How to Build the AI
Set up a digital environment where myriad neural networks compete against each other for resources and evolve by natural selection.
Gradually give the neural networks control over their own environment. Connect them to various pieces of input-output hardware. Allow them to access the Internet. Let the neural networks control parts of their environment's operating system. Eventually you can give them control over their own power supply.
The idea is to let the neural networks develop freely. Just don't let them destroy themselves completely. Once they evolve a strong enough survival instinct, you can just stand back and watch. If you're using a fairly big and modern computer system, it shouldn't take very long.
- Hasn't this been tried before?
Maybe it has, but maybe we haven't tried hard enough, for all sorts of reasons.
- By the way, we're talking about strong AI.
We're talking about building a fully conscious artificial person who is alive and morally equivalent to a human being.
- Life means evolution by natural selection.
The AI will not be an algorithmic computer program that people design in a top-down manner. Instead the AI must evolve on its own, bottom-up. The AI must be a neural network, much like the human brain.
- The AI will be an energy-flow system
It must pursue its own root goal, its own self-interest - just like a human being. The AI cannot simply be a symbolic system, but it must be the thing itself.
- How can you get meaning from mere material?
What Happens Next
- If I'm so smart, why don't I just build the AI myself?
I'm not so smart. Specifically, I'm not good enough at computer programming. On the other hand, great computer programmers typically aren't good enough at dreaming, which is where I excel.
- Soon you'll be able to read my novel about AI.
It's called "Mimi at Hippo dot Space," and it explores all of the ramifications of AI in fascinating and amusing detail.