View this PageEdit this PageAttachments to this PageHistory of this PageHomeRecent ChangesSearch the SwikiHelp Guide

Miscellanious Interesting Papers

Cox, M. T. (2005a). Metacognition in computation: A selected research history and summary. Submitted to Artificial Intelligence CoxAIJ-resubmit.pdf
J. Orwant. EGGG: Automated programming for game generation.
Terra Nova commentary

MM comments (5.23.05):
EGGG is a high-level declarative language for describing game tokens, game state and rules. He’s developed an operational ontology of games; a game is described by specifying such things as: history (how much history player history has to be stored to play the game), synchrony (whether players move all at once or in order), topology, endings (what determines when the game is over), board (the playing surface) and so forth. The system then generates a program implementing the game based on these declarations. It is not a game generator in the same sense that AI story generators are generators. It doesn’t reason about the high-level structure of the game; that structure is completely determined by the human-defined declarations (unlike, say, Minstrel, which generates stories based on a high-level theme). I can imagine using EGGG as the lower-level knowledge representation over which some higher-level reasons. The representative games the author uses in his EGGG examples are: Tic Tac Toe, Chess, Poker, Crosswords, Tetris, Rock Paper Scissors, Deducto, Mammon (text-based stock trading game). I wonder how hard it would be to define a platformer or side-scrolling shooter in EGGG? His representative games are all discrete-turn games (Tetris is the most “real time” of these games, but even here there’s a senses of handling each piece one at a time).

One of the interesting claims about EGGG is that it lets game designers quickly explore design space. It operationalizes the rapid-prototype approach of Will Wright’s team. Perhaps this is a way to support rapid prototyping of AI – by providing a high-level declarative language within which game AI approaches can be tried. But at what abstraction level is it useful to think about and describe behavior?

EGGG provides a generic minimax mechanism for building AI opponents for turn-based games. The Minimax mechanism has heuristics for automatically assigning values to pieces and board positions based on the declarative description of the game.

EGGG automatically learns hidden-markov models of the player’s behavior and uses these to predict player behavior.

Orwant mentions METAGAME as related work: METAGAME is a mathematical framework for generating programs that can play any “symmetric chess-like” game. It is in many ways the inverse of EGGG.
Metagame is mentioned in Orwant's dissertation. Since it's focus is "symmetric chess-like" games, it has limited interest (to MM), but is an example of an abstract framework for describing games.
Henry Lieberman, Hugo Liu, Push Singh, and Barbara Barry (2004). Beating Common Sense Into Interactive Applications. Artificial Intelligence Magazine, 25(4): Winter 2004, 63-76. AAAI Press.
AIMag-BeatingCommonSense.pdf
Daria. A net-art system that dynammically generates new drawings. It's a "distributed software system" that collectively acts as an artist." It's unclear how much real AI is going on and how much is "conceptual AI" (fancy artspeak parading as AI).
Here's the technology text from the extermely irritating-to-interact-with website:



Link to this Page