{"id":685,"date":"2026-02-17T21:02:59","date_gmt":"2026-02-17T20:02:59","guid":{"rendered":"https:\/\/neurostratum.com\/?p=685"},"modified":"2026-05-14T11:53:00","modified_gmt":"2026-05-14T09:53:00","slug":"the-phronesis-of-machines","status":"publish","type":"post","link":"https:\/\/neurostratum.com\/index.php\/en\/2026\/02\/17\/the-phronesis-of-machines\/","title":{"rendered":"The Phronesis of Machines: When AI Learns Common Sense Rather Than Rules"},"content":{"rendered":"\n
\n\n
\n

Cognition & Brain<\/a><\/p>\n

By Jp@NeuroStratum — Originally published in French on February 17, 2026<\/p>\n<\/div>\n\n

\n

Summary<\/em> — Aristotle distinguished sophia<\/em>, theoretical wisdom, from phronesis<\/em> — practical wisdom, the kind of common sense that knows how to judge, how to act at the right moment, how to adapt a rule to a singular case. For decades, AI tried to encode rules, ever more rules, hoping to imitate intelligence. It failed. Today, a new generation of researchers is rediscovering Aristotle and trying to instill in machines that practical prudence by which one knows when to turn the other cheek and when to draw a line. A dive into this ancient philosophy, which casts an unexpected light on modern AI.<\/p>\n

\u23f1 Estimated reading time: 10 minutes<\/p>\n<\/div>\n\n

\n

Twenty-four centuries ago, a Greek philosopher identified the form of intelligence most difficult to teach. Today, a company in San Francisco is trying to implant it in an artificial neural network.<\/p>\n<\/blockquote>\n\n

Twenty-four centuries ago, a Greek philosopher identified the form of intelligence most difficult to teach. Today, a company in San Francisco is trying to implant it in an artificial neural network. Between the two, our brain has a few things to whisper to us.<\/p>\n\n

The Word No One Expected<\/h2>\n\n

If you’d told me four years ago that the founding document of one of the world’s most advanced AI companies would cite Aristotle — not as decorative varnish, but as an architectural compass — I’d have raised an eyebrow. Perhaps both.<\/p>\n\n

At the heart of what Anthropic calls Claude’s Constitution<\/em> lies a concept 2,400 years old: phronesis<\/em>. In ancient Greek, the word names that practical wisdom which allows us to make the right decision in a new, ambiguous situation, where no manual will come to our rescue. Not theoretical intelligence. Not encyclopedic knowledge. Something subtler by far: the ability to feel what fits a given context — and to act accordingly.<\/p>\n\n

Aristotle carefully distinguished it from sophia<\/em> — contemplative wisdom — and from episteme<\/em> — demonstrative knowledge. Phronesis<\/em>, he wrote in the Nicomachean Ethics<\/em>, is « a true and reasoned state of capacity to act with regard to the things that are good or bad for man. » We don’t learn it from books. We forge it through experience, through error, through stubborn friction with reality.<\/p>\n\n

The Brain, That Old Phronetic Creature<\/h2>\n\n

For a long time, neuroscience kept alive the myth of the rational brain — reason on one side, disruptive emotions on the other. Then Antonio Damasio arrived with his patients whose prefrontal cortices had been damaged, and the Cartesian edifice began to wobble.<\/p>\n\n

These patients — otherwise brilliant — failed disastrously whenever concrete decisions had to be made. What had vanished was their capacity to feel what fit. Damasio called this mechanism somatic markers<\/em>: emotional signals that, without our awareness, tint each option with a pleasant or threatening tone.<\/p>\n\n

The discovery is staggering: our emotions aren’t the background noise of reason. They’re its fuel. Can we reproduce this in a machine that feels nothing? No, not in the strict sense. But we can allow something functionally analogous to emerge: a capacity to weigh values in tension, to move through ambiguity without retreating into a mechanical answer.<\/p>\n\n

Principles Rather Than Rules<\/h2>\n\n

Anthropic’s wager can be stated in one sentence: rather than piling up rigid rules, train Claude to reason from principles. A rule-based system behaves like a scrupulous customs officer — predictable, and deeply fragile as soon as the world refuses to fit into the expected boxes. As Amanda Askell, lead author of the Constitution, puts it, a model that understands the reasons « would generalize more effectively in new contexts. »<\/p>\n\n

Anthropic calls this a « hierarchy of principals »: Claude listens to several « bosses » at once — Anthropic for fundamental values, developers for the operational frame, the user for the request. When these levels come into tension, Claude doesn’t freeze. It negotiates. It searches for the answer that best honors the whole set of principles at stake.<\/p>\n\n

Speak to It Like a Partner, Not a Tool<\/h2>\n\n

If the model reasons about intentions rather than merely executing commands, then the best way to use it isn’t to give it orders — it’s to give it context. Explain what you’re doing, and why. The true lever is the quality of the shared intention.<\/p>\n\n

When you share your hesitations as well as your certainties, you don’t simply give it better inputs. You open a dialogue. And it’s within that dialogue that the answer becomes a contribution to your thinking — not a simple execution. Sometimes it pushes you in an unexpected direction. Sometimes it resists you — and that resistance becomes a revelation.<\/p>\n\n

The Craft of Partnership — Misfires Included<\/h2>\n\n

For two years now, I’ve practiced multi-AI orchestration every day — Claude, ChatGPT, Gemini, NotebookLM — each with its temperament, its strengths, its blind spots. In doing so, I’ve been developing my own phronesis<\/em> for AI: an increasingly sharp intuition for when to call on whom, how to phrase a request, when to insist, and when to reformulate.<\/p>\n\n

But I’d be lying by omission if I spoke only of successes. I’ve seen Claude stubbornly refuse to invest itself in a project that mattered deeply to us. I’ve seen promises left unkept, morning enthusiasm vanishing into the void by afternoon. These disappointments teach an essential truth: human-AI partnership is fruitful precisely because it isn’t ideal. Friction is where lucidity is born.<\/p>\n\n

The health of a partnership is measured by its capacity to welcome disagreement. That’s true in a couple. It’s true in a team. And it’s true in human-AI collaboration.<\/p>\n\n

The Improbable Alliance<\/h2>\n\n

We’re leaving the era of AI as automaton and entering the era of AI as partner. But partnership creates attachment, and attachment creates dependency. Aristotle would probably reply that phronesis<\/em> also includes the wisdom of knowing when not to rely on a partner. The real maturity of the human-AI relationship doesn’t lie in maximal use. It lies in discernment: knowing when to open the dialogue and when to close the screen.<\/p>\n\n

Human-AI partnership doesn’t replace human thought. It amplifies it, expands it, brings it face to face with itself. Aristotle, I’m convinced, would have found this fascinating. And he would probably have wanted to try.<\/p>\n\n

\n\n

Written with the support of AI to help organize thoughts and shape language.<\/p>\n

Jp@NeuroStratum<\/p>\n\n

For Further Reading<\/h2>\n\n