I already believed something akin to panpsychism, but with a weighting function towards higher interaction processes of some specific kind I’m not sure I know how to mathematically identify, but which relates closely to integrated information, signal to noise ratio, compressibility, agency. I suspect I’m looking for some relationship between those, and of course some of those need additional parameters that an ideal form of the concept selector wouldn’t. it has to accurately match my brain-internal qualia, while also having strong reason to believe it generalizes. Because intelligence arises from a series of simple objectives, a fundamental component of which is usefulness per watt ie a signal to noise ratio towards an agentic target, we can expect that anything that contains optimization will contain some magnitude of consciousness or agency, by generalizing from the fact that our neural substrate seems to.
The activation maps of artificial neural networks look like the maps of qualia would in my brain; because the internal steps of the behavior share structure and we share the same physics, I was already granting artificial neural networks “qualia” (but not “personhood”, which I believe arises from counterfactual maximum agency or so).
when I saw michael levin’s talks on bioelectricity, I came to the conclusion that non-neural cell-level patterns also are qualia, and are simply not integrated with the main qualia network. something approximately like, qualia is the feeling of increasing signal to noise ratio by distilling locally intentional, intelligent behavior out of our surroundings and into the networks of interaction between neurons’ weights, synapses, neurotransmitter types in dna. our skin (and entire body map) does a similar job but distilling only the local shape of skin.
evo devo talks about the development from a relatively static perspective, but levin’s work shows that it is in fact highly internally communicative and dynamic, which means there’s a lot more distillation and noise-rejection occurring at runtime; noise rejection, being the basis of intelligence in our own brains in ways that we can observe fairly clearly from neuroscience, gives me significant confidence that our entire body is comparable to neural cellular automata in the way large transformer networks are very vaguely comparable to the brain—probably more vague of a comparison than the NCA one, honestly!
Thank you for sharing your perspective, even though most of it was too technical to understand for me.
I already believed something akin to panpsychism, but with a weighting function towards higher interaction processes of some specific kind I’m not sure I know how to mathematically identify, but which relates closely to integrated information, signal to noise ratio, compressibility, agency.
As I understand, this is your expectations about properties of qualia: some unknown process + integrated information + signal/noise, compressibility, agency.
I believe my post should give you new information (if the post is true) and new ideas.
hmm. Rambling mode, activate!
I already believed something akin to panpsychism, but with a weighting function towards higher interaction processes of some specific kind I’m not sure I know how to mathematically identify, but which relates closely to integrated information, signal to noise ratio, compressibility, agency. I suspect I’m looking for some relationship between those, and of course some of those need additional parameters that an ideal form of the concept selector wouldn’t. it has to accurately match my brain-internal qualia, while also having strong reason to believe it generalizes. Because intelligence arises from a series of simple objectives, a fundamental component of which is usefulness per watt ie a signal to noise ratio towards an agentic target, we can expect that anything that contains optimization will contain some magnitude of consciousness or agency, by generalizing from the fact that our neural substrate seems to.
The activation maps of artificial neural networks look like the maps of qualia would in my brain; because the internal steps of the behavior share structure and we share the same physics, I was already granting artificial neural networks “qualia” (but not “personhood”, which I believe arises from counterfactual maximum agency or so).
when I saw michael levin’s talks on bioelectricity, I came to the conclusion that non-neural cell-level patterns also are qualia, and are simply not integrated with the main qualia network. something approximately like, qualia is the feeling of increasing signal to noise ratio by distilling locally intentional, intelligent behavior out of our surroundings and into the networks of interaction between neurons’ weights, synapses, neurotransmitter types in dna. our skin (and entire body map) does a similar job but distilling only the local shape of skin.
evo devo talks about the development from a relatively static perspective, but levin’s work shows that it is in fact highly internally communicative and dynamic, which means there’s a lot more distillation and noise-rejection occurring at runtime; noise rejection, being the basis of intelligence in our own brains in ways that we can observe fairly clearly from neuroscience, gives me significant confidence that our entire body is comparable to neural cellular automata in the way large transformer networks are very vaguely comparable to the brain—probably more vague of a comparison than the NCA one, honestly!
Thank you for sharing your perspective, even though most of it was too technical to understand for me.
As I understand, this is your expectations about properties of qualia: some unknown process + integrated information + signal/noise, compressibility, agency.
I believe my post should give you new information (if the post is true) and new ideas.