Neuron-Like Computer Hardware Finally Gets Software
Over the last few years, engineers have proposed a number of arcane materials, such as graphene or quantum computer chips, as alternatives for the silicon chips that run modern computers. With the development of actual software to go along with the futuristic hardware, circuit components called memristors have now taken the lead in the race to replace silicon. Even more promisingly, memristors behave like neurons in many ways, allowing scientists to use this software to create a kind of digital brain.
As first reported in IEEE Spectrum, the Modular Neural Exploring Traveling Agent (MoNETA) project looks to combine the artificial intelligence software research of Boston University's (BU) neuromorphics laboratory with the memristor technology developed by Hewlett-Packard. DARPA, the U.S. military's mad scientist wing, will underwrite the project, all with the goal of producing viable artificial intelligence within the next couple of years.
Memristors are special because unlike the electrical components in a regular chip , their resistance to electrical flow depends on the current that previously flowed through the component. Through this relationship, a memristor can "remember" a piece of information, just like a biological neuron. By connecting memristors in a more networked fashion than resistors in a silicon chip , computer scientists could create a system not too dissimilar in function to a flesh-and-blood brain.
Now, before everyone starts out on a quest to kill SkyNet, remember that artificial intelligence doesn't mean "as smart as a human." MoNETA's current research focuses on basic navigation and resource acquisition, a level of intelligence far below that of even an ant.
And rather than fueling android dreams of mechanical pets, the BU researchers see memristors neural networks as a path towards ensuring that computers continue to speed up at their current rate.