Scientists plan to implement aspects like pattern recognising

[chhitiz]

i was reading this article on AI and i have a question:
how do scientists plan to implement aspects like pattern recognising(and not just simple pattern recognition, but even linking images from two totally unrelated fields, like we sometimes link mental images, i hope u get it), or spontaneity and creativity?
how far have they gone?

 

[junglebeast]

Recognition and classification algorithms are quite good if features can be extracted that are invariant under the desired transformations. In the case of recognition in images, this is quite difficult because nothing is really invariant under projection and illumination change, so even recognizing identical objects can be extremely difficult. However, we have some stuff that works OK.

Artificial Intelligence is separate, and deals with the logical aspects of reasoning rather than interpretation of sensory information to recognize patterns etc. It is assumed that some other "hardware" has already processed this input and parsed it into more meaningful information tokens which can then be dealt with in a logic framework.

So far AI has yet to live up to its name. There is nothing that resembles true intelligence, creativity, or emotion -- actually, it is yet to be shown that this is even possible, as it may be nothing more than an illusion in humans. We do have machines that can solve simple planning problems, but only if those problems can be posed into a very formalized in rigid framework -- and we don't really have a good way of transforming observations into this logic framework. Because we don't have machines that can actually think, we can't even begin to work on adding things like spontaneity or creavity..although if we did have machines that could think, this would probably be a trivial addition. Instead what you may see are toy problems where randomness is used to give the cursory illusion of human-like qualities.

 

[chhitiz]

what about common sense? the amount of common axioms that we take for granted is so huge that it don't think that can be stored in a small memory. is it even possible for us to make a computer learn, like we make a baby learn?

 

[junglebeast]

what about common sense? the amount of common axioms that we take for granted is so huge that it don't think that can be stored in a small memory. is it even possible for us to make a computer learn, like we make a baby learn?

A simple program that records all its input in memory could be considered "learning" the data. A simple one pole moving average filter could be considered to learn the average of a stationary series. A neural network can "learn" arbitrary associations through back propagation or other update schemes. Hebbian learning can be used as well.

Basically, learning is trivial...but it is only meaningful if you have an intelligence that can use that learned information.

Common sense information (with exception to natural instincts) is learned through experience in humans, and any good AI system would also need a system for learning such as this that would allow it to develop without needing a bunch of axioms being manually programmed in. Common sense rules are not strictly "if then" statements, either..everything is relative and has a degree of uncertainty.

 

[cswave]

Unless the computer replicates the human brain to a very high resolution, it is not likely that it will ever seem completely human to a trained observer. But there are scientists currently trying to replicate the human brain, cell by cell, in a computer. And if nanotechnology ever does become advanced enough, then maybe someday the contents of ones brain could be downloaded into a computer.