DARPA in Tom'n'Jerry robo-brain quest
First they came for the mice; and I said nothing
Pentagon boffinry chiefs are on the verge of inking a deal which could see US forces equipped with "neuromorphic" imitation brain modules, each potentially as intelligent as a cat.
Wired magazine reports that HRL Labs of Malibu initially announced a firm deal on the feline-grade artibrain contract, referred to by the US military as Systems of Neuromorphic Adaptive Plastic Scalable Electronics (SyNAPSE). However the company later corrected this to say that actually negotiations with the government continue.
One need hardly say that the US gov tentacle in question is none other than DARPA, the famed warboffin bureau which often sneers at ivory towered, white coated mainstream science, preferring instead the straitjacket, the padded cell, the dungeon laboratory and the insane foam-lipped laugh.
The SyNAPSE project seems like a good example of this. According to DARPA, regular electronic computers are no good for dealing with real-world problems.
Modern electronics has evolved through a series of major developments (e.g., transistors, integrated circuits, memories, microprocessors) leading to the programmable electronic machines that are ubiquitous today ... these machines are of limited utility in complex, real-world environments ... As compared to biological systems for example, today’s programmable machines are less efficient by a factor of one million to one billion ... biological neural systems (e.g., brains) autonomously process information in complex environments by automatically learning ... neuromorphic electronic machines would be preferable in a host of applications ...
For instance, your regular cat can easily jump up onto a fence requiring only binocular video to do so. A robot felinoid combat system using normal computers would need some kind of 3D mapping lasers or something, would still be likely to cock up the jump, and its brain would be so large and power-hungry that the droid moggy would be the size and weight of a bull moose - probably destroying the fence even if it were successful.
Hence the quest for the artificial neuromorphic self-programming cat-bonce box. DARPA says in its proposal that the "final deliverable" hardware would be "a multi-chip neural system" which would enable robotic platforms to perform "at 'cat' level".
Intriguingly, the DARPA boffinry chiefs specify that before the "go" decision to develop a cat-level neuromorphobrain can be given, the selected contractors must first develop a "'mouse' level" system. However, they don't want to see any kind of bizarre, multimillion dollar robotic Tom and Jerry antics, no sir.
The animal tokens (e.g. “mouse”) ... are indicators of complexity; they are not intended to specify particular architectural requirements or environmental behaviors.
But there is a chilling warning buried deep in the DARPA outline (word doc). The big brains at the Pentagon aren't just looking to put cats and mice out of a job. They are also looking for "a means to scale the complexity of these tasks over the entire range of mammalian intelligence".
Read the Wired report here, and tremble at the imminent obsolescence of your puny mammalian intelligence. ®
Sponsored: Hyper-scale data management