Talk of clusters and distributions brings to mind an inherently correlational/statistical definition of classes. However, one can also think of it in a more causal sense; penguins are birds because they descend from birds (and so automatically end up having the sorts of traits that birds have).
These causal definition would be invariant to the sorts of cluster-remaking that you suggest, because you can’t change the past.
But then how do we make sense of the relationships of belonging that you describe?
Basically, by reversing the hierarchy: penguins are not panguens, instead panguens are penguins. Or more specifically, they are penguin-mammal hybrids, descending from both classes through genetic engineering. So the genetic engineering lets you violate some things that have been invariant (well, approximately, ignoring horizontal gene transfer, which AFAIK has played a relatively smallish role?) from evolution, such as species always descending from a single species, rather than merging multiple distinct species together.
In the case of lizard-mammals, I guess the causal perspective is somewhat trickier. In a way the selection procedure introduces a class of “non-human mammal” that didn’t exist before? And then the mammal-lizard hybrids descend from both of these.
I don’t know whether preserving the categories against declustering/reclustering like this is at all useful, but I though it should be mentioned that it can be done.
It seems like this would have issues once you want the AI to e.g. influence the world in a way that increases the number of happy humans—then you can’t just say “and also ‘happy humans’ have to have an origin causally separated from AI intervention.”
You wouldn’t say “and also ‘happy humans’ have to have an origin causally separated from AI intervention”, you’d say “and also ‘happy humans’ have to have an origin causally downstream from humans”.
Hm. But that doesn’t seem to stop the AI from creating molecular smiley faces whose precise form is causally dependent on what humans are like. The reason the causal origin argument helps for birds is because we can specify a model of the causal origin, and it’s just evolution with no room for AI intervention.
Talk of clusters and distributions brings to mind an inherently correlational/statistical definition of classes. However, one can also think of it in a more causal sense; penguins are birds because they descend from birds (and so automatically end up having the sorts of traits that birds have).
These causal definition would be invariant to the sorts of cluster-remaking that you suggest, because you can’t change the past.
But then how do we make sense of the relationships of belonging that you describe?
Basically, by reversing the hierarchy: penguins are not panguens, instead panguens are penguins. Or more specifically, they are penguin-mammal hybrids, descending from both classes through genetic engineering. So the genetic engineering lets you violate some things that have been invariant (well, approximately, ignoring horizontal gene transfer, which AFAIK has played a relatively smallish role?) from evolution, such as species always descending from a single species, rather than merging multiple distinct species together.
In the case of lizard-mammals, I guess the causal perspective is somewhat trickier. In a way the selection procedure introduces a class of “non-human mammal” that didn’t exist before? And then the mammal-lizard hybrids descend from both of these.
I don’t know whether preserving the categories against declustering/reclustering like this is at all useful, but I though it should be mentioned that it can be done.
It seems like this would have issues once you want the AI to e.g. influence the world in a way that increases the number of happy humans—then you can’t just say “and also ‘happy humans’ have to have an origin causally separated from AI intervention.”
You wouldn’t say “and also ‘happy humans’ have to have an origin causally separated from AI intervention”, you’d say “and also ‘happy humans’ have to have an origin causally downstream from humans”.
Hm. But that doesn’t seem to stop the AI from creating molecular smiley faces whose precise form is causally dependent on what humans are like. The reason the causal origin argument helps for birds is because we can specify a model of the causal origin, and it’s just evolution with no room for AI intervention.