For years, artificial intelligence remained confined to factories, laboratories, banks and hospitals. It moved billions of euros, controlled industrial plants, detected tumours in CT scans, forecast electricity consumption, optimised logistics routes, identified targets in military systems and contributed to the design of new drugs. It was already an enormous technology, but hardly anyone talked about it.

Then, within a few months, artificial intelligence suddenly became the most discussed subject on the planet. Newspapers, television, governments, philosophers and entrepreneurs. Even the Pope. Everyone began asking the same questions: will it become conscious? Will it replace us? Will it be more intelligent than humans?

ChatGPT-3 was not as great a technological leap for the world and humanity as the chatter in the press might lead us to believe.

The total amount of AI operating in the world at the time of its launch increased only very slightly with ChatGPT-3. Indeed, the difference between before and after ChatGPT lies not in the technological leap, but in the fact that an artificial intelligence began talking to us.

Of course, it was not only the interface that changed. Language models represent an enormous technical advance. They have required an unprecedented availability of data, gigantic computing power and architectures that have changed the way neural networks are trained. Yet the cultural turning point does not lie in these aspects, invisible to most people. It lies in the fact that millions of individuals began conversing every day with a system capable of responding in natural language.

From that moment on, we began treating artificial intelligence as we treat human beings.

I have been writing about this since 18 January this year, calling AI a “technological subject” to make clear the paradox of an entity without subjectivity that is nevertheless perceived as possessing it. 

It is a deeply rooted cognitive reflex. When something uses language, constructs arguments, remembers the context of a conversation and seems to understand what we say, our brains spontaneously attribute intentions, emotions and will to it. It is the same mechanism that leads us to insult the satnav when it takes the wrong road or to thank even a basic voice assistant such as Alexa or Siri.

Except that with ChatGPT and its imitators, the illusion is much more powerful.

AI has become a kind of linguistic mirror. And as we look at the reflection, we have begun asking ourselves questions about consciousness, autonomy, the rights of machines and the future extinction of humanity.

Yet if we step outside the public debate for a moment, the landscape looks very different.

Artificial intelligence remains, above all, a collection of specialised tools. Thousands of models work silently within industrial processes, energy systems, hospitals, banks, logistics networks and research laboratories. Many do not even produce a sentence. They receive data, make a prediction and return a number. They are often more profitable than the major chatbots, but hardly anyone knows their names.

This is why we now use the same expression, "artificial intelligence", to refer to profoundly different realities. In technical language, the term encompasses machine learning, computer vision, expert systems, reinforcement learning, probabilistic models, control systems and much else. In everyday language, by contrast, AI has become almost synonymous with ChatGPT, Claude or Gemini.

This confusion, this use of the same expression for such different phenomena, has also fuelled a particular kind of rhetoric. Every new version of a model is presented as a step towards an imminent and immensely powerful general intelligence. Every improvement in linguistic capabilities is interpreted as evidence of growing awareness. Artificial emotions, self-awareness and future rebellions are discussed, often long before anyone has clarified what these concepts actually mean.

It is interesting to observe how very different attitudes coexist within the AI sector itself.

On one side, there is an almost messianic narrative. AGI is always just around the corner. Superintelligence seems a matter of hours away. Every technical advance is interpreted as a sign that humans are about to be overtaken.

On the other, there are companies that have been using artificial intelligence for decades to solve concrete problems, and do so in a very practical way. IBM belongs to this tradition. Listening to figures such as Dario Gil or Alessandro Curioni, a far less spectacular perspective emerges. Artificial intelligence is described as a powerful technology, set to transform processes, organisations and scientific research, but operating within the constraints of the real world: data quality, computational costs, reliability, responsibility and integration with existing systems.

Perhaps this difference also depends on the type of client. Those who sell visions often address the general public. Those who work with large businesses, hospitals, banks or governments, as IBM does, have to deliver results. An algorithm that reduces the energy consumption of an industrial plant or a city by three per cent is worth far more, economically and ethically, than a chatbot capable of producing a song that sounds as though it were written and sung by De André. Even if it makes fewer headlines.

In the end, the truly revolutionary element of recent years may be far less mysterious than it seems. On one side, we have an impressive growth in computing power. On the other, enormous quantities of data. In between, we find increasingly sophisticated statistical models. LLMs add a linguistic interface that makes this capability accessible to anyone.

It is precisely that interface that changes our perception. When a machine communicates through language, we almost immediately stop seeing computation and begin imagining a mind, even though we know perfectly well that there is none.

Perhaps the public debate on artificial intelligence reveals the extraordinary ease with which we attribute subjectivity to anything that speaks to us, through a reflex that seems to come straight from the Pleistocene and from our need, in order to survive, to attribute subjectivity and intention to as many beings and things as possible, since attributing subjectivity to a bush was far less dangerous than failing to attribute it to a sabre-toothed tiger.