Humanoid robot learns how to run

[Althornin]

I believe emulating the brain with computer hardware is the wrong way of going about things. Just for starters, a computer will probably never be able to compete with brains on a neuron-by-neuron basis; there's just too many of the little buggers and with too many cross-connections.

The brain evolved out of a specific need using tools available to nature. No point in trying to copy it using dead matter... It'll probably just be slow and clumsy and extremely expensive.

I'll go on record saying you are 100% wrong.
neural nets (emulating the brain, in a way) are the most likely way to achieve AI.
all we require is great enough silicon density to simulate sufficient interconnections, and unless you believe we are gonna hit a "wall", that time will be around 2036+- some years, hence the predictions....

Take a massive neural network, and attach some inputs (random) and train it. It could easily learn to do quite complex tasks. I've only dealt with (written) very small neural networks, but the results are very good when training.
If they scale properly, good AI is inevitable.

 

[London Geezer]

I believe emulating the brain with computer hardware is the wrong way of going about things. Just for starters, a computer will probably never be able to compete with brains on a neuron-by-neuron basis; there's just too many of the little buggers and with too many cross-connections.

The brain evolved out of a specific need using tools available to nature. No point in trying to copy it using dead matter... It'll probably just be slow and clumsy and extremely expensive.

I'll go on record saying you are 100% wrong.
neural nets (emulating the brain, in a way) are the most likely way to achieve AI.
all we require is great enough silicon density to simulate sufficient interconnections, and unless you believe we are gonna hit a "wall", that time will be around 2036+- some years, hence the predictions....

Take a massive neural network, and attach some inputs (random) and train it. It could easily learn to do quite complex tasks. I've only dealt with (written) very small neural networks, but the results are very good when training.
If they scale properly, good AI is inevitable.

What we're saying is that you could have an artificial brain which has the same number of transistor (neurons), or more than a normal brain. In the end that artificial brain will still need what our brain has "learnt" in the last 20000 years or so (or 6000 for the Creationists).

Our brain is what it is today because of centuries of evolution, dictated by the environment and by our own body. Our movements are much more linked to our nervous system than most people think.

Of course, the Artificial brain can "learn". But without a body as complex as ours it can only learn and remember how many States are in the USA and the names of the governor in each one of them in 0.003 seconds. Which might be useful, but i think we can all agree we are much more than that.