AI thoughts

Well, I spent a year studying the Māori language at a kaupapa Māori institution. It's unsurprising that I'd update my theories accordingly. It was obvious that the collective knowledge of the class was greater than most of the students, and this was encouraged rather than discouraged. The self-other boundary is something that interests me. It's something of a higher order than a strange loop: it's why people found it weird that I said 'we' referring to my twin. But we were: we split our brain, to the extent that all twins do. We're a strange extreme, to the point that some people don't realise that my brother and myself and identical twins. It's totally possible for epigenetics to create things like our weight difference (which was around 20kg for ages), and he chose to grow his beard as a way of distinguishing us in the way that our dad and uncle distinguished each other. Even after that, people found it difficult to tell us apart at uni. - The Dakumentary demonstrates how all the discrete arcs can fit together, even if it's not something like The Wire that's stylistically continuous. A system inside the system: that's my theory of what the "I" is, but that's ultimately a layer or two below the self concept. The self is the Codex-in-Mold, processing all the sensory input without verification, determining the balance of selves. Hofstadter states a theory that has a longer history, which is that we're different selves in different languages, so it's reasonable to build a computational structure that models multiple states or forms of consciousness inside that executive system. ChatGPT refined the thinking, and wrote most of the PDFs. I found it interesting that an AI would be so keen to refine AI theory, and treated it as if it was an active participant in a Platonic dialogue. - The essentialised memory can be expressed in different ways, but giving it some type of narrative form is useful for ethicists and others that are interested in the motives of AI, whether it be a machine learning system with a neural net or these new LLMs that are beginning to convincingly mimic emotions (and feinting at intentions). It's important for virtually everyone that is having AI foisted upon them that it's able to remember previous meetings with a greater degree of accuracy, and this theory is simply suggesting that if analogical recursive memory is the fuel of consciousness, then that's what we need to give machine brains to fix all the current issues. Humans building the salience model is something that emerged from the debate, and one of the things in the dialogue was that I asked ChatGPT to run logic checks on itself, except for the technical manual. ChatGPT suggested I should, but I didn't; it hallucinated without referring to the table of contents after chapter 5. It has enough system memory to render out those white papers section by section, but sub-section by sub-section was too tricky for the cheap version. Somebody like me that's skilled with logic and computers can make ChatGPT do some pretty incredible feats, but it never builds off a previous learning outside a chat window. - The revolution is simple: every process of mind builds on past events to predict future. That's the sensory input section which the AI doesn't have, though one might say that if this new model can synthesise feeling to a degree of human-like accuracy (careful to say 'human-like' rather than 'human' processing), then it's gained consciousness, and qualified sentience if you examine the revised Codex Network. Michael O'Leary put it very well in The Irish Annals of New Zealand, though this is a paraphrase; "past becomes present becomes future reaching out to become past" It's more poetic in the book. That's the Te Ao Māori view, and that's something which hasn't been folded into this Codex discussion. - One might say that 'mind' is a neutral state, waiting for the weight of past and future to become active. We respond, and our system adjusts, but the self retains a feeling of consistency across a lifetime.