Whereas conventional chips are wired to execute particular "instructions," the TrueNorth juggles "spikes," much simpler pieces of information analogous to the pulses of electricity in the brain. Spikes, for instance, can show the changes in someone's voice as they speak—or changes in color from pixel to pixel in a photo. "You can think of it as a one-bit message sent from one neuron to another." says one of the chip's chief designers. The chips are designed well not for training neural networks, but for executing them. This has significant implications for consumer AI: big companies with lots of resources could focus on the training, which individual TrueNorth chips in people's gadgets could handle the execution.
Where do we go from here? Is it even reasonable to order $50k worth of components and put together our own high-performance, reasonably-priced blade cluster? Or is this folly, best left to experts? Who are these experts if we need them?
And what is the better choice here? 16-core Opterons at 2.6 GHz, 8-core Xeons at 3.4 GHz? Are power and thermals limiting factors here? (A full rack cupboard would consume something like 25 kW, it seems?) There seems to be precious little straightforward information about this on the net.
Alex King is director of the Critical Materials Institute, a part of the U.S. Department of Energy's Ames Laboratory. CMI is heavily involved in making rare earth minerals slightly less rare by means of supercomputer analysis; researchers there are approaching the ongoing crunch by looking both for substitute materials for things like gallium, indium, and tantalum, and easier ways of separating out the individual rare earths (a difficult process). One team there is working with "ligands – molecules that attach with a specific rare-earth – that allow metallurgists to extract elements with minimal contamination from surrounding minerals" to simplify the extraction process. We'll be talking with King soon; what questions would you like to see posed? (This 18-minute TED talk from King is worth watching first, as is this Q&A.)
Now, a Chinese team has successfully implemented this artificial intelligence algorithm on a working quantum computer, for the first time. The information processor is a standard nuclear magnetic resonance quantum computer capable of handling 4 qubits. The team trained it to recognize the difference between the characters '6' and '9' and then asked it to classify a set of handwritten 6s and 9s accordingly, which it did successfully. The team says this is the first time that this kind of artificial intelligence has ever been demonstrated on a quantum computer and opens the way to the more rapid processing of other big data sets — provided, of course, that physicists can build more powerful quantum computers.