Are memristors the future of Artifical Intelligence? DARPA thinks so
By Joseph L. Flatley 
posted Jul 14th 2009 5:54AM
posted Jul 14th 2009 5:54AM
New Scientist has recently published an article that discusses the memristor, the long theorized basic circuit element that can generate voltage from a current (like a resistor), but in a more complex, dynamic manner -- with the ability to "remember" previous currents. As we've seen, HP has already made progress developing hybrid memristor-transistor chips, but now the hubbub is the technology's applications for artificial intelligence. Apparently, synapses have complex electrical responses "maddeningly similar" to those of memristors, a realization that led Leon Chua (who first discovered the memristor in 1971) to say that synapses are memristors, "the missing circuit element I was looking for" was with us all along, it seems. And of course, it didn't take long for DARPA to jump into the fray, with our fave DoD outfit recently announcing its Systems of Neuromorphic Adaptive Plastic Scalable Electronics Program (SyNAPSE -- cute, huh?) with the goal of developing "biological neural systems" that can "autonomously process information in complex environments by automatically learning relevant and probabilistically stable features and associations." In other words, they see this as a way to make their killer robots a helluva lot smarter -- and you know what that means, don't you?
Read - New Scientist: "Memristor minds: The future of artificial intelligence"
Read - DARPA: "Systems of Neuromorphic Adaptive Plastic Scalable Electronics"
Read - New Scientist: "Memristor minds: The future of artificial intelligence"
Read - DARPA: "Systems of Neuromorphic Adaptive Plastic Scalable Electronics"
Has pandoras box been opened?
AI that never forgets? Ha, girlfriends are scarier than that.
I knew a girl at school called Pandora. Never got to see her box, though.
I've seen a stripper named Pandora. Got to see her box when she was on stage.
But these Pandoras you speak of-
were their boxes OPEN?
Its ok if Pandora's box was opened. Just so long as she can close it.
DARPA is actually spending most of its resources on getting the acronyms to fit. All the rest is just "science".
Defense Agency Responsible for Providing Acronyms.
@switchbitch: +1
Aww... the military and their acronyms... (gonna get us all doomed eventually)
Can someone just explain this a bit better because the article has confused me. From what I can tell, these things are kinda useless, as they adapt to the present current flowing through them. So if the current changes, they change. Which means that remembering what the previous current was is irrelevant, as the next current to pass through it will just overwrite that. I can't be right on this, so I'd like to know what I'm missing here...
Well, you have some options.
(1) Physics II (Electricity and Magnetism); it's a 1 semester course, credited no different than biology. Lab included.
OR
(2) http://en.wikipedia.org/wiki/Memristor#Memristor_theory
Wow, how tremendously helpful. Either take a physics course or try to decipher a page that you'd probably need at least an A-level in physics to understand. Ta muchly...
You can still read a memristor without changing its resistance by measuring the the change in voltage across the memristor using a small alternating current.
Think of a memristor like a box. You have to open the box to put stuff in the box and take stuff out, but you only have to shake it to know how much stuff is inside.
This enables processors to save a wider array of values, forget 1's and 0's, it's more human like now (or will be).
Couldn't we do that already? While the electronics we have are designed for digital processes, they are capable of functioning in an analogue manor. Capacitors can store information (via charge) to various quantities. However, they're rather crap at it as the signal degrades, and dealing with capacitors require additional components (transistors). And refreshing a degrading signal can only be done digitally to any degree of reliability. And refreshing takes constant energy. And they're much slower. And if you want to make a circuit who's response changes depending on its previous response takes more components still. So all in all, as far as I can see, we've replicated the functions of memsistors for a very long time, but they're all a bit crap by comparison. Why DARPA suddenly thinks this is the key to replicating the brain I have no idea. At best I'd think it means that they can use less components to replicate the same functions, but the cost is still offset by the fact they have no clue as yet how to mass manufacture memsistors. By the way, hybrid analogue/digital processing may be a new efficient system for lossy computations such as video compression.
Either way, I thought the brain's power came from it's three dimensional wetware architecture enabling incredible component densities that can take advantage of incalculable constructive parallelism. Basically the connections are in 3d, there's 100 trillion components, the circuits can rewire themselves, and almost every part can communicate with almost any other part almost immediately. Compare this to a computer chip, where the arrangement is flat, there's a billion components, the hardware's fixed and parallel pipelines are almost completely cut off from each other. I see how a component that remembers might better replicate a synapse that remembers, but the rest is still a ways off.
I for one... meh, who am I kidding!
I am actually wondering who the hell comes up with these acronyms? Seriously, some of them are billiant! It's like the British Army and it's love of TLA's (three letter abreviations).
hmmm intel's font
While billions of dollars are spent for machines that will eventually go insane and start spraying people with lasers and napalm- using children as fuel- I hope that there is some underground resistance that is making advances with electromagnetic pulse weapons, or else we're all screwed.
I mean that seriously.
There better be some fatass geek muthafucka in his mom's basement, who watches too many sci-fi movies and truly believed all this intelligent-power-hungry-robot shit before Terminator 2 et al, who is working on some crazy anti-intelligent-robot weapons. Because if there's not some eccentric reject making some seemingly useless weapons NOW, then by the time old ladies' heads start lopping off from knee mounted razor blade cannons, it's too late. It's not like the movies where one can just go into some basement and MacGyver some "this just might work..." weapon. We need that one EMP dude who also has made communication devices that are based on encoding/decoding messages by means of fluctuating the resonant frequency of plants through song.
If you're out there reading this, saviour of The People..... Good luck, and godspeed. And also start working out and washing your face, cuz nobody's ready for an obese and acne covered hero.
"Systems of Neuromorphic Adaptive Plastic Scalable Electronics Program"
...or simply, SKYNET.
.
@peepeepants
I predict that the one that will eventually save us all is none other than Benjamin J Heckendorn.
I don't mean to spoil the fun but as far as I'm concerned robots don't do anything unless they are told to do something. Maybe the people you should be worrying about are the ones who will be controlling the robots that are doing the killing. It will be a human being on the other side I can guarantee that. As much as I dig movies like Terminator or irobot there is no evidence suggesting robots will ever kill humans unless we program them to do so. I think the real problem is within ourselves. Lets stop killing each other mmkay?
Yeah, computers never do things you didn't tell them to. Oh wait.
You're right about how it will function and the method used will be determining the time it takes to change to the resistance from the new current, thereby allowing you to calculate what the previous resistance was.
Clearly meant as a reply to dboobis. Let's see if this one actually replies to myself or not...
Ah now I see. Thank you for actually being informative.
I'm not too worried about the robot apocalypse anytime soon. To become more and more advanced, these robots are all using tricks to just mimic the human brain. Eventually, the latest and greatest super-bots will be just as susceptible to errors and confusion and humans. And then we will finally realize... Robots feel love too.
OK I get the AI thing, scientists are pretty sure our brains work in ternary (-1,0,1) rather then binary like a computer (0,1)
but this is just a ternary computing platform? uh...ok
didn't the russians have one of these in the 60s? yes..they did, the Setun and the Setun-70
Whats the difference between this and Setun-70? I would really like to know how this will path the way to AI and more then another Ternary computer. (I'm not being cynical for once, I'm actually curious)
Another step closer to true A.I. not the behaviour gamers claim to be A.I.
I wanted to write some usefull apps here but there are way too many things we can do with A.I.
Main question is "How do we balance the A.I's freedom and still keep it from going haywire?
It's going to take a whole load a smart people to program these things, maybe even computers programming computers.
"We have only bits and pieces of information but what we know for certain is that at some point in the early twenty-first century all of mankind was united in celebration. We marveled at our own magnificence as we gave birth to AI."
Cue thunder clap.
Re: peepeepants
Sorry to say that EMP's are not going to save you from robots.
Ever hear of Rockwell Collins?
Besides, if your tax dollars paid for a killing machine shouldn't you be the one to reap the benefit of being killed by it?
If they could be used for synapses then it could possibly be used for regular nerves. That combined with the metallic muscles they mentioned a few months back could be used to make fully functional prosthetic limbs.
Do you mean Artificial? not artifical.
The resistor that does not forgive, and does not forget.
darPa initiative ^^
Not one mention of "The Pipes" yet? I'm dissapointed.
There was a fascinating article in ieee spectrum about memristors:
http://www.spectrum.ieee.org/semiconductors/processors/how-we-found-the-missing-memristor/
it's exciting to think of the possibilities this would unlock...
1. wall thickness of enlarged tube should be thinner than one not enlarged, if i am not mistaken.
2. we are already capable of incredible amounts of ARTIFICIAL INTELLIGENCE, or has nobody been paying attention?
tomo