3 ms·
He remarks about the need to simulate the entire brain. However, at the moment, artificial neurons do really well while being a tiny minuscule proportion of the
by return0 11y ago
He remarks about the need to simulate the entire brain. However, at the moment, artificial neurons do really well while being a tiny minuscule proportion of the real neurons in the brain. It's possible neural nets capture the level of description required to reproduce intelligence and we don't need to go any deeper. Learning in real neurons is extremely more messy and probably more distributed, but that does not mean that all this mess is necessary to improve current models.
- visarga 11y agoI agree, we might do OK with less than the power of the brain. On the other hand Hinton also says that we need to invest more computation in order to achieve AI with common sense. So, it would seem, even if we don't need as much power as the brain has, our road ahead is still tied to the advancement of computation. It's not a matter of guessing the right learning algorithm.
- fla 11y agoOn the other hand, the human brain cosumes approximately 20 Watts.
- return0 11y agoyou mean it's power efficient? Consider that it takes the brain, let's say, ~5 years to learn the stuff that a visual NN learns in days.
- argonaut 11y agoHe did not say we need to simulate the entire brain. He said neural nets would need about the same number of parameters. Not the same thing.