I’d push back on that, we could use this to invent a new cognitive architecture approach to AGI (route perception directly to memory state and use that as the world model) and see if it improves RL performance. In fact I think the world models paper did this. You could conceivably have a generative / predictive world model, instead of a perceptual / remembered one. I think Karl Friston is more in the former, predictive camp, which implies your consciousness isn’t a memory, but rather a prediction (Metaverse in your head, so to speak) so it’ll be interesting to see how AGI eventually pops up, will they consciously experience their perceptions, or their predictions? Perhaps perception and prediction are two sides of the same coin and we just need to figure out what material the coin is made of.
Also for neuro, maybe it means Alzheimer’s would show up on perception tests before it shows up on memory tests, if memory is implemented by perception. And since we mostly look at “memory centers” of the brain in that disease, perhaps we could find issues in perceptual areas, too. Maybe we look at the hippocampus when we ought to look at V1, olfactory region, sensory gyrus, etc — if Alzheimer’s follows the clogged sink metaphor, then the clog could be in a different brain region entirely.
Personally I don’t like to think about it that way because that implies consciousness has “lag!” However, we’d have to evolve perception before we could evolve to remember that which is perceived.
One big problem with this article is the fact our vision etc has a lot of holes. If we consciously experience a mere memory, then there’s a whole lot of in-painting going on in the perceptual system to fill those holes, which makes it quasi-generative, and thus less of a memory and more of a prediction of stuff we didn’t actually see. Which means memory and prediction and perception are all the same. That’s a can of worms because it means when we experience a moment, we’re actually predicting the past…
Also for neuro, maybe it means Alzheimer’s would show up on perception tests before it shows up on memory tests, if memory is implemented by perception. And since we mostly look at “memory centers” of the brain in that disease, perhaps we could find issues in perceptual areas, too. Maybe we look at the hippocampus when we ought to look at V1, olfactory region, sensory gyrus, etc — if Alzheimer’s follows the clogged sink metaphor, then the clog could be in a different brain region entirely.
Personally I don’t like to think about it that way because that implies consciousness has “lag!” However, we’d have to evolve perception before we could evolve to remember that which is perceived.
One big problem with this article is the fact our vision etc has a lot of holes. If we consciously experience a mere memory, then there’s a whole lot of in-painting going on in the perceptual system to fill those holes, which makes it quasi-generative, and thus less of a memory and more of a prediction of stuff we didn’t actually see. Which means memory and prediction and perception are all the same. That’s a can of worms because it means when we experience a moment, we’re actually predicting the past…