Elon Musk has funded a new artificial intelligence project aimed at finding new ways to compare artificial intelligence and the human brain.
Instead of figuring out how quickly hardware can do calculations, the project measures how quickly a brain or computer is able to send messages within its respective network. This standard could become very useful in measuring the development of AI as it begins to compare with the intelligence of the human brain.
The project itself is called AI Impacts, and it was developed by two Ph.D. students from Carnegie Mellon University and the University of Southern California, Berkeley. The two students have even developed a preliminary method for comparing AI to a brain, which they call traversed edges per second, or TEPS. TEPS essentially determines how rapidly information is passed along a system.
Typically, the TEPS benchmark uses computers to create a graph and search through it. Of course, that's not possible when it comes to the human brain, so the team instead compares the performance of the computer with how often the brain sends electrical signals through its neurons.
"A big pragmatic benefit of measuring the brain in terms of communication is that it hadn't been done before," said Katja Grace, a researcher at Berkeley's Machine Intelligence Research Institute. According to Grace, the method "provides a relatively independent estimate of the price of computing hardware roughly comparable to the brain."
The project was given $49,310 from the Future of Life Institute in Boston, after which the research team was one out of 37 to get a slice of the $7 million in funding for artificial intelligence from Elon Musk. Musk himself is known for funding projects designed to further develop smarter AI, while minimizing possible dangers surrounding it.
The TEPS record is currently held by IBM's Sequoia computer, which achieved 2.3 x 1013 TEPS. AI Impacts estimates that the human brain is around 30 times as powerful as IBM's machine.
Using the calculation, the team was able to figure out how expensive the human brain's performance is in terms of current computer performance, pegging it at around between $4,700 and $170,000 an hour. The team says that TEPS prices could improve by a factor of 10 every four years, meaning that a computer that is $100 an hour to operate could have power comparable to the human brain in between seven and 14 years.
It's important to mention that the researchers are quick to point out a number of "ifs" and "buts" in their equation. For example, the researchers don't know how quickly TEPS' performance might develop when it comes to the computer hardware needed. It's entirely possible that development could significantly slow down in the future.
Despite this, the TEPS benchmark could prove very useful in comparing artificial intelligence with the brain in the future.
Via: IEEE Spectrum
Photo: Karl-Ludwig Poggemann | Flickr