This week, IBM scientists and their university partners announced a breakthrough in cognitive computing research: the simulation of a brain the size of a cat’s. Using 144 terabytes of RAM and almost 150,000 processors, scientists were able to model a neural cortex with 1 billion simulated neurons and 10 billion synaptic connections. The researchers have subsequently been awarded $16.1 million for completion of phase 0 of DARPA’s SYNAPSE project. The goal of SYNAPSE (Systems of Neuromorphic Adaptive Plastic Scalable Electronics), according to DARPA, is “to develop electronic neuromorphic machine technology that scales to biological levels.” In layman’s terms, their goal is to put simulated brains onto microchips.
While a disembodied brain on a microchip is itself interesting as an object of study, the real question is: what happens when this brain is embodied? The unstated intention of the SYNAPSE project is the embodiment of simulated mammalian brains into robots. Before we lose ourselves to rapturous visions of snuggly pet purrbots, we should consider the possible range of behaviors of not only the feline brain, but of mammalian-size brains in general. Cats are predators, as well, after all. The brain controls an animal’s body and behavior, and much of its function would be superfluous to a robotic replica (presumably, for example, these robots would not able to mate and multiply). If individual behaviors can be added and removed willy-nilly, then any given mammalian behavior could be isolated and reproduced onto a microchip. Feline predation is just one possibility, and a scary one at that. Imagine hordes of 14ft-tall steel tigers rampaging through your suburban neighborhood, programmed with insatiable hunger and an instinctual urge to maul and devour bipeds (did I mention that robots can be carnivorous?).
The presentation of this cortical simulation as the equivalent of a feline brain lends an aspect of cutesyness to the project, but this cutesyness, of course, is just a facade. It allows us to envision it as something benign. Right now, there’s no harm in a disembodied brain with no behaviors sitting on a hard drive in Livermore, CA, being trained to distinguish images of corporate logos to the awe of of fawning scientists. But what happens when they let the cat out of the bag (sorry)? Such a virtual brain could be trained with any given set of potentially distasteful behaviors and embodied in a machine of practically limitless size and strength. This is a DARPA project, after all, and as much as we might want snuggly purrbots, we should be aware that what we might get is marauding swarms of giant predatory killbots.
Joe McFall received his undergraduate degree in Linguistic Anthropology from Emory University in 1998, and went on to receive two Masters degrees, in Linguistics and Artificial Intelligence, from the University of Georgia in 2005 and 2007. He holds a lifetime interest in culture and counterculture, fringe science, paranormal, alternate history and conspiracy theory. He works as a software developer and lives in Atlanta, GA. He is also, alongside our buddies Raymond Wiley and Austin Gandy, one of the hosts of Disinformation: The Podcast, Disinformation World News, and Out There Radio.