|
Post by steinarhugi on Jan 31, 2009 14:50:57 GMT -5
What method does CadiaPlayer, the RU GGP engine, use?
In the end of the learning part he mentions new approaches, or "learning procedures", to the representation and manipulation of large amounts of information. What are those?
|
|
|
Post by Björn Vignir Magnússon on Jan 31, 2009 16:16:34 GMT -5
If 2 solutions to a problem seem to be the same in the eyes of a rational agent, how is it determined which one should be picked?
Would a random factor be usable or should a human make the final decision?
|
|
|
Post by Hjalti Magnússon on Jan 31, 2009 18:36:11 GMT -5
Has any progress been made in parallel processing and if so has it been helpful to AI?
Is it self-consciousness paramount for an intelligent agent, i.e. is it not possible to explore the world and have meaningful conversations without being self-conscious?
|
|
|
Post by Alfreð Már Alfreðsson on Jan 31, 2009 18:46:27 GMT -5
1. Has there been any major breakthrough with computers being able to learn since this article was written?
2. You think computers will ever be able to learn as humans do?
|
|
|
Post by Haukur Jónasson on Jan 31, 2009 22:49:22 GMT -5
(Sorry this is a bit late, I hope it doesn't matter too much...)
1) Christian mentioned hearing vs. vision. Has any research been done on using echolocation for robots, instead of relying on sight for 3d-spatial sense?
2) Are any points in this article obsolete, or severely changed? It is written over twenty years ago, and technology has advanced fast in those twenty years.
|
|
|
Post by arnists on Feb 2, 2009 10:29:34 GMT -5
I have a partial answer to a question posted on top of page 2.
Cadiaplayer uses statistical methods and has no knowledge representation to help it along. Given the type of games and length of rounds it is not surprising than the epistomologically aware programs.
The games presented are 'small' and not very emergent.
The length of each round is 30 seconds.
Yngvi S. Björnsson acknowledges that Cadiaplayer does not handle well games that have loops, like chess for example.
As more complex games are presented into the competition and presumably the length of each round increased, statistical methods will not perform as well as epistemic methods. Yngvi agreed with this position wis hen I spoke with him in early January.
My thoughts about this topic are easier to posit with a declaration than a question/query.
Self-consciousness is much easier to attain than general consciousness. Any embodied agent that has a model of itself and sensors that pick up it's action in the world it inhabits is self-conscious. You can even make a simple bitstream self-conscious. In a certain silly way the bitstream of a Turing machine is self-conscious. General Consciousness (capital C) is more tricky. The Penrose-Hammeroff conjecture of quantum superposition decomposition giving rise to general consciousness has a beauty about it worthy of many a debate.
|
|