Patterns, Facts, and AI

Replying to this thread.

I think if you scan through a hundred papers and find a pattern in those papers, and use that pattern to make something, you haven't copied anything.

Harder to do: you can also scan through a hundred papers, find the one original thing, and use that to make something. Then the question of whether you've copied is a matter of degree.

In neither case does it matter whether a machine or a human does this. What makes something ethical or not is the act, not the technology.

Anyhow. I scan though a hundred papers (more or less) every day to produce my newsletter. I try to find the original and highlight that. And my own original work is based on patterns I see in the data.

I passed over Doug's article because (in my view) it had been done before (not the least of all, by me, in 2019). There's no ethical issue or blame here; most of what is produced in the world (including most of what I produce) is not original.

Again, it's not the tech.

I think the hardest of all to produce is something original, based on a pattern no one has seen, that is useful to people.

I think that if a machine did that, there would be no real issue with the fact that it was a machine that did that, because we'd all be too busy trying to take advantage of this new knowledge.

But it's hard for a machine to find a new pattern, because there's so much pattern recognition already in human discourse. Useful is also really hard.

And just so, as Belshaw comments, "in some circles not being rabidly 'anti-AI' gets you tarred and feathered." This is a pattern of discourse. It gets magnified and reflected back and forth.

Certain patterns (eg., 'AI copies') get reified until they become 'fact'.

As @poritzj says, it's the data used to find the patterns that matters, for all sorts of statistical reasons. But who among the human pundits is honest about the complete corpus of material they draw upon?

That's where the ethics (if ethics = good science) comes into it, if at all.

If your data is bad - if it's not diverse, if it's not informed, if it's propaganda - then your pattern recognition is bad, and you reify the wrong things into facts the promotion of which is actually harmful.

That's my main criticism of most anti-AI writers: that their data is bad. They draw from popular and commercial press, or from (say) commercial publishers with a vested interest.