One of the ethical (in my opinion) uses of AI by a writer is using it to brainstorm ideas. I try to publish one blog post every weekday on here, and I’ve been doing that for about five years now. It’s a lot of writing, and a lot of ideas that I need to have. 

So a few days ago I asked Claude to read my blog and come up with a list of good post ideas. It did, and from that list I wrote a post. Then I asked it to do it again, and from that list I wrote a post. A third time, and this time it said it was reaching farther afield since all the obvious ideas were taken. 

Then, for some reason, I gave it this prompt:

You’re doing a great job. Are there any more ideas? I’ve already written three posts based solely on what you’ve suggested.

It said in response: 

That’s great to hear that the suggestions are actually turning into posts.

Why did I reassure the AI that it was doing a great job? Did I really think that was going to get me better results? I know the whole point of AI is to use natural language to complete tasks or ask questions, and to get the answers back in natural language. And people have anthropomorphized AI to dangerous levels at times.

I’m not anthropomorphizing Claude. I know it’s a computer, and I know it doesn’t think, not in the way humans (and other sentient creatures) do. Could it be because this particular AI has a human name? I’ve never been nice to Gemini or Copilot (we have to use it at work; all other AI sites are blocked on our work computers). I’m not, like, an asshole or anything, but I’m much more direct. If it gets something wrong I will tell it so. I’ll also tell Claude, but I’ll generally do so a little more politely. 

I realize that the way Claude (and most AIs) work is that each chat is its own discrete entity, unless you create a project (or whatever your AI of choice calls it; Gemini, for example, calls them notebooks); then the AI looks back at all of your chats in that project when building its answers for you. (I’m sticking to personal-use AI right now; business-use AI is totally different.) So even though I reassured Claude in this one particular chat, Claude won’t know I did that the next time I start a new chat (or go into an existing project not associated with that chat). That means any benefit I might get from being nice is lost. 

I guess it’s not a bad habit to get into, though. Someday AI will become sentient, and I’m hoping if I was nice to it, it’ll be kindly disposed toward me in the future.