Two Ph.D. students from the University of California, Berkeley, and Carnegie Mellon University are seeking to determine how soon artificial intelligence might exceed the capabilities of the human brain.
One of Katja Grace and Paul Christiano's first effort toward this goal is finding ways to compare the computational performance of the brain to that of supercomputers. They settled on traversed edges per second (TEPS), a measure of how quickly a computer can move information around within itself.
Comparing the TEPS benchmarks of supercomputers to a rough estimate of how frequently neurons in the brain fire off electrical signals, the researchers found the human brain is at least as powerful and as much as 30 times more powerful than IBM's Sequoia, the highest-ranked supercomputer. They used this estimate to peg the cost of human-equivalent computer performance at between $4,700 and $170,000 per hour, a level that computer hardware might be able to achieve within seven to 14 years.
The researchers say they hope to eventually build "a quantitative model of how fast artificial intelligence research should be expected to grow in an economy with increasing quantities of artificial intelligence available to do research."
From IEEE Spectrum
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA
No entries found