complexity become understanding? Does producing the correct answer mean that a system comprehends its significance? And if a machine can imitate certain aspects of human thinking, is that sufficient reason to conclude that it possesses the same cognitive abilities? Apparently, there is a considerable distance between technical functionality and human intelligence, a distance that marketing prefers to measure in monthly subscriptions.

As early as 1950, the British mathematician Alan Turing raised the question of whether machines could think. Rather than becoming trapped in an endless debate over the definitions of thought and intelligence, he proposed his famous imitation game. The idea was to investigate whether a machine's behaviour in a text-based conversation could become sufficiently convincing that a human evaluator would struggle to distinguish it from another person. Here, however, an intriguing contradiction emerges. A successful conversation may demonstrate convincing linguistic behaviour, but it does not, by itself, establish whether human-like understanding or consciousness exists behind that behaviour. Yet today, we frequently encounter claims suggesting that the ability of language models to converse naturally is almost conclusive evidence of machine intelligence. So, if a system manages to convince me that I am speaking to a thinking being, must I necessarily accept that it actually thinks? What fascinating scientific logic. Particularly when the observer's confusion starts being treated as evidence of the observed system's abilities.

This, to me, is one of the most interesting problems of contemporary technological culture. Human beings are inclined to search for meaning, intentions, and intelligence wherever they recognise familiar patterns of behaviour. Fair enough, but when we receive a logically structured sentence, when a system responds politely, or when it uses language associated with understanding and empathy, we readily begin attributing the qualities of a genuine conversational partner to it. Yet what we perceive and what the system actually does are not necessarily the same thing. And once we become willing to recognise intelligence in convincing behaviour, my next question becomes unavoidable. Is the machine deceiving us, or are we simply demonstrating a remarkable talent for manufacturing our own illusions?

Compass Insight | Who Is Actually Supposed to Think?

I am not particularly troubled by machines becoming increasingly capable of imitating human behaviour. What concerns me is the ease with which people begin accepting convincing imitation as evidence of intelligence. There are important distinctions between computation, generation, understanding, and consciousness, and those distinctions do not disappear simply because marketing has decided to package them under one impressive label.

And if scientific progress demands evidence, why are we so willing to accept advertising promises as scientific conclusions? A machine can generate answers without it being established that it understands them as a human does. But a person who trusts those answers uncritically voluntarily abandons one of the most important advantages of human intelligence: the ability to question.

The irony? We called the machine intelligent, and now we have to remind the human being to think.

What a magnificent triumph of human intelligence! We created an algorithm capable of imitating aspects of thought, called it intelligent, turned it into a multibillion-dollar business, and now voluntarily entrust it with our own judgement.

All that remains is to award it a Nobel Prize for convincing humanity that thinking is an unnecessary expense.