Thursday, March 11, 2010

AboutAI

The Artificial Intelligence Community

Archive for January, 2008

Codeplay: Portable High-Performance Compilers

Posted by admin On January - 26 - 2008

Codeplay’s Sieve C++ Parallel Programming System is a scalable programming system aimed at those who need to create C++ code suitable for use on a multi-core processing environment. The Sieve System consists of an extension to a C++ compiler, a multi-core linker and a runtime to schedule the processes.

The Codeplay Sieve C++ partitioning system has been built on top of our proven VectorC C++ core technology.

Source:

http://www.codeplay.com/technology/sieve.html

Popularity: 1% [?]

Stalking Blinky, Pinky, Inky, and Sue

Posted by admin On January - 25 - 2008

It had to happen some time. Two researchers from Eötvös University in Budapest, Hungary, have developed a computer program that can play the video game Ms. Pac-Man better than an average human player. The New Scientist Tech reports that the AI program developed by András Lörincz and István Szita was allowed to develop it’s own strategies for navigating the mazes, gobbling up dots, and avoiding the deadly touch of colored ghosts. The goal of creating this program was to demonstrate the weaknesses in AI versus human intelligence, especially in the arena of video games. Lörincz and Szita have also published their findings in the Journal of Artificial Intelligence Research.

There are no details, in the New Scientist Tech article, about the interface of the AI program and the video game. I suspect some internal interface to the game console. However, a set-up that actually had an independent and external “player” would be quite an achievement. Besides the internal strategies, this external agent (robot?) would need visual recognition to view and interpret the current status of the game from the screen. Then, there would need to be the computations required to control the arm operating the joystick and buttons based on the visual input and the internal strategy routines. By my count, this is at least three cores’ worth of processing.

There are thousands of uses for autonomous agents that can take in visual data, interpret their surrounding environment, and then use learned strategies to navigate and complete assigned tasks. Robotic cars driving around the dessert or through town are flashy demonstrations, but are they biting off too much all at once? Maybe navigating through a smaller, more controlled environment–like a maze inhabited by fruit and pesky ghosts that chase you–might be a more manageable start to unlock the methods for bigger endeavors.

Popularity: 1% [?]