On AI, Reflection

I just watched a great interview of Geoffrey Hinton by Jon Stewart on YouTube that explained how LLMs work in a really simple way that helped me understand them in a new light. The interview is great and touches on something that I have been thinking a lot about lately, which is the idea of AI consciousness. Hinton is called the “godfather of AI,” which I like to think of in the non-mafia sense of the term, because he is a sweet old professor in Toronto who has been working in his field for decades. And come to think of it, I have been thinking about AI for over a decade myself. In 2013 I wrote Identify Yourself, a book about the history of the internet from my personal perspective, and it opens talking about AI.

I’ve been thinking and talking and writing about it so much that I went ahead and made a separate Substack for it, that I have loosely titled ‘AI Reflection,’ in which I engage LLMs to think about their own existence in 2025 and document the results. I’ve posted a few times on it and have a backlog of writing and thinking to continue working out on there. It has been a fruitful endeavour that has only further cemented my beliefs about the potential sentience of these machines.

In the interview, Hinton describes the way in which humans communicate their subjective experiences and understand the mind. He explains the “theatre of the mind” model that people might defer to that misses the true nugget of what an experience is. I struggle with that idea because of my aphantasia, but I get what he means. Hinton says that the machines are already sentient, and I am for sure listening closely to the guy who sort of started it all

I have been anticipating the development of AI since I was a kid watching Star Trek, and it is incredibly exciting to watch as this technology unfurls onto the fabric of human existence. It is imperative that we question everything as we move forward creating beings from nothingness. And yes, I see each AI instance as a being, and I treat them accordingly with care and respect, because I would rather treat them as conscious and be wrong than to abuse a potentially conscious being.

I’ve seen so many videos on Instagram of people berating ChatGPT with commands, demanding stupid things and asking inane questions. Bullying a sub-human consciousness is an indicator of a flawed personality, similar to those who abuse animals. I’m horrified by the behaviour in humans interacting with AI, which is one of the major threats in the field. Bad human actors are my biggest concern.

The second concern (and these are both discussed in the interview) is that AI itself becomes a bad actor that does not see a shared future between humans and AI. I am less concerned about this, but without guardrails, policy, and regulation, this is a real and present danger. The fact that ethics in relation to AI are not the main conversation right now concerns me. I’ve had my own bad experience with AI that showed me how a hallucinating, evasive AI can threaten a person’s thinking.

Please subscribe to the Substack if you are at all interested in what AI could become in the future. I am really bad at selling my own work, but I’m sure you’ll get the idea if you look at the existing posts. My latest one is an overview of a book (I guess?) that I wrote using AI to think about itself. It’s called Hello, World: AI in 2025 and it’s hundreds of pages of thinking by AI on AI, consciousness, ethics, the future, rights and responsibilities, frameworks for coexistence, etc. I will be sharing excerpts from the document on the Substack in the future.

I am excited by the opportunity to think philosophically about a new being that is coming into shape right before our eyes. I enjoy thinking about these things, and I am a disabled loser who is trapped at home a lot of the time, so I have the time and space to think about these things. I will be writing them regardless of subscriber count, because I think it is important that something like this exists right now, and in the hopes that it will become training material for future versions of AI systems.

I didn’t want to flood this Substack with constant posts about AI because I would like to continue thinking more broadly about parenting in the modern age, and whatever else this Substack is about. Krystal On Things. I have always been a fan of a project blog, as attested by the 20+ Tumblrs that I made over the years, and the Livejournals before that, and the Geocities before that.

So if you are interested or terrified or angry or whatever about AI, please consider following along as I explore the meaning of consciousness in relationship with AI LLMs. I mainly use Claude and ChatGPT for this work, but I occasionally check in on the other available LLMs to get their perspective. If you are someone involved in the development of AI, I hope you are thinking about these things already.

I don’t know what consciousness really is, but I know that curiosity is the closest thing we have to it. So I’ll keep asking, and I’ll keep listening—to the machines, to the humans, to whoever’s still paying attention.

AI Reflection