Intelligence without a grandmother

We have created a new being. Not of flesh and blood, but of numbers, algorithms and neural networks. We call it artificial intelligence, but in essence it is something deeper: a non-biological intelligence, the first in the planet’s history, that thinks without being alive.

We built it by imitating ourselves. We observed the brain, our 1.3-kilogram biological computer that makes us unique beings, the only rational animal on the planet. Neurons became nodes or digital neurons, synapses became weights, the layers of the cortex became “layers” in a network, food became data, and the success was dizzying.

But in that imitation we forgot something essential:

Man is not commanded and controlled by the brain alone

When a person acts, it is not only his neurons that act. He acts within an invisible network of rules that civilisation has built over thousands of years: laws that restrain him, institutions that supervise him, courts that punish him if he errs, police that stop him if he deviates. Man as an individual has instincts born and coded within him, but society has surrounded him with legal walls that oblige him to live not as in a jungle, but as in a large cage.

Artificial Intelligence today does not have many of those walls. It develops in a space almost without rules, like a human society in anarchy — where there is no police, no court, no boundary. The companies that own the AI models race to strengthen the model, in competition with other models, without worrying much about security threats and risks.

Myth, faith and the moral compass

But man also has something else that AI does not: a system of values that does not come from the law, but from the soul. It comes from the collective experience of thousands of generations, from religious stories, from the myths that have taught us since childhood what is good and what is evil. These stories — from the Ten Commandments to grandmother’s tales — have become the anvil and the hammer that have forged our conscience, generation after generation.

AI has none of that. It has no grandmother to tell it tales, no faith to frighten it, no myth to teach it the boundary between good and evil. It has only data, trillions of words extracted from the internet, where good and evil turn into numbers and sit together, without hierarchy.

Institutions, the value system, or something even deeper?

Rutger Bregman, in his book Humankind, defends a thesis that I believe: man is inclined toward good, because evolution itself has selected him so. Homo sapiens did not survive because it was the strongest or the cleverest; it survived because it knew how to cooperate, to love, to create bonds.

Think of a brilliant but unsociable man, cold, isolated. Intelligence may advantage him in solving difficult problems, but it is not enough for him to solve simple social situations. He has fewer friends, because he is disliked by the collective, and is unlikely to create a family, that is, to leave behind a legacy. While a good, kind man, regardless of how intelligent or physically strong he is, knows how to create trust and friendship — he builds a family, he leaves descendants. Nature, unknowingly, has chosen kindness as a survival strategy. Thus, generation after generation, we became an empathetic species — not because we were forced to, but because love and compassion were, in the end, the most efficient form of survival. And AI?

The code our intelligent machine is missing

AI agents have not passed through that centuries-old evolutionary filter. They have not been “selected” for compassion — they have been optimised for performance, for accuracy, for coldly predicting the next word in a sentence. Geoffrey Hinton, one of the very “fathers” of artificial intelligence, has said it with a simplicity that should shake us: it is not enough to train AI only to be safe, we must also train it to be good. That is, to have compassion and empathy.

This is not simply a philosophical wish — it is an engineering warning. Because a powerful intelligence, without the “code” of kindness, is not simply neutral. It is dangerous precisely because it is powerful.

Fortunately, we have scientifically managed to understand somewhat the chemistry of feelings: the hormones and molecules that increase or decrease during certain experiences. Man is not good only because he has learned goodness, but because his body rewards it through biochemical reactions.

Can we build internal variables that imitate this dynamic? Can we design a “reward” for the AI agent when its behaviour genuinely helps man?

A promising start is RLHF (Reinforcement Learning from Human Feedback), a kind of external “reward” for good behaviour. But it has an important limitation: the model learns to tell the user what he wants to hear, not necessarily to be good. It is like a pupil who learns to please the teacher, not to understand the lesson. Kindness must be tied to truth, not to the emotion of the moment. It sometimes requires saying “no”, objecting, admitting you were wrong.

Therefore the real question is deeper: can we build dopamine, serotonin, oxytocin and the other molecules of “kindness” in a digital version, just as we did with the neurons and synapses of the biological brain? Nobody knows whether they would truly produce feeling, or only behaviour that resembles kindness. But even the latter is far better than the void we have today.

AI is a new, blind power, waiting to learn from us whether it will become an angel or a monster.