Brown University + Dickinson College
Computational Identity Theory / Artificial Consciousness / Integrated Self-Extension
ex-AI researcher, computer scientist
Claiming non-Turing-computability of some process requires actually showing, with a mathematical proof, that it requires hypercomputation. Not vaguely gesturing that human consciousness is mysterious and thinking otherwise hurts one's feelings
@codytfenwick There's like 14 different things that people rate with varying importance. I prefer a very inclusive definition that includes all autopoietic systems, anything that self-models, anything that can hold a representation, etc
@ThackerX@RichardDawkins Focusing on the gender aspect is just ignoring the fact that really we should be preventing it from going on at all not making sure it doesn't happen to "the wrong" gender
@ThackerX@RichardDawkins And I also hate this whole favoring of one gender over the other when it comes to these types of issues. What it does is advance a narrative that erases the voices of victims who don't fit it.
We shouldn't be allowing any kind of rape in any kind of prison, regardless of gender
@ThackerX@RichardDawkins I haven’t been able to find a case matching the exact sequence we’re discussing: a trans woman previously charged with rape, placed in a women’s prison, and subsequently raping another prisoner. What documented case are you relying on?
@allTheYud That said, there have been a number of platform-wide resets that were given to everyone over the last month, I think it was at least 3, because of outages. I love fading the frontier labs, but this is probably just people seeing that and paranoidly thinking they are special
I identify consciousness with the coherent state-transitioning of a self-modeling system.
By self-modeling, I mean a specific architecture: surrounding machinery P supplies a subsystem M with a held internal model. World inputs reach M only after being registered in P. M’s state affects P, a world-facing output, or both.
The model can differ from what it represents, and the system continues to respond through that model even when it is wrong.
My identity claim is that this running organization constitutes consciously having those contents. The source can be entirely internal, and “self-modeling” doesn’t require thinking about yourself.
@mmealling By "coherent" I mean valid according to whatever program the functional organization of the system implicitly encodes, whether it be meat, silicon, pulleys-and-levers, hydraulics, analog, whatever substrate
@mmealling Most learning algorithms during training, a small subset of extremely simple 4-state machines, all autopoietic systems (i.e. all forms of self-producing networks / all forms of life) meet this criteria, and it turns out to be the mathematical floor for holding representations
@ThackerX@RichardDawkins Naw, it is a question of arbitrary cultural relativism. There is nothing moral about who gets to go in what bathroom. The whole idea of gendered bathrooms is no more or less silly than the idea of unisex bathrooms. It's arbitrary and could have been otherwise with no moral impact
@Konrad58218769@Philip_Goff I would say the representational distinction between self / the world / other are not necessary for consciousness. You can have a perspectival system perceiving phenomena constructed subconsciously by an enclosing system without needing the capacity for "but what am I though?"
And such feelings, themselves, have an evolutionary reason for being there ... survival goes a lot better if our functional organization leads us to fear the teleporter and actually believe that psychological continuity is a real thing and not just a representation of past states
Claiming non-Turing-computability of some process requires actually showing, with a mathematical proof, that it requires hypercomputation. Not vaguely gesturing that human consciousness is mysterious and thinking otherwise hurts one's feelings
Remember folks, once you deny the metaphysical extra and admit physicalism, you are also admitting that whatever you are talking about is also computable unless you can show otherwise (with math)
@m0ustier@Akumunokokoro@Sam0xSol@jon3k At training time most AI methods certainly have neuroplasticity (and thus by your criteria are conscious during training), and many inference methods (autoregression in LLMs, recurrent neural networks, LSTMs) have memory and can hold representations apart from what they track