Today IBM announced a “cognitive computiing chip” that emulates the human brain. Designed with help from DARPA and four major universities, the prototype may be the first step towards reasoning computers.
Dharmendra Modha, project leader for IBM Research, said the new chips have parts that behave like digital “neurons” and “synapses” that make them different than other chips. Each “core,” or processing engine, has computing, communication and memory functions. “You have to throw out virtually everything we know about how these chips are designed,” he said. “The key, key, key difference really is the memory and the processor are very closely brought together. There’s a massive, massive amount of parallelism.”
Read the Full Story.
I have yet to read more detailed information on this chip, but until my fears are shown to be unfounded I think it’s necessary to toss a bit of cold water over the sensationalism in this announcement.
Saying that this chip ‘thinks like a brain’ is like saying the blank canvas I have at home ‘looks like the Mona Lisa.’ The article says that memory and processing are brought very close together, which is certainly great and is pretty much what we see with distributed memory models, albeit on a much, much smaller scale.
It also mentions digital ‘neurons’ and ‘synapses’ but doesn’t go into detail – my guess is they’re VERY simplified models and there are real questions as to whether you can get ‘brain-like’ function out of overly simple models. I’d also assume they’re using the same neuron model for all neurons, despite the brain having multiple types of neurons. There’s also no talk of the sorts of structure one sees in the brain. While some contend that emergent properties may spring up with a ‘critical mass’ of neurons, most neuroscientists will tell you that the brain -every brain!- has a specific structure which, it stands to reason, exists for a reason. I’d guess this chip has, roughly, a very simple multi-layer ‘neural network’ (from computer science, not neuroscience) layout, as opposed to some true mirroring of structures seen in the brain.
In short, this is a neat experiment at putting lots of processing elements with minimal local memory together on a chip, and that sort of chip allows you to do some pattern-based processing that often has the unfortunate misnomer of ‘neural network’ computing. It’s neat. It’s not simulating the brain.
On a final note, the article ends mentioning IBM’s 2009 announcement about simulating a cat’s brain – this other bit of wildly exaggerated PR led to a bit of a spat between Dharmendra Modha and Henry Markram, a neuroscientist working with IBM’s Blue Brain project. It’s useful to read, so here’s a link:
http://www.eweek.com/c/a/IT-Infrastructure/Rival-Scientist-Calls-IBM-Cat-Brain-Simulation-a-Scam-384096/
I’ve met both men and they’re both perfectly kind and interesting people (and I’m not connected to either project), and I firmly believe that science needs to happen at both the high-level and the low-level scales in order to be productive. But doing something at the high-level and making sweeping statements that outrageously over-state the nature of the work does a disservice to the name of science, especially given the increasingly scientifically-illiterate public.