7 Sep 2008
Beyond AI
No computer has ever been designed that is aware of what it’s doing, but there is no reason why it should not be possible to build a large distributed system of computers which is aware of it.
Even if we know how to build a plane or a computer, it is nearly impossible for a single person to do it from scratch. It’s an engineering problem, and obviously even experienced engineers need the right building blocks and the right tools to do it. Achieving true AI is similar: even if we know how to do it, it is nearly impossible for a single person to do it right now.
As far as one can tell today, to achieve real AI, you will need to cross a complexity threshold. It is unclear how many servers and datacenters you will need to cross the complexity threshold for a new kind of AI, but the number is probably above 100,000 servers – roughly the number of servers which Microsoft and Google use already today in their datacenter for search purposes.
We have also increasingly complex virtual worlds and MMORPGs which resemble more and more the real world. If we combine both, the power of the datacenter with the virtual worlds, then new forms of AI should be possible. We would need..
* ..the combined processing power of a
large number of huge data centers..
* ..to control an adaptive agent which
lives and is embedded in..
* ..a virtual world which is complex
enough to support AI
see also our wiki page: Beyond AI, the Flickr datacenter photo is from the Flickr user jspaw