Slashdot Mirror
MIT Creates Chip to Model Synapses
MrSeb writes with this excerpt from an Extreme Tech article: "With 400 transistors and standard CMOS manufacturing techniques, a group of MIT researchers have created the first computer chip that mimics the analog, ion-based communication in a synapse between two neurons. Scientists and engineers have tried to fashion brain-like neural networks before, but transistor-transistor logic is fundamentally digital — and the brain is completely analog. Neurons do not suddenly flip from '0' to '1' — they can occupy an almost-infinite scale of analog, in-between values. You can approximate the analog function of synapses by using fuzzy logic (and by ladling on more processors), but that approach only goes so far. MIT's chip is dedicated to modeling every biological caveat in a single synapse. 'We now have a way to capture each and every ionic process that's going on in a neuron,' says Chi-Sang Poon, an MIT researcher who worked on the project. The next step? Scaling up the number of synapses and building specific parts of the brain, such as our visual processing or motor control systems. The long-term goal would be to provide bionic components that augment or replace parts of the human physiology, perhaps in blind or crippled people — and, of course, artificial intelligence. With current state-of-the-art technology it takes hours or days to simulate a simple brain circuit. With MIT's brain chip, the simulation is faster than the biological system itself."
The problem is not providing such components, nor get them to work like the original nor getting it into your head. The real problem I see is interfacing with the rest of the brain.
Because, let's face it, that's something every coder knows: Interfacing, working and supporting legacy systems just sucks.
get them to work like the original
Is this really something that we could do in the foreseeable future ? My understanding is that the brain programs itself (or we program it if you like) during the first years of our lives (5 to 7) for the most part. An empty new 'brain part' would act just like some parts of the brain act after a stroke I suspect, meaning that it'll take years and years to (re)train it.
Similarly, children that grew up with animals alone, with little or no interaction with other humans (there were some cases) are never able to learn to speak fluently, because that part of the brain never fully develops (ie. is never programmed).
AFAIK we don't know enough about how the brain works to pre-program such components and it would need to be strongly tuned to the destination brain, otherwise it won't work very well or at all. We know about the lower-level stuff (neurons, synapses) and some things about the higher-level (regions and general functions), but not much in between (though, I'm not a specialist).
Even so, I can see some medical uses for this, for people with disabilities. Though nothing like what you see in 'Ghost in the Shell'.
The analog nature of the neuron isn't really the key to making "artificial brains" - the problem is simply scale.
Agreed.
We will never be able to produce enough of these chips and tie them together well enough to produce anything conventionally interesting
Shall we cue here all the "never" predictions of the last century? By the year 1900 there were lots of experts predicting we would never have flying machines, by 1950 experts were predicting the whole world would never need more than a dozen computers.
Moore's law, or should we say Moore's phenomenon, has been showing how much electronic devices scale in the long run.
I think you have to credit MIT researchers for knowing better where the cutting edge is than you, and the writers of the article for including the 1960s in this paragraph:
'Previously, researchers had built circuits that could simulate the firing of an action potential, but not all of the circumstances that produce the potentials. “If you really want to mimic brain function realistically, you have to do more than just spiking. You have to capture the intracellular processes that are ion channel-based,” Poon says.'
More than just spiking; from my AI lectures years ago I recall that the McCulloch-Pitts neuron model of the was a spiking model (excitatory inputs, inhibitory inputs, thresholds) etc.
I might be out of date, but: the event itself requires the neuron's action potential to reach a threshold, then the synapse fires. It either fires or it does not. On or off. But the process of reaching the firing threshold is analog, since the physical geometry of the neuron and of its afferent neural feeds (inputs) determines at what point the neuron will fire. Neurotransmitter quantities in the synapse are also modifiable though eg by drugs and natural up/down regulation of receptors, enzymes or re-uptake inhibition. So a neuron is an analog computer having output with various amplitudes of on/off.
MIT’s chip — all 400 transistors (pictured below) — is dedicated to modeling every biological caveat in a single synapse. “We now have a way to capture each and every ionic process that’s going on in a neuron,” says Chi-Sang Poon, an MIT researcher who worked on the project.
Just because you finally can recognize the letters of the alphabet doesn't mean you can speak the language.