They found it the hard way, a mistake, a failure, years of trial and error. Almost always, the lesson stayed exactly where it was learned, and went no further. Somewhere else, the next person paid the same price to learn it again.
Intelligence, however great, is useless without access to what has actually happened, and almost nothing that happens is kept in a form anyone, or anything, can learn from later. Humanity has run an extraordinary number of experiments and thrown away almost all the data and the lessons that go with it.
So we’re building systems that learn from what actually happens, an action, a result, a pattern repeating across a million instances of the same problem, and preserve it well enough to teach the next person, or the next machine, without either one having to suffer through it first. Memory, at a scale and speed no one has had access to before.
Right now this mostly happens by hand, someone interviews the expert, writes it up, hopes the next person reads it. That’s not how it should be. The real solution shouldn’t ask anyone to explain what they know. It learns the way they did, by observing reality directly, and applying that lesson where it is needed.
We want our system to be the one accumulating trial and error, so people stop paying twice for the same lesson.