On CBS MoneyWatch: Report: Tiger to Pay Wife $60 Million
BNET Business Network:
BNET
TechRepublic
ZDNet
TalkBack 3 of 4:
Next »
« Previous
re:
"Which is a semi-guided form of self-improvement of software."

More than "semi" currently. Essentially they attempt to move towards pre-set goals, which are pretty well-defined and simplistic.

Also, genetic algorithms have some inherit problems:

-The requirement for many children and many generations for even small improvements means that it often takes a long time for "something" to happen, especially when it has already found a good solution.

-So far, I've yet to see complex behavior emerge from genetic algorithms. All of what I've seen are basically "poke around in the dark" solutions that are quite simple and just optimize a small set of variables. Genetic algorithms do not yet create anything that resembles a step-by-step problem solving process. This is a huge problem for solving more complex tasks which may require many steps to solve, and may not have partial solutions to assist in deciding whether one mutation is colser to solving the problem than anotehr.

-IMHO, it's a mistake to think that simply throwing computational power at the problem can solve it. From what I've seen, they find a good local maximum and stay there. After a certain point, they simply just keep repeating the same thing, with improvements diminishing rapidly. They do not attempt to find alternative solutions for the problems presented, even though their current solution may not completely or even optimally solve the problem.

"I don't think it is far fetched that they will be capable of self-growth within the next 40 years."

I think it is. From what I've seen, current genetic algorithms do not innovate, they simply optimize.

And I have yet to see true sef-replication from raw materials, like humans do. What I've seen is essentially self-assembling. They assemble themselves, but they cannot create their own parts from raw materials.

Those cubes at Cornell are funny, but nothing at all like true self-replication.
Posted by: CobraA1   Posted on: 10/06/05 You are currently: a Guest | Members login | Terms of Use

Alert moderator to an offensive message

Subscribe to this discussion via Email or RSS

Singularity  CobraA1 | 10/05/05
We already have genetic algorithms  Zinoron | 10/06/05
re:  CobraA1 | 10/06/05
Bit of forest for the trees.  Zinoron | 10/06/05

What do you think?

SponsoredWhite Papers, Webcasts, and Downloads

advertisement
advertisement

SmartPlanet

Click Here