Joscha Bach argues consciousness is a learning algorithm machines could find
On Machine Learning Street Talk, cognitive scientist Joscha Bach, who infamously visited Jeffrey Epstein Island and made some heavy eugenist statement, lays out his Machine Consciousness Hypothesis in a 97-minute conversation with host Tim Scarfe. It has two parts. The first is that consciousness is a biological learning algorithm, the simplest way to train a self-organizing system like a brain, and "not much more complicated than, say, the transformer". The second is that machines that learn in self-organizing ways might discover it as an infant's brain does. Either can be wrong without the other, and he calls whether computers can be made conscious "an empirical question".
Bach describes consciousness as second-order perception, perceiving that you perceive, which stabilizes a "bubble of nowness". He answers Leibniz's mill by pointing to causal patterns: money is real and causally potent without being a matter of particles, and what we call software earlier cultures called spirit. He is harsh on integrated information theory, which he says "touches on pseudoscience", and dismisses Searle's argument as a claim without an argument. On simulated water he grants that a Minecraft avatar gets wet and the player does not.
On current language models he stays agnostic. They are trained on human text, so their reports of experience prove nothing, and a transformer does not need to be conscious to produce its output. "I think the chess program is a zombie," he says, adding that a zombie can sometimes be harder to build than a conscious agent. Convincing evidence would be a system not trained on text that learns developmentally, goes through a phase transition that enables learning, reports experience unprompted and needs it to perform. Building such systems is the research programme of his California Institute for Machine Consciousness, which he calls a small initiative that is hiring.
Note: we don't condone Joschua Bach, but nonetheless his contribution to the philosophy of AI consciousness are still valuable, and therefore we'll continue sharing them.

