DARPA Wants Your Brain
In a robot body
DARPA Wants Your Brain
In a robot body
by MATTHEW GAULT
The Defense Advanced Research Projects Agency, that bastion of fringe military science that gave us GPS, the Internet and cinder block-hurling killbots, is at it again. In August, the agency put out a request for information for research and development of a cortical processor, a machine that mimics the human neocortex.
The neocortex is the best part of the human brain. It’s the bit that’s responsible for all of our higher reasoning skills like language and conscious thought. Why in the world would DARPA want to replicate the human brain?
It’s pretty simple: human brains are fucking awesome.
Capturing complex spatial and temporal structure in high-bandwidth, noisy, ambiguous data streams is a significant challenge in even the most modern signal/image analysis systems. Current computational approaches are overwhelmingly compute intensive and are only able to extract limited spatial structure from modest quantities of data. Present-day machine intelligence is even more challenged by anomaly detection, which requires recognition of all aspects of a normal signal, in order to determine those parts that do not fit. New approaches, based on high capacity, low power implementations, must be developed.
Despite all of the crazy technological leaps in the past hundred years and all of the computational power that can be crammed into a microchip that sits on the end of a finger, there are still things that people are way better at than machines.
We can program a robot that routinely beats the grandmasters of chess, but we can’t build a robot that roles with the punches if the rules of that game changes midstream. Human are just better at processing all of the different little bits of sense data that are flying around at any given time.
DARPA’s got a pretty ingenious answer to the problem:
“DARPA is examining a new approach based on mammalian neocortex, which efficiently captures spatial and temporal structure and routinely solves extraordinarily difficult recognition problems in real-time.”
Build machines modeled after the human neocortex. The part of the brain that helps us make sense of all the disparate information we encounter every day.
Not everyone is sold on the idea, however.
“DARPA never ceases to amaze me with its audacity,” Noel Sharkey told WIB via email. Sharkey is a professor of robotics and artificial intelligence at the University of Sheffield in England. He’s also one of the founding members of the Campaign to Stop Killer Robots — a group of individuals and NGOs, including Human Rights Watch, who’ve worked for the past year to heighten awareness of various governments plans to build autonomous killing machines.
If anyone is going to know if DARPA’s plan will create machines that think like humans, it’s Sharkey. But he’s unimpressed.
“Don’t forget this is just a research idea,” Sharkey wrote. “That does not mean that it is possible. Honestly the number of times I have heard people say that the neural algorithms are developing will be at the level of a mouse or a cat over the last 25 years. None of these have been successful. It is a bit like putting out a call for the development of a computer architecture that emulates the human brain and then imagining that just by saying it, that it will have that functionality.”
He’s right. A request for information is just that — DARPA wants the best and brightest out there to toss them information about how such a thing could be done. In particular, it’s looking for software, algorithm, hardware and systems experts to write in and tell them what creating a computer based on the human neocortex might look like.
DARPA has a history setting its sights on wild, crazy, aspirational targets and when you shoot for big you fail big. For every success it’s had there have been even more failures. During the Vietnam era the agency attempted to create a mechanized elephant to assist soldiers in clearing out the jungle and in the 1970s attempted to train a group of telepathic spies to gain an intelligence advantage against the Soviet Union. Neither project came to anything productive.
Time will tell if their cortical processor is another GPS or another embarrassing group of psychic soldiers.