>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.
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.