I was recently asked to give a talk on higher-order theories to Google DeepMind. I was all set to give my usual spiel about higher-order theories and their empirical testability, but then I found out they wanted a talk on AI consciousness and higher-order theories. I figured that since I had originally planned a chapter on higher-order theories of consciousness, functionalism, and AI, for my book (but never got around to it due to my sabbatical running out), and that I had been to a conference recently which I was writing up as a blog post, and which had sparked some ideas on the topic, I thought this would be a good chance to get straight about what I thought. I started with powerpoint slides and a rehearsal of the basics based on the draft of the blog post I had been working on, then a rough draft written based on my oral run-through with the PowerPoints.
From there I started to think that I should talk to some AIs to write the paper. I don’t usually use AI in my research, though I have experimented with it to see if it is useful (which so far I had not found it to be. My book was entirely written by Humans and my previous attempts to use it for literature searches have been foiled by hallucinations). Things were different this time around. After a read through and initial discussion, what resulted was an intense week of discussion and revision. I started with a draft that was about 6000 words and now the paper is much larger! Granted I did not use the most advanced models, but my overall experience was that the AI was useful for developing the ideas I had and would often give me a useful objection or counterpoint. But then it would suggest a revision as a way to handle it and I would implement a different one. Or when I responded by saying ‘good point, but doesn’t…’ it was almost always the case that the suggestion I had/edit I made was better than the one suggested by the AI model. So I really did have the feeling that I was getting decent feedback and that it was helping me develop ideas (and connect it to my previous published writing) that I originally had but also that I as able to out-think it in sense of better understanding the argument and points being made. This may not last long but my overall impression was that it would be a useful tool for someone like me who teaches upwards of 15 classes a year to increase the amount of ideas that get developed into full blown papers. To test that, I had it analyze my blog and there were several posts that had arguments that were never published by me but which someone else published (on their own) later. Had I been teaching less (and more interested in publishing) I could have developed those. After this little vanity project it spontaneously suggested to me a decent trade book idea based on my blog: pairing events from my autobiography with a philosophical argument that the story brings up. This was a kind of obvious idea but it had never occurred to me! So all in all, I came away convinced that there is a case to be made for AI use in writing.
I know that there is growing pushback against AI-slop and AI philosophy, and I still think that writing is the best tool for self-discovery that we have and that ultimately this is not something AI can do for us. Nor would I want to replace the effort/thinking that went into those un-pursued blog posts. Thinking through the issues has always been the most enjoyable part of philosophy for me, and I see writing (especially revising and rewriting) as essential to that thinking process. So something would be lost if you just had AI generate the paper for yourself (something being debated ver at Daily Nous as I write this, haha). Still, I will let this work stand for the record. I think the paper is good. The original idea I had was interesting and discussion with AI really did improve the paper in many ways. I wrote this mainly as a way to get straight what I wanted to say for the talk and to test my ideas against current models and frankly I do not need a publication at this point in my career, so I guess I’ll just leave it here and see what people think.
