4 ms·

I would say it probably doesn't take a lot of computational mass to simulate a person. We're each doing it right now with a glob of meat the size of a grapefrui

I would say it probably doesn't take a lot of computational mass to simulate a person. We're each doing it right now with a glob of meat the size of a grapefruit, and that's just what evolution was able to come up with. So simulating a universe where minds are a low-level primitive, and everything else is simulated lazily and using lots of shortcuts is reasonably plausible. But that's not the kind of world we find ourselves in. Our minds appear to be constructed of the same stuff as the rest of the world. If we're to accept the simulation argument, then our definition of "simulated reality" must imply a physics implementation that is high enough fidelity to support minds as high level structures - you have to actually be able to build compact working brains in the simulation. Alternatively, we'd have to suppose that a brain is actually just a very convincing facade, usually sitting inert until someone observes it, at which point it reflects the inner workings of the person's mind at whatever level of detail is necessary to convince the observer that it is the mechanism behind the mind. Both of these seem far less plausible.

jamesaguilar 13y ago
> But that's not the kind of world we find ourselves in. Our minds appear to be constructed of the same stuff as the rest of the world. If we're to accept the simulation argument, then our definition of "simulated reality" must imply a physics implementation that is high enough fidelity to support minds as high level structures - you have to actually be able to build compact working brains in the simulation. Devils advocate, but there could simply be level-of-detail heuristics, where the mind is explainable in terms of physics, but not implemented that way (unless you look). Similarly, for people who take damage to their brains, the model for that person's new brain could simply be derived during that simulation step and substituted. That's just a high level view of how a lazy simulation could still produce minds that appear to by implemented using physics.
- mistercow 13y ago
  That's what I was saying in the "Alternatively, ..." part. But this does suggest an avenue of testing the simulation hypothesis that might be more useful than looking at purely physical anomalies. The physics engine might be simplified, but edge cases will likely be subtle and difficult to distinguish from gaps in our theories of physics. But if the simulation is using heuristics to gauge whether a brain is being observed, and making it correlate to the mind primitive backing it as needed, then we should eventually be able to construct scenarios where those heuristics are wildly inaccurate. If we could, for example, trick the simulation into presenting a completely inactive brain while its was conscious, that would be informative.
  - jamesaguilar 13y ago
    Agreed although presumably our hypothetical simulators have sysadmins and could simply ratchet up the fidelity around anyone who is looking.
    - mistercow 13y ago
      Yeah, there are a number of ways that it could be thwarted. Any civilization that can achieve this kind of universe simulation would probably also have solved general AI, so the "heuristics" we're trying to fool may in fact be an intelligence far more clever than ourselves. But evidence is never perfect.
      - jamesaguilar 13y ago
        Actually in a universe where time doesn't necessarily flow forward, evidence could be perfect. They could just delete branches where the simulation is discovered and revert back to the point where the bit leading to the discovery was made. But I see what you are trying to say.
  - pixl97 13y ago
    Who says we haven't tricked the simulation, and who says we can be tricked? If we do exist in a simulation, not understanding the motivations and nature of the perpetrators of said simulation leave us with no means to tell if our results are unadulterated. Lets say you find incontrovertible proof that we are being simulated. You are so exited you pull through a red light on the way home and are killed by a bus. In the mean time the hard drive with your data on it crashes and the data is lost. Another researcher attempts to follow your work, performs the same experiment and gets a negative result, we are not being simulated (little does he know he's running on Universe.exe patch level 4.0.2113). A universal virtual machine could operate just like virtual machines we use now. If the occupants of said simulation screw stuff up we can always be rolled back to a checkpoint with no knowledge of such an occurrence. Slight changes could be made to the problematic actors in the simulation and chaos theory would likely make the simulation diverge.
eli_gottlieb 13y ago
>I would say it probably doesn't take a lot of computational mass to simulate a person. We're each doing it right now with a glob of meat the size of a grapefruit, and that's just what evolution was able to come up with. And in this case, evolution was exceptionally clever, since it managed to design the world's most powerful, most parallel computation devices in maybe a couple kilograms of meat. Consider that the computational power necessary to emulate a human brain accurately is conjectured to be measured in petaflops. I mean, sure, you could invoke some kind of high-level computational heuristics, like we did to emulate the Nintendo 64 back in the day, but when dealing with living things I wouldn't call that a very good idea.
- mistercow 13y ago
  >And in this case, evolution was exceptionally clever, since it managed to design the world's most powerful, most parallel computation devices in maybe a couple kilograms of meat. That is almost certainly not the case. The estimate for emulating a brain is about 37 petaflops. The fastest supercomputer is currently China's Tianhe-2, at nearly 34 petaflops. You're certainly going to have more than 10% overhead for emulating a brain (more on this in a moment). But beyond that, the brain is a specialized computing device, and at certain specialized tasks, it is currently unrivaled. However, this may largely be a matter of algorithms rather than raw power. >Consider that the computational power necessary to emulate a human brain accurately is conjectured to be measured in petaflops. Emulation requirements are an exceptionally poor indicator of computational power. Emulating an SNES loosely requires a CPU capable of around 350 MIPS (FLOPS don't work for obvious reasons). Emulating an SNES accurately requires a CPU capable of over 50,000 MIPS. The CPU of an actual SNES does 1.5 MIPS. It has other components working in parallel, but we'd be generous to estimate 10 MIPS for the whole system. Why the discrepancy? Well first off, there's the overhead of simple translation. But more importantly, the SNES has several components working in parallel, and the timing of how these components interact is critical. And that's emulating hardware with a handful of components, designed by rational people. The brain has massive parallelism, and was created by a blind optimization process. So even if whole brain emulation can really be accomplished with power around 37 petaflops, doing the actual work of a human brain (without emulating it) can be done with far, far less. The emulation overhead is probably orders of magnitude greater than that of an SNES, but if it were the same, we'd be looking at single-digit teraflops to implement a mind as powerful as a human's. That's achievable right now for a few thousand dollars.

jamesaguilar 13y ago

> But that's not the kind of world we find ourselves in. Our minds appear to be constructed of the same stuff as the rest of the world. If we're to accept the simulation argument, then our definition of "simulated reality" must imply a physics implementation that is high enough fidelity to support minds as high level structures - you have to actually be able to build compact working brains in the simulation. Devils advocate, but there could simply be level-of-detail heuristics, where the mind is explainable in terms of physics, but not implemented that way (unless you look). Similarly, for people who take damage to their brains, the model for that person's new brain could simply be derived during that simulation step and substituted. That's just a high level view of how a lazy simulation could still produce minds that appear to by implemented using physics.

mistercow 13y ago
That's what I was saying in the "Alternatively, ..." part. But this does suggest an avenue of testing the simulation hypothesis that might be more useful than looking at purely physical anomalies. The physics engine might be simplified, but edge cases will likely be subtle and difficult to distinguish from gaps in our theories of physics. But if the simulation is using heuristics to gauge whether a brain is being observed, and making it correlate to the mind primitive backing it as needed, then we should eventually be able to construct scenarios where those heuristics are wildly inaccurate. If we could, for example, trick the simulation into presenting a completely inactive brain while its was conscious, that would be informative.
- jamesaguilar 13y ago
  Agreed although presumably our hypothetical simulators have sysadmins and could simply ratchet up the fidelity around anyone who is looking.
  - mistercow 13y ago
    Yeah, there are a number of ways that it could be thwarted. Any civilization that can achieve this kind of universe simulation would probably also have solved general AI, so the "heuristics" we're trying to fool may in fact be an intelligence far more clever than ourselves. But evidence is never perfect.
    - jamesaguilar 13y ago
      Actually in a universe where time doesn't necessarily flow forward, evidence could be perfect. They could just delete branches where the simulation is discovered and revert back to the point where the bit leading to the discovery was made. But I see what you are trying to say.
- pixl97 13y ago
  Who says we haven't tricked the simulation, and who says we can be tricked? If we do exist in a simulation, not understanding the motivations and nature of the perpetrators of said simulation leave us with no means to tell if our results are unadulterated. Lets say you find incontrovertible proof that we are being simulated. You are so exited you pull through a red light on the way home and are killed by a bus. In the mean time the hard drive with your data on it crashes and the data is lost. Another researcher attempts to follow your work, performs the same experiment and gets a negative result, we are not being simulated (little does he know he's running on Universe.exe patch level 4.0.2113). A universal virtual machine could operate just like virtual machines we use now. If the occupants of said simulation screw stuff up we can always be rolled back to a checkpoint with no knowledge of such an occurrence. Slight changes could be made to the problematic actors in the simulation and chaos theory would likely make the simulation diverge.

eli_gottlieb 13y ago

>I would say it probably doesn't take a lot of computational mass to simulate a person. We're each doing it right now with a glob of meat the size of a grapefruit, and that's just what evolution was able to come up with. And in this case, evolution was exceptionally clever, since it managed to design the world's most powerful, most parallel computation devices in maybe a couple kilograms of meat. Consider that the computational power necessary to emulate a human brain accurately is conjectured to be measured in petaflops. I mean, sure, you could invoke some kind of high-level computational heuristics, like we did to emulate the Nintendo 64 back in the day, but when dealing with living things I wouldn't call that a very good idea.

mistercow 13y ago
>And in this case, evolution was exceptionally clever, since it managed to design the world's most powerful, most parallel computation devices in maybe a couple kilograms of meat. That is almost certainly not the case. The estimate for emulating a brain is about 37 petaflops. The fastest supercomputer is currently China's Tianhe-2, at nearly 34 petaflops. You're certainly going to have more than 10% overhead for emulating a brain (more on this in a moment). But beyond that, the brain is a specialized computing device, and at certain specialized tasks, it is currently unrivaled. However, this may largely be a matter of algorithms rather than raw power. >Consider that the computational power necessary to emulate a human brain accurately is conjectured to be measured in petaflops. Emulation requirements are an exceptionally poor indicator of computational power. Emulating an SNES loosely requires a CPU capable of around 350 MIPS (FLOPS don't work for obvious reasons). Emulating an SNES accurately requires a CPU capable of over 50,000 MIPS. The CPU of an actual SNES does 1.5 MIPS. It has other components working in parallel, but we'd be generous to estimate 10 MIPS for the whole system. Why the discrepancy? Well first off, there's the overhead of simple translation. But more importantly, the SNES has several components working in parallel, and the timing of how these components interact is critical. And that's emulating hardware with a handful of components, designed by rational people. The brain has massive parallelism, and was created by a blind optimization process. So even if whole brain emulation can really be accomplished with power around 37 petaflops, doing the actual work of a human brain (without emulating it) can be done with far, far less. The emulation overhead is probably orders of magnitude greater than that of an SNES, but if it were the same, we'd be looking at single-digit teraflops to implement a mind as powerful as a human's. That's achievable right now for a few thousand dollars.