3 ms·

The first three possibilities depend in large part on improvements in computer hardware. Progress in computer hardware has followed an amazingly steady curve in

The first three possibilities depend in large part on improvements in computer hardware. Progress in computer hardware has followed an amazingly steady curve in the last few decades [17]. Based largely on this trend, I believe that the creation of greater than human intelligence will occur during the next thirty years. Yes, well, single-core clock speeds are a dead end now, and it's not at all clear how to take advantage of multiple cores in the same way. Why are we still excited about this?

simonb 17y ago
Google would like a word with you.
russell 17y ago
The human mind is highly parallel. Anything that approaches it in intelligence will also be highly parallel. Even if clock speeds were a million times faster, parallel would be the only way to manage the complexity. The clock speed of a neuron is only 5ms and we do pretty well.
- frisco 17y ago
  The "clock speed" of a neuron is actually closer to 50ms-100ms (only about 10-20Hz for meaningful purposes) -- but you have about 100 billion of them, and they're all working in parallel. The brain's building blocks are really slow relative to silicon; action potentials move at speeds measured in m/s.
- swolchok 17y ago
  That's parallelism at the "micro" level. Compare it to the parallelism in superscalar or EPIC processors (incidentally, EPIC is also hard to take advantage of). This argument doesn't show that many-core architecture is necessarily going to help.
  - randallsquared 17y ago
    If you can put enough processors and memory close enough, it seems as though you should be able to model a few hundred or thousand neurons on each processor, to assume the worst case. We've already moved past the need for neuron-level parallelism, so from here on the question is what order of magnitude of neurons can be modeled on a single processor. (Again, assuming the worst case that we have to model neurons to get intelligent systems before we can figure out exactly how to produce intelligence in a less simulation-oriented manner).
- goodside 17y ago
  "The human mind is highly parallel. Anything that approaches it in intelligence will also be highly parallel." Birds fly by flapping their wings. Any man-made object that approaches the flying ability of a bird will also flap its wings.
  - Confusion 17y ago
    That analogy fails, because man-made objects have not yet approached the flying ability of birds (or insects). Man-made flying objects are highly energy inefficient and slow and clumsy as a result. We cannot make something that flies like a bird or something that flies like an insect.
jodrellblank 17y ago
What are you suggesting - that Intel's Pentium D, Core Duo, Core 2 Duo, Core 2 Quadro and Core i7 are all exactly the same computing power as a similarly clocked Pentium 4, just with more cores? Clock speeds in current aircooled personal CPUs are sitting where they were several years ago, but branch prediction, pipelining, caching, size reduction, dedicated MIMO/SSIMD units, integrated memory management units, and I don't know what else have all resulted in the ability to compute more in the same time (i.e useful speedup). Why are we still excited about this? Why not? It's exciting.

simonb 17y ago

Google would like a word with you.

russell 17y ago

The human mind is highly parallel. Anything that approaches it in intelligence will also be highly parallel. Even if clock speeds were a million times faster, parallel would be the only way to manage the complexity. The clock speed of a neuron is only 5ms and we do pretty well.

frisco 17y ago
The "clock speed" of a neuron is actually closer to 50ms-100ms (only about 10-20Hz for meaningful purposes) -- but you have about 100 billion of them, and they're all working in parallel. The brain's building blocks are really slow relative to silicon; action potentials move at speeds measured in m/s.
swolchok 17y ago
That's parallelism at the "micro" level. Compare it to the parallelism in superscalar or EPIC processors (incidentally, EPIC is also hard to take advantage of). This argument doesn't show that many-core architecture is necessarily going to help.
- randallsquared 17y ago
  If you can put enough processors and memory close enough, it seems as though you should be able to model a few hundred or thousand neurons on each processor, to assume the worst case. We've already moved past the need for neuron-level parallelism, so from here on the question is what order of magnitude of neurons can be modeled on a single processor. (Again, assuming the worst case that we have to model neurons to get intelligent systems before we can figure out exactly how to produce intelligence in a less simulation-oriented manner).
goodside 17y ago
"The human mind is highly parallel. Anything that approaches it in intelligence will also be highly parallel." Birds fly by flapping their wings. Any man-made object that approaches the flying ability of a bird will also flap its wings.
- Confusion 17y ago
  That analogy fails, because man-made objects have not yet approached the flying ability of birds (or insects). Man-made flying objects are highly energy inefficient and slow and clumsy as a result. We cannot make something that flies like a bird or something that flies like an insect.

jodrellblank 17y ago

What are you suggesting - that Intel's Pentium D, Core Duo, Core 2 Duo, Core 2 Quadro and Core i7 are all exactly the same computing power as a similarly clocked Pentium 4, just with more cores? Clock speeds in current aircooled personal CPUs are sitting where they were several years ago, but branch prediction, pipelining, caching, size reduction, dedicated MIMO/SSIMD units, integrated memory management units, and I don't know what else have all resulted in the ability to compute more in the same time (i.e useful speedup). Why are we still excited about this? Why not? It's exciting.