Human cognition is limited in terms of memory, logic, and arithmetic ability, but the brain's flexibility allows people to handle tasks that flummox computers. For example, supercomputers can be designed to master a single game such as chess, but computers cannot easily learn new games--a skill at which even children are adept.
However, Thomas Walter Murphy VII earlier this month unveiled a program that can learn multiple games without any specialized prior knowledge. Available online, the program defeats original Nintendo games such as Super Mario Brothers, learning tricks and trying strategies.
As with all artificial intelligence (AI), the program has difficulty scaling up, in this case to more sophisticated games. Researchers wonder whether an AI system could approximate human flexibility in learning new tasks, and current efforts focus on using big data to imitate humans who know how to complete the task. However big data breaks down when mimicry is not an option, and real-world situations are often nuanced and open-ended.
One approach to making AI more flexible would be to examine the human gift of analogy, say authors Douglas Hoftstadter and Emmanuel Sander. They note that people can apply knowledge of one first-person shooter game to a similar game, for example, whereas a computer will approach each game as entirely new.
From The New Yorker
View Full Article
Abstracts Copyright © 2013 Information Inc., Bethesda, Maryland, USA