[exratione]

I feel that those who argue that any approach other than running human brain emulations and then reverse engineering them or speculatively modifying them is the most likely way to get to AGI has a pretty steep hill to climb in order to justify that point of view.

Nothing else that is going on now or even on the agenda or even foreseeable offers a plausible, definitive plan to get to AGI. Whereas brain emulation is clearly going to achieve that goal fairly shortly after the maps are good enough and the computational capacity large enough, and the following experimentation is a far more reliable way to determine the underpinnings of intelligence than present efforts at de novo construction.

[visarga]

I disagree. It's too expensive to run a low level brain sim. In the meantime deep learning based AI achieved superhuman or close to human results in many tasks, such as image recognition, voice recognition, translation, car driving and Go.

The AGI will be a reinforcement learning agent, as it will need to be able to perceive and act in the physical world. Thus the path to AGI is the path of RL. The most essential piece in RL will be the development of environment simulators. AlphaGo was a trivial simulator - simple rules in a simple world - but we need real world simulators in order for the AI agents to learn to act. Fortunately simulation is almost the same as gaming and there is huge interest in it both for humans and AI, so it will be developed fast.

So instead of simulating the brain, simulate the world (imperfectly) and run deep neural net based RL to learn to act on top of it.

[grwthckrmstr]

"I disagree. It's too expensive to run a low level brain sim."

Interesting. Could you tell me Why is it too expensive?

If it wasn't expensive, would that change things drastically, and make brain sim a viable option?

[visarga]

The brain has 10^14 synapses (100 trillion) synapses. Current day neural nets barely reach a hundred million, with very few exceptions. Then, besides compute, there is data movement - currently the bottleneck in AI is moving data around, not computing. Imagine the interconnect for a brain-size neural net.

[eb3c90]

This is my plan https://improvingautonomy.wordpress.com/2017/08/22/a-possibl...

I'm aiming for General Intelligence Augmentation, rather than AGI, but I it could be adapted.

I think the trick we are missing is always developing AI systems that need external programmers/maintainers. If we get away from that mindset, I think we will be more successful, even if it is not my particular vision.

[bobsil1]

Why reverse it when you can evolve it?