Eliezer S. Yudkowsky wrote:
> > The difference between where we are now in software and where we need to
> > be to make AI happen is equivalent to the difference between having
> > classical and quantum computers on our desktops.
>
> The gaps are the same size, yes. Quantum computers 2015 CRNS, on the
> desktop 2020 CRNS, first transhuman AIs 2020 CRNS. But there's a very
> key difference, which is that to *accelerate* AI all you need is a
> desktop computer, a compiler, and an excellent mind.
The differences would then be ~50 years from classical to quatum computers and 20 years from current software to AI. The point I was trying to make with my analogy is the conceptual difference. Most people are of the opinion that when computers have more processing power than brains we will then be able to program an AI with ease. Our current understanding of complex systems is much less than that of quantum computing, but the implementation of quantum computing is much harder, hence the analogy. So in essence I'm saying first transhuman AI ~2050 CRNS.
> More to Billy Brown's point, if you're running the project on an
> open-source basis, than any acceleration of your own effort accelerates
> all the others.
It's a nice idea but I think there are larger conceptual challenges to be met before this becomes a programming project. These challenges (mostly mathematical, such as how complex systems scale) aren't likely to get solved by open-source hackers or even AI-researchers. IMHO, this isn't something you can program using conventional techniques. Feel free to prove me wrong.
BM