Linked by Eugenia Loli on Sun 9th Sep 2007 21:06 UTC
Geek stuff, sci-fi... Technologists and investors gather at the two-day Singularity Summit in San Francisco to discuss the benefits and risks of advancing artificial intelligence--and what to do in the event that machines one day out-think humans.
Thread beginning with comment 269949
To read all comments associated with this story, please click here.
Strong (general) AI
by michield on Mon 10th Sep 2007 13:23 UTC
Member since:

We do have an idea how to implement general AI. The theory is well-defined by Marcus Hutter:
Universal Artificial Intelligence: Sequential Decisions Based On Algorithmic Probability

The AI model AIXI converges fast towards the best solution that there is for a problem. The only problem in the approach is that it requires infinite resources, therefore corners have to be cut. There are approaches that can do this and they show behaviour that is expected of intelligent systems. The systems that will be implemented in the future will be able to redesign themselves.

Basically the theory is this:
1) Make many random monkeys (programs), and give more time to the monkeys that best approach your program in the simplest way (Occams razor and Kolmogorov complexity).
2) Select the monkey that writes a word well and give him a banana (reward).
3) Breed this monkey into variants that write lines well.
4) Breed the succesful monkey (give it more probability) and give him a banana once he writes shakespeare.
In the meanwhile data-mining like techniques are used to build a world model and a model that checks what consequences actions of the robot itself are on the environment. This self-improving model will be used to prove which strategies (programs) best approach the goal.

These techniques are used by people who have succesfully learnt robots to tie knots or when you give them the order "fetch ball", run
to the ball and take it, then bring it back to the one who gave the command. The techniques can also be used for driving or stocks.
Look for example to Novamente ( ) or NARS (non axiomatic reasoning system).

A book I am reading now by Goertzel called Artificial General Intelligence is very good, it contains an overview of much research in the field of strong general AI.

Reply Score: 2

RE: Strong (general) AI
by axilmar on Tue 11th Sep 2007 10:23 in reply to "Strong (general) AI"
axilmar Member since:

Call me back when an android like Star Trek's lt Cmdr Data exists. Until then, talk is cheap.

Reply Parent Score: 1