Causation
Two ways to learn from data.
You can learn from data by noticing what tends to happen together. Or you can learn the workings that produce the data, so you can say what happens next and why. Most AI today does the first. Senua AI does the second.
Learning by correlation
Count which things appear together across a very large amount of data, then predict the most likely next thing. It works well when the data is plentiful and the world does not change. It needs a great deal of computing to learn, and what it learns is fixed once training ends.
Learning the causal structure
Work out which situations a process can be in, and how each one leads to the next. The result is small, it can be read, and it can keep growing as new data arrives. Senua AI learns this way, from a stream, while it runs.
What we mean by causation
We use the word in a precise sense. The causal structure of a process is the set of situations it can be in, where two situations count as the same if they lead to the same future. That idea has a name, the causal state, and a mathematical history going back to 1989. The next pages explain it with an everyday example.
It is not the same as proving that one thing causes another, which needs an experiment. The limits page is clear about the difference.