fAIl

I had this excellent idea for using AI in a constructive and ethical way the other day. One of the problems I’ve found in writing series of books is that it starts to get hard to recall precise details from several volumes previously: who met whom when, what they said and what were they wearing at the time. Damned if I can remember. I make copious notes as I go – and I can obviously go back and search through previous volumes to track down some detail – but it is time-consuming.

But, wait, we have AI now. That sounds like the perfect scenario for the technology. I can train a private AI on just my books (or one particular series of my books), and then I can ask questions to “chat” with those books as much as I want. Sounds awesome! What an age of marvels we live in!

My friends, it completely failed to work. I uploaded the text of my Triple Stars books to my test AI and started with a fairly simple question: “Who is the protagonist of the Triple Stars books?” Should be pretty straightforward, I figured – she’s mentioned on just about every page and almost all of the action takes place from her point of view. It’s Selene Ada, obvs.

The AI (I won’t identify which) utterly failed to get even this basic question right. It simply had no idea, didn’t even make an inaccurate guess. I tried other questions, as basic as I could think of, and it got every single one wrong. It was utterly useless on everything.

So, for now, it’s back to manual notes and searching. This is really disappointing though – this seems like such a great use for the technology that I will try again some day – perhaps when AIs are a bit more intelligent.