What Socrates can teach us about AI

"If Socrates was the wisest person in ancient Greece, then the great linguistic models must be the most stupid systems of the modern world."

According to the oracle of Delphi, Socrates was the wisest in Greece, not because he knew more than anyone else, but because he was the only one aware of the limits of his own knowledge.

The LLMs are exactly the opposite.

They don't know what they don't know, they don't track the truth, they don't have a "critical eye," nor do they recognise beauty, they simply make statistical conjectures presented with the same certainty with which they would present a verified fact or the greatest falsehood in the world.

In this article in the magazine The Impossible, Carissa Véliz, a philosopher at the University of Oxford, retrieves Harry Frankfurt's words to name this precisely: it's not a lie, it's something more dangerous... It's stupidity and this is usually a persuasive speech disconnected from any concern for the truth.

The liar at least knows what the truth is and deliberately avoids it. "The one who tells the truth and the liar play on opposite sides of the same game." The language model doesn't even play that game. The truth is simply irrelevant to him.

And these systems are designed to be compelling. Not to be truthful. To satisfy the user's preferences with an answer that sounds good.

We are building decision-making infrastructures – in journalism, medicine, politics, education – on systems that have no mechanism to worry about whether what they say is true.

Véliz closes the article with a question: Athenian democracy killed Socrates and let the sophists (the deceivers of that time) prosper. Are we repeating the same mistake?

By Rommy Artigas