A “powerful enough” AI isn’t magic: it cannot recover information that no longer exists
Technically, it can of course—through inference. Any information we have recovered about our history—history itself—is all inference used to recover lost information.
Even with successful cryonics, you still end up with a probability distribution over the person’s brain wiring matrix—it just has much lower variance, requiring less inference/guesswork to get a ‘successful’ result (however one defines that).
Agreed with your last paragraph that crossing the uncanny valley will be difficult and there is much room for public backlash. It’s so closely related to AI tech that one mostly implies the other.
A “powerful enough” AI isn’t magic: it cannot recover information that no longer exists
Technically, it can of course—through inference.
Sounds like Hollywood image enhancement, where a few blurry pixels are magically transformed into a pin-sharp glossy magazine photograph.
I could point out that if you can infer the information, then by definition it still exists, but the real point here is just how powerful an AI can be and what inferences are possible. Let’s say that yesterday I rolled a dice ten times without looking at the result. Can a “powerful enough” AI infer the numbers rolled? Is the best-fit reconstruction of someone’s mind, given an atom-by-atom scan a century from now of a body frozen by Alcor today, good enough to be a mind?
I could point out that if you can infer the information, then by definition it still exists,
This is not real?y true.
When typing the above sentence, I removed a letter and replaced it with a ?. You can probably infer what the originally intended letter was, thus using inference to recover information that did not exist anywhere in your physical locality.
But yes this is a terminology/technicality, and agreed that
the real point here is just how powerful an AI can be and what inferences are possible
Let’s say that yesterday I rolled a dice ten times without looking at the result. Can a “powerful enough” AI infer the numbers rolled?
Yes and no. A powerful enough AI in the future can recreate many historical path samples (ala monte carlo sim) through our multiverse.
Of course, if the information was just erased and didn’t effect anything, then it doesn’t matter. It literally can’t matter, so the AI doesn’t even need to infer/resolve that part of space-time—any specific choice for the die roll is equally as good, as is an unresolved superposition . There may be a connection here to delayed choice quantum eraser experiments.
Is the best-fit reconstruction of someone’s mind, given an atom-by-atom scan a century from now of a body frozen by Alcor today, good enough to be a mind?
I imagine that will completely depend on the details of their death, the delay, and the particular tech used by Alcor at the time they were frozen.
That being said, in a century powerful SI seems quite possible/likely. There are huge economies of scale involved in simulations. It is enormously less expensive—in terms of per human reconstruction cost—to do a historical simulation/reconstruction for all of the earth’s inhabitants at once.
The SI would use DNA (christendom has done a great job over the millenia at preserving an enormous amount of DNA), historical records, all of the web data from our time that survives, and of course all of the alcore data. It could have the equivalents of billions of historians working out the day by day details of each person’s lives before constructing more detailed sims, etc etc. It would be the grand megaengineering project of the future, not some small scale endevour.
Technically, it can of course—through inference. Any information we have recovered about our history—history itself—is all inference used to recover lost information.
Even with successful cryonics, you still end up with a probability distribution over the person’s brain wiring matrix—it just has much lower variance, requiring less inference/guesswork to get a ‘successful’ result (however one defines that).
Agreed with your last paragraph that crossing the uncanny valley will be difficult and there is much room for public backlash. It’s so closely related to AI tech that one mostly implies the other.
Sounds like Hollywood image enhancement, where a few blurry pixels are magically transformed into a pin-sharp glossy magazine photograph.
I could point out that if you can infer the information, then by definition it still exists, but the real point here is just how powerful an AI can be and what inferences are possible. Let’s say that yesterday I rolled a dice ten times without looking at the result. Can a “powerful enough” AI infer the numbers rolled? Is the best-fit reconstruction of someone’s mind, given an atom-by-atom scan a century from now of a body frozen by Alcor today, good enough to be a mind?
This is not real?y true.
When typing the above sentence, I removed a letter and replaced it with a ?. You can probably infer what the originally intended letter was, thus using inference to recover information that did not exist anywhere in your physical locality.
But yes this is a terminology/technicality, and agreed that
Yes and no. A powerful enough AI in the future can recreate many historical path samples (ala monte carlo sim) through our multiverse.
Of course, if the information was just erased and didn’t effect anything, then it doesn’t matter. It literally can’t matter, so the AI doesn’t even need to infer/resolve that part of space-time—any specific choice for the die roll is equally as good, as is an unresolved superposition . There may be a connection here to delayed choice quantum eraser experiments.
I imagine that will completely depend on the details of their death, the delay, and the particular tech used by Alcor at the time they were frozen.
That being said, in a century powerful SI seems quite possible/likely. There are huge economies of scale involved in simulations. It is enormously less expensive—in terms of per human reconstruction cost—to do a historical simulation/reconstruction for all of the earth’s inhabitants at once.
The SI would use DNA (christendom has done a great job over the millenia at preserving an enormous amount of DNA), historical records, all of the web data from our time that survives, and of course all of the alcore data. It could have the equivalents of billions of historians working out the day by day details of each person’s lives before constructing more detailed sims, etc etc. It would be the grand megaengineering project of the future, not some small scale endevour.