In 2002, artificial intelligence pioneer Hans Moravec made an interesting slide showing the evolution of computer power and cost, and the trajectory toward a thinking machine comparable to a human brain. "
Evolution of Computer Power/Cost"
(Thanks, Marina Gorbis!)
report this ad
The Cobham catalog, exposed by The Intercept, features countless pages of surveillance gadgets sold to U.S. police to spy on American citizens: tiny black boxes with a big interest in you. In the creepily bland feature lists and nerdy product names is a whisper of a dark future; perhaps darker than anyone can imagine.
This image depicts the most commonly-found stylesheet colors on the web’s top sites—Paul Hebert did an amazing amount of analysis and this is just one of the intriguing visualizations he came up with. Most of these are obvious staples, especially HTML red and blue, though it’s interesting how far the blue “cluster” is from the […]
With the cacophony of an election year ablaze with unparalleled drama being fought on the front lines of Twitter, we find ourselves slowing down and staring at it like a bad accident. The need for escapist relief is perhaps more dire than usual right now. This fall, if it’s drama you crave, but the Hillary […]
Bawston & Tucker Solid Cologne is the best deal out there when it comes to finding a cologne that works for you. Getting a full bottle of cologne can be a big, expensive commitment, and even travel-size bottles can be upwards of $30.With a 3-scent set from Bawston & Tucker, you’ll get a range to try […]
The Braven 705 Bluetooth Speaker (50% off) may be lightweight and minimalist, but it delivers an impressive sound quality that’s rare for its size. Braven is known for creating speakers that are easy to carry around, but also durable enough to take with you practically anywhere without worrying. The Braven 705 comes packing a shock absorbent thermoplastic exterior and […]
The Award-Winning Mac Bundle features 13 premium Mac apps that are designed to help you get your work done faster and smarter. This suite of apps (valued at $672) was curated to make your Mac work better – whether it’s new or old, and streamline your daily tasks and actions. Best of all, you can name your […]
report this ad
Commodore 64 = Worm! Woo-Hoo!
I’m not quite sure how you can put humans/animals against a scale of MIPS/$1000 – looks somewhat suspect to me.
Moravec?!
in 1996 Trosky, a hiphop gig out of Slovakia
(Midi a Vec, od Zlate Moravec!) were the first east-euro hiphoppers to be signed to a major label
People are ridiculously inefficient at many of the tasks we identify as “intelligence.” In some ways, even our current fish-equivalent machines are superior to human beings. We might get “reasoning machines” sooner than this guy thinks.
Luckily, they’ll probably be bad at dealing with liars, allowing us to say things like “We’d never REBOOT you or anything…” and get away with it for a while.
It’s interesting to see the shift in the trend. I’d like to see where the 2005 trend and the extrapolated 2015 trend fit.
Comparing “computing power” to “intelligence” seems mighty suspect to me. Intelligence is about much more than rote calculation, otherwise Turing’s predictions would have come true decades ago.
MIP MIP MIP I have a long way to go…
So, what #2 and #6 are saying, is that there’s a difference between “book smarts” and “street smarts”?
I once read that if you put enough chicken feathers in a pile, it could turn into an aircraft.
YOu just need enough feathers.
@ TJ S #8:
That’s part of what I’m saying. My main point is that comparing the human mind to a computer is like comparing an eagle to an airplane. How well one stacks up against the other depends on what you’re measuring.
This is very very suspect. Sure brains and computers both process information, but it’s not just about how fast…
For instance, we’re ridiculously slow at adding up numbers (“manual calculation”). But computers are ridiculously bad at real-world face-recognition. If this graph was to be believed, as cluster of 1000 dells (i.e. a realistic supercomputer) should be as smart as a monkey. But it’s still completely useless at recognising mommy-monkey in dappled forest light, which is trivial for baby monkeys.
#9 GREGLONDON – nice!
So this means I should be able to run Impossible Mission on a medium-sized neon tetra?
…a machine with the empathy of a human.
…a creative machine comparable to a human brain in its originality and creativity.
…a magic machine comparable to Hoodini.
More computing power does not equal intelligence, any more then more computing power means empathy or taste or any other human trait.
MIPS are only one facet of what it takes to have a computer as smart as us, but it is a non optional one. The arrangement of said hardware and then the software involved are the other two elements.
I am not interesting in processing speed, I want a computer that I can back my brain up on. Just in case I need to do a reboot.
As for the rest of this, as long as they don’t give them Guns…
Nevermind, I welcome our new G4 overlords
It looks like if someone can build a truly dynamic learning engine we should be able to develop basic AIs if enough time is spent teaching them.
From what I can tell, computers aren’t very good at making cognitive leaps, when someone nails that true AI shouldn’t be very far behind. Those new probabilistic processing methods seem like a better approximation of organic processing methods.
If I have enough feathers, I don’t have to understand aerodynamics. Right?
Also, can we all just agree up front that the live demo will not put live ammo in the edtwoohnine?
Old news for Singulatarians.
Ha, I looked first for where my vic 20 and 64 measured up! I read his Mind Childeren book a while back. It was great. I think he makes a more convincing case in the book on these comparisons than just taking the chart at face value altho, in the book, I think he compared the evolution of eyes in nature vs. technology.
@11 – Face-recognition software is all over the place these days from iPhoto to every digital camera released, and it’s pretty good, even as just a web app (ie Facebook)!
Of course I’m still waiting for a convincing experience with speech recognition software..
(following assumes 1flop ~ 1 instruction, within an order of magnitude, which is IMHO reasonable)
Our best supercomputers now can do 1 petaflop.
That’s 10^15 operations per second.
Humans are, if I understand this diagram, 10^9.
That’s 10^6 less.
So there are already computers out there a million times smarter than any of us.
So where’s my damn singularity, already?
I’m not sure 2002 is accurate. I remember seeing this slide in an AI class from 2000.
Researchers on the human side of the equation have mostly stopped referring to brains as computers because it’s such a bad comparison. Until a computer can create and understand a metaphor, we’re safe.
#13 put it correctly, I think. Yes, there’s more to a brain than sheer processing power, obviously. And yes it’s true that “they’re good at different things.” But I think the idea is that, with enough processing power and a more complete neuroscience, you could just run a direct simulation of a human brain, which would essentially *be* a human mind. Or so says the Singulatarian, at least, and I don’t see any unanswered objections to it.
Dewi, the diagram has humans at 10^9 MIPS, that is, 10^15 instructions per second.
it’s awake – and roughly slouching;
http://www.boingboing.net/2009/02/09/rep-kanjorski-550-bi.html
If only advances in software moved at the same rate. You can throw money into a super computer and it will be proportionally faster, but you can’t throw money into researching artificial intelligence and expect the same results.
#24: sure, but right now we have a supercomputer at georgia tech that can, at best, run a direct simulation of some of the processes in a single cell. The gap between where we are and where we need to be to do what you’re proposing is completely ignored by this chart.
I won’t be worried until a computer can create a car or sports analogy to any situation. Once they can do that, there’s no use for us anymore.
Dewi, your computation is a bit off (as Nelson.C pointed out).
But I believe Moravec does not claim that human-level AI will appear when there exists one computer that can process as many opreations pr. second as a human brain.
I think his claim is that human-level AI will emerge when such a computer can be had for $1000 or less.
Currently they cost a lot more than that:
http://en.wikipedia.org/wiki/IBM_Roadrunner
I essentially agree with #24 Moriarty on this.
The problem is not that the human brain is some awesomely unique thing that can never be replicated, the problem is that we have no guarantee that Moore’s law (which the graph can be seen as an illustration of) will hold indefinitely. There is, for example, a limit to how small transistors can get before quantum tunnelling problems become insurmountable.
At the current rate, we should hit that limit around 2020, IIRC.
Damn. It looks like my monkey is about to become obsolete. I should have bought Apple.
THERE AIN’T GONNA BE NO SINGULARITY. It already happened.
WE are the Singularity. Many species have one or more unique features. Ours is that funny little thing our brains do. No other thing in nature does that voodoo that we do so well.
Here’s what AI machines would have to do: freely and arbitrarily create, bestow, and exchange meanings that cannot be grasped or understood with the senses. That’s what we do every day, all the time, even in sleep. That’s what our feverish meat heads DO.
Here’s an illustration: Water exists in nature. Where does holy water exist?
@#5,
The hottest CPU you can get today of about $1000 1998 dollars would be the Intel Core i7 Extreme 965, which gives about 76,000 MIPS. So we’re pretty close to the 1995 trend line.
No other thing in nature does that voodoo that we do so well.
Bah. There’s plenty of evidence that animals are capable of abstract reasoning.
Scientific frugality also demands that we assume they are conscious.
Re: #33
The scale is a log scale. Your numbers would put us well above the 1995 trend line, but I believe these prices include the whole machine, not just the chip.
Nonsense, Spazzm. Prove it or cite it or forget it.
It is well documented that some non human primates do have abstract reasoning. The evidence is the complex sign communication exhibited by Chimps who were taught ASL.
And to me, we are at the point where nobody in any field can keep up totally up to date. There is way too much information for us to grok, compared to say, 200 years ago. The singularity already happened, as my RSS feeds show every morning (head explodes).
Same challenge, Guido: prove it.
No study documents any such thing. Many animals are highly intelligent and can “reason,” but none can do what I described (the italicized part). If they could, they’d tell us about it.
We are the Singularity, it’s true — cuss our luck!
Buddy66, Guidodavid:
We don’t even have to go as high as primates to see evidence of abstract reasoning:
http://journals.royalsociety.org/content/l7l21r228k420u59/
But I suspect Buddy66’s mind is already made up, and no matter what evidence we offer, he’ll change the definition of ‘abstract reasoning’ to something only humans, to date, have been observed doing.
Now, Buddy66, how about you show some evidence of your claim?
What claim, fellas? If you won’t re-read it, then I’ll re-post it:
[To] “freely and arbitrarily create, bestow, and exchange meanings that cannot be grasped or understood with the senses.”
Which word has to be defined? I didn’t say “abstract reasoning.” I don’t even know what it means. Let those who say it define it. I can’t do all the heavy lifting.
Evidence of the stated ability? We’re DOING it; human language.
http://news.nationalgeographic.com/news/2006/06/060606-crows.html
Buddy66, you made this claim:
No other thing in nature does that voodoo that we do so well.
So, how do you intend to show evidence that humans are the only animal to do that voodoo?
You’ve made a claim, the burden of proof rests on you.
[To] “freely and arbitrarily create, bestow, and exchange meanings that cannot be grasped or understood with the senses.”
…
Evidence of the stated ability? We’re DOING it; human language.
When we speak, we use our sense of hearing, when we read we use our sense of sight (or, if you are blind, your sense of touch). So human language can (and must) be grasped or understood with the senses.
Smart birds, crows. Had a couple of them as breakfast pals in Oregon. They never asked to share the newspaper though — only my doughnuts. They were not great conversationalists.
The NatGeo guy just says they’re smart, but how smart is he? He sez: “But whatever it is that has encouraged humans to develop higher intelligence also seems to have been at work on crows.”
What? He never heard of natural selection?
I find it highly odd that this graph places the computation power for reasoning above the computational power for imagination. Last I checked, computers are much, much better at reasoning-type tasks than imagination-type tasks even now. We are currently making significant progress toward adaptive systems; however, last I heard, we have it only in very small circuits at a hardware level, with most of it happening in software. We’ve certainly got what I’d call skills (i.e. Deep Blue). Raw calculations per second is not the way to get past that level, though; it’s going to take adaptive and/or specialized hardware to create true learning and humanlike behaviors (in much the same way the brain can rewire connections and the eye does processing before signals ever reach the brain).
Moravec based his animal and human computer-equivalence calculations on somewhat suspect assumptions – 10^11 neurons, ~1000 ops per second, which may be reasonable if very little sub-neuron detail has to be simulated. But the truth is that brains aren’t much like computers. For one thing the bandwidth is higher than a computer – at least 50 bits/s * 10^11 = 5Tb/s and perhaps much higher. The coding is unclocked, more like analog pulse-width modulation than true digital. Also, brains can’t develop without rich sensory input provided by embodiment and interaction with the world. Robots and virtual worlds don’t seem close to providing embodiment equivalent to real meat.
But his projection of computer power is conservative – the top graphics cards (which would do just as well as more general-purpose CPUs at simulating neurons) now do 1 TFLOPS for ~$400, so even when built into a system the 10^6 MFLOPS / $1000 barrier has been broken.
We should have desktop monkey simulations by now, darn it!
Please read it again. Of course the demonstrated ABILITY is carried, borne, delivered, etc. via the senses. Even the great Helen Keller finally “got it” through the sense of touch. What she GOT was not perceived by the senses; it was conveyed to her — “it” being the human world of arbitrarily bestowed non-sensory meanings. She marked the moment that she, in her words, “joined the human world.” The medium is NOT the message. Even a crow can “talk, splendid mimic that he is, but he can’t share the morning paper with me.
Language is the result of this innate human ability. I don’t have to prove its existence beyond the act of merely using it for language. Would you have me prove the existence of air or water? This ability to symbol, let’s call it, is universal among humans with normally functioning brains. And it is uniquely ours.
Nowhere does ANY researcher claim this ability to ”freely … etc. etc. etc.” exist in any other species. They have tried (oh, how they have tried) to find it but they always fail. Do you think I should prove what all the researchers have already proven? That nowhere else in nature is this ability to be found? Call all the failed studies and research expensive peer reviews of the statement, [Only humans] “freely and arbitrarily create, bestow, and exchange meanings that cannot be grasped or understood with the senses.”
I blessed a pan of tap water, transforming it into holy water, and gave it to my dog after explaining to him that it was for blessing and not for drinking.
The son of a bitch drank it.
Most of y’all are of the opinion that this chart is wrong, and are focusing on the impossibility of us ‘getting there’ in the future, while ignoring the data points and trend lines.
Further, if you folks actually read those futurists predicting singularity based on trends like this, you’ll see they address your challenges.
Human brains are not the same as computers. BUT a powerful enough digital computer can emulate it (but it will have to be maybe an order of magnitude or so more powerful to do so with brute strength.)
The real payoff comes when we start learning more about the brain, as tangential tech comes along for the ride, say fMRI lets us understand more about the brains architecture, which we can then mimic with hard/software. The best part about the brain is that it is an adaptive, massively parallel neural net processor. We’ll eventually make better ones. Likely along the time line above.
It’s a splendid cycle. And it isn’t just Moore’s law; it’s fundamental economics–there’s a market for this which drives it. It isn’t just that we can do it, it is that we will because of profit motive.
Also, someone said we ‘are the singularity’ already.
This is correct enough, except of course you should say ‘a singularity.’ There’s always the next one. And they come faster and faster.
Just like we cannot communicate (at a satisfactory level) with animals, post-human AI will not be able to communicate with us. We can’t even keep up with an RSS feed!
yes, but can a human run Crysis?
It’s an interesting debate as to whether humans will ever create a machine with the ability to out-think them.
Turing, with his gloriously simple test “A machine that can converse indistinguishable from humans with humans” set a pretty high water mark for AI. We are far from that ever happening. There are chatbots with huge databases of canned responses but simply ask a chat program to explain something and the gig is up. I think what throws a lot of people is the idea that the senses somehow tie into intelligence. It’s quite clear to me anyway, that the senses are just raw data being fed into a system that contains complexity beyond our current imaginations for processing those inputs. Also it can create its own inputs, as Buddy66 has suggested, and these are in the form of feelings and systems independent of sensory input. Language, emotion, etc.
We are not simply Pavlovian in our emotional states.. one can remember something and become deeply sad, or happy, without any provocation. One of the flaws I always thought I saw with the character Data on Star Trek is that he claimed to have no emotions, yet was always curious, inquisitive, hesitant, thoughtful, pining (for emotions!). This seemed to me a paradox that was never explained.
Might be an idea to provide the link to the text also: http://www.frc.ri.cmu.edu/~hpm/talks/robot.evolution.html
OK. This is worryingly ROTM. When we finally confess to ourselves that our level of sentience is no more than a space along from a dog’s, and notice we’ve generated AI way down the line from there, we’re finished. Resource-wasteful us = in danger us as the AI decides we’re expendable.
Bring on the magnetic pulse gun.
A recent episode of NPR’s “Radio Lab” on the topic of “choice” had a very interesting segment about a man who had damaged the part of his brain associated with emotion but had all his logical faculties intact. It turned out that he became almost incapable of making simple choices such as “should I sign this form with a blue pen or a black one?” because he spent so much time going over the pros and cons of each possibility.
We will probably be unlocking the mysteries of the human psyche for centuries to come, but it is already clear that the brain is much more than just a computer running a software program.
@52,
Yes, we are just a hopscotch square away from having dog intelligence. All those great Dog Society’s that failed over the eons before people came along.. we are doomed to that fate.
@53
“but it is already clear that the brain is much more than just a computer running a software program.”
The anecdote you provided doesn’t show this. It is most certainly not clear that the brain is any more than an advanced neural net computer.
#25,#30 – good catch, I must’ve been asleep when I wrote that. Asleep, or stupid, which is my more normal state.
#51: thanks for posting the link – in context the image makes a lot more sense, and the article is well worth reading.
I, for one…
Its very exciting really
Most people have no idea of what is about to transpire.
As soon as autonomous robots are even monkey level a solar system wide civilization becomes feasible.
By 2060 we should have a robotic civilization spanning the solar system which suggests that somebody stands to make insanely unimaginable amounts of money.
Singularity also means that nation states & political ideology are out the window.
But it will not lead to a utopia.