Overall, I liked this post. Here’s the primary exception:
Even then: could not this same mind, given only the complete row for Physiology, deduce the contents of Chemistry no less readily?
The compression could easily be lossy. Most higher level concepts are relational in some way- to use an example from image processing, we could build cats out of noses out of edges out of pixels. If we have an idea of how a cat acts on the level of catness, that does not necessarily give us all we need to figure out the components of a cat, or how those components appear visually, or what the basics of the visual field are. If we know pixels, it is easy to come up with edges; if we know edges, it may not be clear if there are pixels or if the underlying image is continuous, for example. More mathematically, if I know the difference of two variables, that does not imply that I can determine what those original variables were.
And typically, when a human makes these sorts of maps, they do so in a reductive (i.e. lossy) way. I mean that in the sense that a molecule is a single element in the chemistry map, but many elements in the physics map; a hand is a handful of elements in the physiology map, but a staggeringly massive number of elements in the chemistry map.
If you were given the complete story of some hunter-gatherer tribe on the moral level- what they did, what they thought about each other, what values they held, and so on- do you think that from just this account a superintelligence could determine the fundamental nature of the particles in their universe? That just seems information-theoretically implausible.
Overall, I liked this post. Here’s the primary exception:
The compression could easily be lossy. Most higher level concepts are relational in some way- to use an example from image processing, we could build cats out of noses out of edges out of pixels. If we have an idea of how a cat acts on the level of catness, that does not necessarily give us all we need to figure out the components of a cat, or how those components appear visually, or what the basics of the visual field are. If we know pixels, it is easy to come up with edges; if we know edges, it may not be clear if there are pixels or if the underlying image is continuous, for example. More mathematically, if I know the difference of two variables, that does not imply that I can determine what those original variables were.
And typically, when a human makes these sorts of maps, they do so in a reductive (i.e. lossy) way. I mean that in the sense that a molecule is a single element in the chemistry map, but many elements in the physics map; a hand is a handful of elements in the physiology map, but a staggeringly massive number of elements in the chemistry map.
If you were given the complete story of some hunter-gatherer tribe on the moral level- what they did, what they thought about each other, what values they held, and so on- do you think that from just this account a superintelligence could determine the fundamental nature of the particles in their universe? That just seems information-theoretically implausible.