Science-fiction writers once envisioned pocket devices that could make calls using voice and video, search immense databases, and also think.
Smartphones made the first few goals reality and some recent apps can do low level thinking for you. Now, MIT researchers have come up with a computer chip that will be able to power a truly mobile AI.
The term "neural networks" isn't new to any fans of Star Trek or other science-fiction fare, but now it has also become reality. These networks are "large virtual networks of simple information-processing units, which are loosely modeled on the anatomy of the human brain," according to Larry Hardesty of the MIT News Office. Neural networks are graphics processing units, or GPUs, as opposed to CPUs, which are the brains of computer motherboards.
The new chip, dubbed Eyeriss, is 10 times more efficient than industry standard GPUs today. Vivienne Sze, an assistant professor in MIT's Department of Electrical Engineering and Computer Science, explained that this translates into letting the chip process information and make decisions "locally," rather than needing to work through the Internet. The key is how it lowers the amount of times the mobile device needs to interface with distant memory banks, which usually requires an Internet connection. This saves on time needed to complete the operation and data usage charges, in addition to adding a new layer of privacy for the user.
Eyeriss was partially funded by DARPA, a Departnet of Defense agency that encourages development of technologies for military use. Using Eyeriss in a drone can be help it to recognize targets on the ground and independently relay the information to nearby troops.
Google has been working on neural nets for a while. As far back as 2012, the company was reported by the New York Times to have started working on networking GPUs. Approximately 16,000 processors were let loose on the Internet to search for cats, never having been trained or programmed to know what a cat was. It was able to come up with a "dreamlike digital image" of a cat based on searching different images and YouTube videos. Google has also created TensorFlow, an Open Source Library for Machine Learning. Google hoped making it open source would help other minds to help in the development of better models of neural networks. This type of machine learning is what makes self-driving cars possible.
Is the world ready for a phone with a personality that is able to make decisions it knows you'd want? The technology isn't there yet, but this new chip brings society a big step closer to having personal digital assistants in the future. It's also a lot closer to a "War Games" style scenario of playing "global thermonuclear war." Hopefully, researchers will concentrate more on the former than the latter.