"Web May Hold the Key to Achieving Artificial Intelligence"
Washington Post (09/06/02) P. A1; Cha, Ariana Eunjung; Drezen, Richard A.
The controversy over artificial intelligence and how the Internet could shape it has been re-ignited by the advent of "chatterbots" such as Active Buddy's SmarterChild, a marketing tool that can "talk" to people via real-time text messages and answer questions by tapping into the Web's vast informational resources. A few such bots have progressed to the point that they can pass the Turing test and convince people they are actually human. Some experts worry that the ultimate goal of turning the Internet into a "global brain" by developing the technology further would give computers too much information and power, and lead to a "hive mind" that erodes freedom and individual expression. However, several challenges must be solved before such a goal is even possible: Computers cannot easily read Web pages, and lack common sense. World Wide Web Consortium director Tim Berners-Lee is trying to overcome the first problem by leading an effort to link keywords and tags to text, sounds, and images. Meanwhile, Push Singh of MIT's Media Lab and others are trying to solve the common sense problem by building a "knowledge base" in which volunteers contribute observations about commonsense behavior. At the Free University of Brussels, scientists are leading an international initiative to make computers understand how humans access data online, thus allowing them to comprehend the interactions between people, objects, and concepts. The work was pioneered by Old Dominion University's Johan Bollen, who conceived of a program that studies how people seek out information, and then streamlines the process. http://www.washingtonpost.com/wp-dyn/articles/A43363-2002Sep5.html
"A Theory of Evolution, for Robots"
Wired News (09/05/02); Sandhana, Lakshmi
Chalmers Institute of Technology researchers Krister Wolff and Peter Nordin have conceived of a winged robot that can learn how to fly on its own; such a design circumvents scientists' own lack of knowledge about the mechanics of flight. The robot's creators turned to genetic programming to generate the instructions fed into the machine that helps it learn how to lift off. After being assessed, the most successful liftoff instructions were paired up, while randomly swapping instructions between these optimal pairs sired next-generation "offspring" commands that were transmitted to the robot. Techniques that the robot has learned include standing on its wings, pulling itself up by grasping convenient objects, and a more effective flapping methodology. The University of Reading's David Corne explains that evolutionary computation is key to solving control problems, such as what sequences of movements will allow a robot that is heavier than air to lift off and stay airborne. Meanwhile, engineering difficulties--the fast and flexible movement of joints--also need to be solved. The barrier to developing effective flying robots is a lack of funding for testing designs so that improved models can be developed, according to Corne. Nordin believes that such robots could be in operation within three years if enough funding is provided, and he also says that such research is important if better-performing aircraft are to be developed. http://www.wired.com/news/technology/0,1282,54900,00.html
Posted on Sep 11, 2002, 4:54 PM from IP address 4.43.240.222