Corwin

Edited to remove needlessly inflammatory remark.



[ September 25, 2002: Message edited by: Jeremy Pallant ]</p>

Friar Bellows

In Stanislaw Lem's book, Imaginary Magnitude, which contains nothing but introductions to imaginary books of the future, one of the introductions is to a book on "bitic" literature, or "bitistics", which is literature by non-humans, or in other words, AI. There he describes the so-called Cogito paradox which rests on the question of whether the machine you are conversing with is truly conscious, or is just some perfect simulation of consciousness. Then comes an ironic twist; I'll let Lem describe it: "So the Cogito paradox made itself known to us in bitistics in an ironic and at the same time startling manner: as despair on the part of machines as to whether people really think!"

trientalis

Friar Bellows, thank you so much for your post. I love that book! Wouldn't it be interesting if Golem 13 and Honest Annie could be more than just fiction? Scary, but interesting.

taillessmonkey

Actually, NialScorva, you're incorrect about this. There are no artificial intelligences that function at even a child-like level. It's a bit of a pet peeve of mine, so if I go off on a rant here, bear with me. It's not aimed at at you, it's aimed at that article you posted.

I've seen that Israeli AI "research group" on television a few times with their outlandish claims, and while I strongly suspected that they were full of it, they never gave out enough information for me to be sure. That article does, however.

From the article: "His team of scientists have designed a computer program based on a set of behavioural algorithms that enable the computer to learn language the same way humans do, starting from scratch.

Up until now, computer analysts have provided programs with built-in grammatical rules. This is the first time a program has been based on the developmental language learning of humans."

It clearly says here that their "child" HAL is a system that learns grammar rules. In addition, further down the article, the only purpose Mr. Dunietz discusses using HAL for is user interface.

He also clearly avoids saying that the program is as intelligent as a human baby, saying instead that "the computer's language skills mirror those of an 18-month-old toddler."

From this, I fell fairly secure in concluding that HAL isn't actually an intelligent system at all, simply an extremely inefficient method in building a natural language parser and generator, albeit one that is overly hyped to the uneducated press. I can't decide whether or not the fact that it passes a turning test at an 18 month old level is impressive or not. There are natural language parsing and generation systems that can provide information and hold conversations at a much higher level, but only within specific domains.

There are simply no artificially intelligent systems that reach a level anywhere near that of a humans at all. And there won't be for a long time to come. Even the simplest problems (such as natural language and vision processing) are *impossible* to solve, not only with current hardware and algorithms, but with the hardware and algorithms available in the forseeable future.

The best we can do is process small amounts of information in a limited domain. Generalization is the current Achille's heel of artificial intelligence research. We simply don't have any way to take an algorithm that performs well in a specific domain, and generalize to perform on everything. The more complex problems (cognition and information storage) we've made very little progress on at all. On actual machine consciousness, we don't even know enough about any sort of existing consciousness to attempt to create it.

I don't think we'll see artificial intelligence at the level of a human within my lifetime. I work with some very highly advanced AI systems on a daily basis, and as impressive and amazing as they are sometimes, they don't even register on the scale of intelligence at all when compared with a human being. It's not that they're "dumb", it's worse than that. They simply do not contain much in the way of actual intelligence.

Perhaps if we move off of silicon to biological computers, the new hardware, and the new theories of computation that I would presume would come along with the new hardware medium, amazing progress that cannot be forseen now would occur. Also, progress in neuroscience could open up avenues not currently considered. But I still wouldn't bet on seeing human level artificial intelligence during my lifetime.

-tail

American Agnostic

I would be VERY careful about saying that somethings is impossible considering how many time such claims have been wrong in the past. That Said:

1. We're definatly going to have things like nano-scale filters, ultrastrong materials (carbon nanotubes), chemical/biologic detectors able to be woven into clothing, etc. I think that all the high and lofty promises that proponents make will come to pass within 30-40 years and will fundementally change society. Nanotech may well be the answer to many of today's problems.

2. Human-level intelligence and higher will definitly be achieved by 2050 if not sooner. Why should we think that there is something special about human intelligence? Except for our egos that is.

3. The human lifespan will continue to increase and will probibly hit 100 in developed countries within our lifetimes. With the biotech revolution coming into full swing in the next few decades we should be able to reprogram our bodies to live longer.

4. Absolutly! The only thing to hold us back is lack of motivation! I predict sometime in the next 500 years the first interstellar colonization missions will be lauched (if we survive that long we'll be able to along with eachother without huge defense budgets and have more money for importent issues). In the meantime we'll have colonized the Solar System and almost certainly started to terraform Mars, maybe even Venus.

I know what you mean. I want to stick around and see it happen!