One thousand years ago, were people not already generally intelligent? They had the same human intelligence that we have today. Why are we trying to reach general intelligence by throwing tons of data at AI that humans from a thousand years ago had never seen? Why do we assume that information is what makes us generally intelligent? It seems like we're treating intelligence as though it's just a very large pile of facts. Doesn't AI already know more than us? How many times more information than us does it need to accumulate before we can say that it is generally intelligent? Or is it that AI has more facts than we do, but is not more intelligent than we are? AI learns frames. A frame is a way of understanding a situation. It is a conceptual filter through which it can interpret and respond to a situation. The more “good” data, the stronger the frame. This is why AI can sometimes interpret a scan correctly when a doctor does not. This is almost certainly due, at least in part, to AI having a stronger frame because of the greater amount of relevant data that was thrown at it. But what happens if the frame doesn't exist yet? What happens if AI doesn't have the capacity to formulate the frame? Humans have something remarkable and different. We don't solely operate within existing frames. We also create new ones. We have the ability—although it is not foolproof—to recognize that the current way of looking at a problem is inadequate, and we instinctively discover entirely new ways of understanding it. That is fundamentally different from selecting among previously learned patterns. Even if AI eventually becomes incredibly good at combining existing frames, is that the same thing as originating a genuinely new one? Perhaps AI will be as good as it can be based on how many of the frames humans have created it can learn, how well it can learn them, and how quickly it can apply them. There's another thing to consider. Some human frames may never be reducible to data in the first place, thus locking AI out of truly “grasping” those frames and leaving it only able to mimic them to the best of its ability.

I’m wondering if anyone else is dealing with messy and large databases. For AI agents to properly function long term your data needs to be clean so that it can understand where to go, what everything means and how it all works together. Any recommendations for a tool or way to organize an existing database?