Big Blue to build DARPA cat-brain machine
To scale through 'entire range of mammalian intelligence'
US military attempts to develop "programmable neuromorphic" electronic artificial mouse- and cat-bonce brain podules have now moved into gear, with IBM scooping a $5m contract award.
One need hardly specify that the Pentagon office overseeing the Tom'n'Jerryputer push is DARPA, that reassuring rock of madness in an often tediously mundane techno world. The bonkers bad-boy battle boffins of DARPA eat breakfast every day at the Unfeasible Diner, frequently lunch at the Quite Possibly Unnecessary Hotel, occasionally take their aperitifs at the Unforeseen Consequences Saloon - and now and then get thrown out of the Step Too Far Club late at night.
They've certainly pushed the boat out with the Systems of Neuromorphic Adaptive Plastic Scalable Electronics (SyNAPSE) programme, in which they would like someone to build "autonomous, learning neural systems" which can replicate the capabilities of "biological neural systems (eg brains)".
The idea is to achieve the same kind of computing feats that animal brains can, with machines of similar size, weight and power requirements. A cat, for instance, can jump up onto a fence using only binocular vision; a computer able to take stereoscopic vid and accomplish the same feat with four robotic legs would be so heavy as to crush the fence. Likewise, a human can drive a car in traffic - and so can a computer, perhaps. But that computer currently weighs more than 200 tons and requires power levels typical of a warship, not a car.
Hence DARPA's interest in artificial-brain controller machines. For a first step they'd like the SyNAPSE brainputers to perform "at 'mouse' level", then "'cat' level"; but that's just the beginning. SyNAPSE should ultimately "scale over the entire range of mammalian intelligence". That would of course include Einstein, Alexander the Great, Robert Oppenheimer etc.
Against that kind of requirement, the award last week of a mere $4,879,333 to Big Blue appears positively stingy. However, multiple contracts are normal in DARPA efforts of this type. Bigger money should be forthcoming in later stages of the SyNAPSE programme, assuming it gets that far. ®
Sponsored: Virtual application patterns