Hard Light Productions Forums
Off-Topic Discussion => General Discussion => Topic started by: Grey Wolf on June 15, 2002, 06:43:34 pm
-
http://www.cnn.com/2002/TECH/ptech/06/09/common.sense.computer.ap/index.html?related
I don't know why, but this interested me...
-
I would like to see the results of this; the problem of human "common sense" is now being attacked from two ends. ;)
-
yes, all very well, but what about metaphors and stuff that all students are made to study in English? How does Cyc know when someone's being literal or when they mean something else? humans are complex things.
I don't think AI will work with the stuff we have today... it'd probably need some sort of technological revelation we haven't discovered yet to work...
-
Depends whether its a common or not metaphor and the context in which it is used.
-
There's too much to understand about human speach alone for a computer to understand. You could never get something that runs on 1's and 0's to fully understand the different shades of tones used in human speach to create and effect. Humans can learn and guess about things they dont understand, when a computer comes across something it doesnt know, it'll most likely crash.
In conclusion: computers are tools for humans - they could never be a real human.
pete
-
Agreed. Something that can only execute rules, no matter how flexible those rules might be, will never have an imagination.
-
You people need to read The Age of Spiritual Machines. :D After all, imagination itself is just an execution of rules. ;)
-
Cyc is a dead end. The project has gone on since 1984 and very little useful has resulted from it.
At best we have network scanning tool (that's not as good as a trained sysadmin watching the same traffic), a search engine query tool (that consistently returns less relavent results than Google) and the so called 'Knowledge Server', which has yet to show its worth.
The problem is that the they're trying to model an analog, nondeterministic process (thought, common-sense, drawing inferences based on loosely/nonrelated facts) on a digital, deterministic system. Computers are incapable, at the most basic level, of things like lateral thinking, hunches, etc.
Computers are only capable of discrete, pure values: 0 and 1. How do you model the subtle gradations of 'maybe'? You can't do it without a perfect computer with infinite registers. Common sense is about as likely as artificial intelligence: it won't happen on transistorised hardware.
-
We don't even know how to model a human brain as a pure simulation.
There's an outside possiblity that the unpredictable nature of quantum computer IC's could yeild a more reasonable AI system, but my bet is on neural/carbon based machines.
Of course...the rate everyones' trying to make sure we never get any useful genetics work done might knock that on the head.
-
The problem is that the they're trying to model an analog, nondeterministic process (thought, common-sense, drawing inferences based on loosely/nonrelated facts) on a digital, deterministic system. Computers are incapable, at the most basic level, of things like lateral thinking, hunches, etc.
Computers are only capable of discrete, pure values: 0 and 1. How do you model the subtle gradations of 'maybe'? You can't do it without a perfect computer with infinite registers. Common sense is about as likely as artificial intelligence: it won't happen on transistorised hardware.
That is what fuzzy logic is for, with which it should be fairly easy to model human thoughts. Most neurologists today agree that "lateral thinking" and "hunches" are just the thought processes carried out by the subconscious portion of the brain (the part without good monitoring and control systems) where the results are sent to the main brain without the reasoning, so that it seems "intuitive." The current system of processors may or may not be adequate for this task but really the only thing missing there is a true random generator, which can be done using results from elementary particle movements. (that is, if the probabilistic view of the universe is the correct one)
-
Before you know it, you'll have a SHODAN on your hands, trying to destroy us all.
-
Freaky stuff.
-
Originally posted by CP5670
That is what fuzzy logic is for [rest of speculative comment supressed]
There are no examples of 'fuzzy logic' circuits in existence that do not devolve to two-state switches.
Fuzzy logic is a wonderful idea, but you're not doing it on a transistorised system, like a digital computer.
-
Hmmmm... my father's working a little on that.... I've heard he's using Bayse Nets and Probabilistic Inference to do things similar to that... can't give you specifics though :p
-
There are no examples of 'fuzzy logic' circuits in existence that do not devolve to two-state switches.
Fuzzy logic is a wonderful idea, but you're not doing it on a transistorised system, like a digital computer.
That is perfectly okay though, as long as the system works correctly. Even our own brains work on similar types of electrical pulses and therefore probably use the same 0-1 system at their cores along with a randomizer.
I doubt there will be a sci-fi type of man against machine conflict anytime in the future, but rather that man and his creations will gradually merge over time and eventually become indistinguishable from one another.
-
Originally posted by CP5670
I doubt there will be a sci-fi type of man against machine conflict anytime in the future, but rather that man and his creations will gradually merge over time and eventually become indistinguishable from one another.
Bah, sound like self asimilation:p we all become borgs:D
-
yep, that's what we're headed for; read that book I mentioned earlier. :D