Slashdot Mirror
Artificial Synapse Created For Synthetic Brain
Zothecula writes "It's probably still going to be a while before autonomous, self-aware androids are wandering amongst us. That scenario has come a little closer to reality, however, with researchers from the University of Southern California having created a functioning synapse circuit using carbon nanotubes. An artificial version of the connections that allow electrical impulses to pass between neurons in our brains, the circuit could someday be one component of a synthetic brain."
It's probably still going to be a while before autonomous, self-aware androids are wandering amongst us.
Sure, that's what they'd like us to believe, anyway...
(crap - forgot to post this anonymously!)
#DeleteChrome
If these so-called synapses can't spontaneously self-assemble and inter-connect, they still have a LOT of work to do to achieve real AI. A more likely practical use for this was mentioned in another article about it: repairing damaged tissue in human brains (or perhaps deliberately "re-wire" portions to alter function).
But does this mean that they have discovered a way to "plug in" a computer to a brain? Can these be used as an adapter to "talk" with neurons?
All we need is about a quadrillion of them for a brain.
Maybe if we only used half or may be a third of that?
The first AI capable of acting as a CEO is still an achievement.
...The AI you're speaking of has a distinct lack of the I part. We can fake things along quite nicely, sure, but we're still ages away from true artificial intelligence. Slapping some cameras in front of a giant if/then/else statement ain't it.
"Although there are a multitude of variations in synapses, we have modeled a typical cortical synapse. Action potentials, the signals from other neurons that arrive at the synapses are about a millisecond in duration and about 100 mV in amplitude. Under certain conditions, the synapse responds with an output potential of around 5-10 mV that lasts around 10 ms. Thus the synapse slows and spreads the effect of the action potential, synchronizing its effect with other action potentials, since not all action potentials arriving at the postsynaptic neuron will arrive simultaneously...The resulting postsynaptic potentials produced by many synapses combine to create enough potential (voltage) for the postsynaptic neuron to generate an action potential and fire."
I think the same thing could be achieved with just an RC filter. If I'm following this correctly, the difference here is a "demonstrated variation in synaptic strength, a key neural mechanism associated with memory and learning." Things will really start to get interesting when something like this circuit can be made that is also capable of amplification. That would be a complete artificial neuron.
http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?arnumber=5754178
Deriving 3-d representations of the environment is mostly solved - what's needed is a way to recognize objects in the environment in an abstract fashion. There are techniques for recognizing, e.g., a specific box (one previously seen) in a complex scene, determining its 3-d position stereoscopically (no laser range finding needed), and interacting with the object robotically. What's needed is a way to classify unknown objects by abstract class, e.g., recognizing that there is a box in a scene, even though the specific box in the scene (say, something brightly decorated like a cereal box, or oddly shaped like some sort of product packaging) has never been seen before.
All we need is about a quadrillion of them for a brain.
Who wants to make brains out of these? I'm guessing not the researchers, since we can already make brains the natural way, and there's little advantage that I could see to making one synthetically.
From TFA:
"This is a necessary first step in the process," said Parker. "We wanted to answer the question: Can you build a circuit that would act like a neuron? The next step is even more complex. How can we build structures out of these circuits that mimic the neuron, and eventually the function of the brain, which has 100 billion neurons and 10,000 synapses?"
Uhhh... That number of synapses is off by about 10 orders of magnitude. I assume the number of synapses was meant to be a "per neuron" number, but that's a pretty glaring thing to leave out of that sentence. :-/
"Empathise with stupidity, and you're halfway to thinking like an idiot." - Iain M. Banks
You can even make an AI which appears self aware by giving it desires to do different tasks, but that is kinda wreckless in my opinion.
I prefer my AI's wreckless. If they create wrecks the autonomous vehicle projects should never go AI.
Well, I might have a way, but it only works on a semi spherical planet in a vacuum.
robots can do all sorts of tasks and even understand natural language as a programming language
Not on their own - we'll have to endow the neurons with prior knowledge about how human language works. (The circuits for human language is ingrained into the brain at birth by our DNA; it is only by using the assumptions about language that we are born with that we are able to learn it within 2-3 years. A clump of neurons without these assumptions would find it extremely difficult to learn human language, if not impossible.)
But, if we know enough about language to give robots the necessary prior knowledge, language is completely feasible. We might even be able to achieve similar results with an evolutionary algorithm, that mimics the way language evolved in humans - though that might also be difficult without prior knowledge.
Just because biological brains have synapses, do computer brains need them as well?
Serious question. I don't know where AI is or where it's taking us.
This is more useful to study human neurology.
Help! I'm a slashdot refugee.