Posted on 07/12/2006 6:35:37 PM PDT by annie laurie
US advanced technology solutions firm BBN Technologies has been awarded $5.5 million in funding from the Defense Advanced Research Projects Agency (DARPA) for the first phase of the Integrated Learning Program.
Over the next four years BBN will develop an artificial intelligence capability called “Integrated Learner” that will learn plans or processes after being shown a single example. The total value of the project, if all four years of the development programme are completed, could be up to $24 million.
The goal is to combine specialised domain knowledge with common sense knowledge to create a reasoning system that learns as well as a person and can be applied to a variety of complex tasks. Such a system will significantly expand the kinds of tasks that a computer can learn.
Under the contract, administered by the US Air Force Research Laboratory, BBN’s first year research will focus on military medical logistics planning. The specific goal is completion of a simulation which involves evacuating wounded military personnel and civilians from Fallujah, Iraq to hospitals in Germany and Kuwait.
Successful demonstration of this system will have implications beyond the ability to automate the medical evacuation planning process, by providing the groundwork for automated systems capable of learning other tasks of similar complexity. This will enable a capacity to develop more effective military decision/planning support systems at lower costs and that require less training for human users.
“This programme attacks one of the biggest problems in AI,” said Mark Berman, vice president, BBN Technologies, “The Integrated Learner will combine traditional machine learning techniques with an AI reasoning system capable of understanding behaviour it observes only once. This ambitious goal is necessary because, for many of the current and future complex military tasks that could benefit from automation, there simply are not many examples in existence. Although there has been some research into this area, this will be the first deployed system with the capacity to apply general knowledge and reasoning to a task.”
BBN Technologies is best known for pioneering the development of the ARPANET, the forerunner of the Internet.
Eigentlich läuft die Übersetzungssoftware im Web auf ein Kauderwelsch hinaus, wenn komplizierte Idiome verwendet werden, weil ein Computer den Sinn eines Durchgangs wörtlich übersetzt. Ein Computer, ist zum Beispiel, dazu nicht fähig, Slang wie zu übersetzen, "sind Sie, Baby irre!" Als in diesen Engländern zur deutschen Übersetzung des obengenannten Paragraphen.
As you can see, the Machine Translation fails to capture the idiom of the English slang in German. And there is no way to acertain the German expression is idiomatically as opposed to literally correct. Its the former that makes for a common sense language to language text translation.
(The Palestinian terrorist regime is the crisis and Israel's fist is the answer.)
You raise some good points. And then there's also the possibility that you just have to be alive to be intelligent, no matter how much processing power and memory storage you have at your disposal.
It might be possible that in terms of common sense and actual awareness of itself in the world, even the fastest supercomputer will always be dumb as a bag of hammers.
For all the points I raised, I still think that truely intelligent machines are possible. I just don't think I'll see them in my lifetime or that the current computer architectures/programming paradigms can support them. That doesn't mean there shouldn't be research. A lot of people in the field focus on hardware and reasoning algorithms--not as many focus on how to represent, organize, and associate information (the CYC program is one notable exception)--which is key. I also think there is much promise in using the WWW as a starting point for AI research.
Developing AI is (perhaps) the hardest research problem ever--by several orders of magnitude. When you think about it, it all about people figuring out how they are wired and how to replicate their own archaic programming.
Because of the above, progress will be very slow and lots of $$ will be wasted on dead ends--but that is the nature of research. Unfortunately, you don't know what will work apriori. Also, unfortunately, the field attract a lot of charatans, probably because no one expects really a lot of success.
"All bogus. AI has been living on its promises for nearly 40 years."
Not entirely. They figured out how to play chess.
"Developing AI is (perhaps) the hardest research problem ever--by several orders of magnitude. When you think about it, it all about people figuring out how they are wired and how to replicate their own archaic programming."
It took approx. 50 years to solve the chess problem. Others will be harder.
Slang is a finite set of special cases and sufficiently sophisticated software can catch them. "Groovy" and "baby" together in the same sentence strongly implies the 60's meaning, rather than an infant with grooves in it.
It took approx. 50 years to solve the chess problem. Others will be harder.
Yeah, and in many ways the Chess problem was EASY. A clean, well ordered environment and well defined rules. Of course, the thing that made it difficult was the exponential number of possibilities that had to be searched. Even though they beat the world champ just recently, Chess programs have been able to beat 99.9% of people since the mid-80s.
The best chess thus far is played by ruthless calculating machines, not by some mechanized analog of intuition.
The problem is that computers have no sensory context with which to understand the symbols being fed in. CYC is no exception. All they and the rest are doing is taking meaningless abstract symbols and munging them into other meaningless abstract symbols. The sensory context to gain meaning from the gibberish must come from the machine interacting with the world, not a machine being spoon-fed symbols.
Just to followup on that, try getting a computer to play Go.
Won't run under windows! Too many "features" for AI.
"...I give you, ROBOCOP 2!..." (robot steps through door, screams and rips own head off)
Don't throw the baby out with the bath water. AI is, certainly, the most over-hyped area in Computer Science. I daresay that most of those in the field make one false assumption: that human intelligence can be recreated in silicon. I firmly believe this to be impossible, that a computer is the sum of its parts, but the human mind is much more than the sum of its parts.
That said, AI has been a fruitful field. Have you ever used/seen a handheld that converted the "handwriting" into characters on screen? Those are usually done with artificial neural nets, a methodology invented within the last forty years.
This is an interesting point but I would contend that this is precisely what human beings go through to learn anything. And we make a LOT of mistakes on the way to perfection.
First we learn some very basic fundamentals; spatial sense; sense of being; sense of others. This is infant-hood.
Later we learn to walk, talk, play, which is toddler-hood.
We grow and learn for the next several years, which correlates nicely from childhood through the teens, before becoming truly functional adults. Many never get that far (Dems, LoL!).
True AI will probably take a similar track to learning. Expecting out-of-the-box AI is too much. It might not take 18 years to get a functioning AI, but I'd say at least a couple years of learning would produce a better product.
Skynet may be fiction, but men flying through the air, rocket ships to the moon, and cell phones used to be fiction too.
Sophisticated artificial intelligence is on its way. Just because it hasn't been achieved yet doesn't mean it isn't coming. Moore's Law and related developments mean that our computing/ AI capabilities are increasing not just by addition, but exponentially.
Not that we can see it now, so clearly. But sooner or later the visible effects of the exponential curve are goign to kick in.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.