We’re standing on the shoulders of the giants at HuggingFace.co.
We’re sure you’d like to know how we know the work is written by an AI. The truth is we can’t. It’s true! AIs are, to a very large degree, black boxes. In that respect, their ‘minds’ are a lot like our own.
Philosophers and psychologists have puzzled over how we know things for millennia. We’ll spare you the details, but the short answer is this: nobody really knows.
AIs are similar. When you train an AI, you give it many examples of something you would like it to identify—pictures of cats, say—mixed in with examples of other things—dogs, horses, rats, and so on. The AI guesses ‘cat’ and is told whether is right or wrong. There’s no need to tell it why it’s right or wrong. You just tell it that it’s wrong and let it move on to another guess, over and over again.
The AI makes a model of what a cat is but even it can’t really explain how it knows—and that’s okay! Again, it’s a lot like you. You know what a cat is, but you’ll never be able to define it. Think about it: you know that if a cat had three legs, missing teeth, blue eyes, or whatever, that it would still be a cat. What is a cat? You just know it when you see it.
That’s how it goes with AIs too!
So, when we (or the people at HuggingFace) say that we’re 90% sure this was written by an AI, we can’t explain a whole lot more than that. It means that in other cases that resemble this one, the AI guessed correctly 90% of the time. And that’s kind of all we know.
It’s up to you to determine what your confidence in our confidence is.