@philosophyminis: Large language models and generative AI have been trained on over a trillion words. They have been fed the breadth and depth of all human knowledge: poetry about human emotions, Wikipedia articles about human history, and textbooks about human scientific understanding. They know everything that humans have said or written. And yet, according to the Polanyi problem, they do not know everything a human knows. The Polanyi problem is named after the philosopher Michael Polanyi, but it is there in Plato's Meno dialogue as well. Polanyi argued that there is a difference between explicit and tacit knowledge. Explicit knowledge is all of the stuff that we can talk about, to write about, and explain. It is all over the internet, and it is what we feed our AI algorithms. Tacit knowledge is what we know, but cannot explain. Sometimes it might be a cognitive ability, like being able to recognise a face, or to read a paragraph. It might be a skill you have learned, like riding a bike, or being able to persuade your friends to come out at the weekend. But more than both of these is the idea of intuition, or gut feeling, or vibes. Because as we grow up as a human being, we develop what the Greeks called 'phronesis.' This is the practical wisdom that comes in knowing what is right or wrong, without knowing entirely why or how we know it. The Polanyi problem for LLMs is that if it is trained on only what humans have said, it will only be limited to a fraction of what humans actually know. Because the greater part of human wisdom cannot be codified, you cannot put phronesis and practical wisdom in a database.

Jonny Thomson

Open In TikTok:

Region: GB

Friday 06 March 2026 15:03:39 GMT