Can AI become conscious?
What is consciousness for? And is biological consciousness best understood as a self-organising algorithm that could, in principle, be recreated in machines?
In this talk, Joscha explores consciousness as perception of perception, coherence maintenance, modelling, resonance, self-organisation, and the possibility that machine consciousness may emerge through the right virtual architecture.
Essay: ‘The Machine Consciousness Hypothesis’ by Joscha Bach & Hikari Sorenson: https://cimc.ai/cimcHypothesis.pdf
CIMC: https://cimc.ai
Post: https://scifuture.org/joscha-bach-the… Intro
0:16 Why #consciousness?
0:42 How is it possible for a system to feel or experience anything?
1:19 The mind as a universal modelling system — like a game engine modelling a 3D world with observers in it
1:49 The observer as a simulacrum, a puppet the mind creates to tell itself it’s own story in the world
2:18 The output of the puppet/observer is being used by the outer mind to orchestrate other aspects of the body/mind
2:44 Thinking about consciousness invokes a feeling of an inscrutable mystery
3:10 Important questions — what is consciousness? how does it work? and what is it good for?
3:25 What does it mean for consciousness to be mysterious? Feeling significant, and inscrutable / ineffable
4:20 An music analogy — being able to read music doesn’t make music worse — the more you understand it the deeper the relationship becomes
4:58 Consciousness as meta-perception that generates order — coherence
5:44 The nature of the learning algorithm is tuned to be self organising, generating order / coherence
6:28 Consensus building — like on blockchain — distributed ledgers building consensus on reality
7:24 The genesis hypothesis — an algorithm we are born with that self organises to help us grow a mind — a prerequisite for our cognitive architecture that makes us human — consciousness as a pattern that colonises the mind
8:35 Conductor theory of consciousness — coherence maintenance — steps in when mind is threatened with decoherence / dissonance
9:32 Consciousness is perception of perception — it’s not inference or logical reasoning, it’s a percept — synchronous, resonant with reality
10:13 Reasoning vs perception is synchronous, it’s resonant in ‘real’ time
10:41 Resonance: Patterns try to synch with the cadence of what they are modelling (resonance) — see Grossburg’s Adaptive resonance theory
11:39 the field of AI doesn’t look at this stuff much
12:54 People are oscillating between being surprised at what AI can do, and with scepticism — but since GPT2/3 the world has changed faster than we can keep track — our perspective on AI has changed
13:23 #AI experts are also surprised that AI today turned out the way it did — some experts thought it would work by emulating the modularity of the mind — others expected a master algorithm that would grow from seed (analytical reasoning) to AI with highly adept logical capability with little common sense
14:52 But what we see is AI seems to have common sense — but sometimes makes logical mistakes — it’s a weird idiot savant that lives in an isolation tank — lacking direct coupling with the world — when humans are put into
15:48 For coherent AI, do we need to take them out of the isolation tank?
16:50 Surprising that perceptrons, feedback & optimisation of these where all that was needed for large language models — all the other symbolic AI was not on the critical path
17:33 Consciousness — Is the causal pattern deep and equivalent to what we have in our minds? It’s not important for current AI architectures
18:23 If AI needs to simulate conscious beings, then AI may need consciousness (or some of the structure)
19:17 LLMs don’t think in words or tokens, they think at sort of the same reality that humans do — at which point is LLMs just text, and approaching the deep structure in a human mind
20:10 Is a technical systems being conscious the simplest explanation for what it does? (cat example)
21:06 If consciousness is an algorithm that is less complicated than what’s required to otherwise behave like it’s conscious then perhaps it is
21:33 Machine consciousness hypothesis (MCH) — pt1 biological consciousness is best understood as a self organising algorithm/system in nature — 1st, 2nd and third order
23:41 MCH — pt2 — machine consciousness — the systems of self organisation that lead to discovery of consciousness in the bio organism, can be recreated in modern computers — through a virtualisation layer that allows for similar self organisation
24:24 MCH falsification criteria
If you enjoy this kind of content, please subscribe, like, and share.
Support #SciFuture:
https://buymeacoffee.com/tech101z
Adam Ford
Science, Technology & the Future