By 2026, apparently:
"But because we’re doubling the rate of progress every decade, we’ll see a century of progress–at today’s rate–in only 25 calendar years."
"In line with my earlier predictions, supercomputers will achieve one human brain capacity by 2010, and personal computers will do so by around 2020. By 2030, it will take a village of human brains (around a thousand) to match $1000 of computing."
So, have super computers achieved one human brain's worth of computing? I believe we're only in the petaflop range, which is a factor of 10^10 behind his prediction.
At any rate, I believe AI is logically impossible.
I don't think that's a sensible claim at all. "Proof that humans are not artificial intelligences" doesn't even seem to be a coherent idea.
I'm going to go ahead and assume you've misunderstood what this person was talking about, as the alternative is to say that you're sitting on proof of the impossibility of a very quickly moving field with extremely deep implications for the future of humanity, and I've never once heard of it.
Estimates of the computing power of human brain vary widely, the ones that have some logic behind them range from 0.1 to 1000 petaflops. IMHO, they tend to be very "pro-brain", and should be seen as an upper limit.
Currently, the fastest supercomputer does abt 8 petaflops. This might be already 80 times more than the power of a human brain. And it could be less.. but where is your 10^10 factor from?
i would not say it is impossible. predicting the futuere is so difficult and things are changing so fast. i agree with the author "kurzweil" that we will first enter a hybrid phase between machines and humans. later, machines will surpass humans. the progress of tech evolution is getting faster and faster and i can't see right now what should stop this. i'm not a scientist, but what i can read in the article looks very logic and therefore i agree with kurzweil.
I have to disagree with Kurzweil here. I remember in one of the Kinect videos, the researcher who worked on the mic array explained that the billion neurons allowed such a level of processing that we could extract info from the noise easily.
The amount of computational resources needed to combat information overload on one sense organ is immense. To be able to pull that off with 5 organs, regulate body function, make inferences with ease, apply vast swaths of info to fields completely unrelated to the underlying information is just immense.
Although, I can see a budding brain peripherals market in about 2050+. It is just that we need a lot more than computational resources to emulate the brain.
One interesting quote I took away from a cognition class (the only one I considered legit in our Psych dept.) was that we tend to compare our brain to the most advanced tech of the moment. My gut feeling says that the current computational models we use are not adequate representations of the human mind and potentially a socratic approach also might fall short.
Just to pick on one piece of your argument : the difference between 1 sense organ and 5 is only a 5-fold increase. Even worst case it's like a 25-fold increase (network effect, etc). But a 25-fold increase should take less than a decade - as demonstrated by the power of the processor in your phone...
But I'll definitely agree that true AI is a difficult target to aim at, since AFAIK no-one knows what the important magical bit of intelligence is.
But unless you've got evidence that we've started to 'turn the other corner', doesn't your graph suggest that the level of technology is going to increase by the same factor as it did since the (say) 1950s?
And even then, maybe the AI level of technology is below the level to which we might be tending towards?
No, no it won't. This is the same statement that has been made consistently since the invention of computers. But the fact is, that despite our increases in computational power, and our high degree of success with specific tasks, we still don't even know what general "intelligence" really means, beyond "what humans do."
Until we have a reasonable definition for intelligence and consciousness (which I would argue are related), we'll always be moving the goalposts, and machine intelligence will always be a "few decades" away.
The following two statements are not incompossible:
1. At each time T, we believe that "machine intelligence", according to the definition in use at time T, is a few decades away;
2. At each time T, we're right.
In other words, the fact that in thirty years we'll still be talking about how we'll have machine intelligence one day, has nothing to do with the fact that by that point we'll already have machine intelligence as we currently conceive it.
The following two statements are not incompossible:
1. At each time T, we believe that "machine intelligence", according to the definition in use at time T, is a few decades away;
2. At each time T, we're right.
They're not incompatible, but the second one isn't true.
It's true that a few decades ago some folks thought that a computer capable of playing good chess would be true "machine intelligence". But they also thought such a computer would be able to carry on a sensible conversation and generally act like a person.
It turns out that playing a decent game of chess was easier than we thought, but building a HAL-like computer (man, I just realised that movie is nearly fifty years old!) is still incredibly difficult.
Still, I shouldn't even be talking about "building" an intelligent computer. The hardware is the easy part, the software is the hard part. Surely by now, all the supercomputers in the world are quite capable of simulating a rat's brain. But are they? Hell no, nobody has ever written a rat-brain simulation and nobody has the foggiest idea how to start.
Ouch! I got beat up over this. I still think it's funny. This man vs. machine appears on HN from time to time, and it probably will continue for another three decades. I've just gotta ask, are we really accomplishing anything here?
I feel like this is pretty non sequitur. He seems to imply that above some magic number of transistors intelligence will simply manifest itself. Intelligence is more than just computational power.
Indeed. I can believe that (maybe) sometime within the next few decades we'll have hardware theoretically capable of running a simulation of a human-level intelligence. I don't see any indication that we'll have the relevant software, nor any particularly good idea of how to write it.
He might be right, though. Intelligence might arise as an emergent property of some complex system. For example, automated trading.
There's no telling really. It also depends on how you define intelligence. It doesn't necessarily have to be (explicitly simulated) human intelligence.
(Mind that intelligence in humans wasn't created on demand either. It arose as an emergent property of the mammalian brain, due to evolutionary pressures)
It always seemed self-evident to me that to attain something like human intelligence, machines would have to share human human visual and auditory senses and be bootstrapped to a point where some great influx of knowledge could take place.
We now have that body of knowledge, audio, visual, and instantaneous worldwide communications.
I have to agree with Kurzweil that, surely machines that broadly exceed humans cannot be far away, though it may be their poetry will leave something to be desired.
When I look at history I don't see people that expected incorrect rates of growth, I see people that predicted incorrect directions of growth.
When nuclear power was new people predicted the future would be shaped by massive amounts of clean/cheap power. When air travel was new we predicted that everyone would fly everywhere.
The singularity is our version of the fusion powered car, or the personal jetpack.
Very good point, but computing devices don't have scaling problems that the other technologies you listed have. Nuclear would have had a similar generality if, for example, you could safely run a fission power device on your phone, for example -- but you can't; it just didn't scale down at the required safety level. Similarly, air travel doesn't scale down as people thought it would (still no flying cars) even though it works very well on large scale (airlines, combined, have near monopoly on long-distance travel). Computing scales both up (supercomputers) and down (cell phones). It may be that current silicon chips won't scale down to cellular level, but it also may be that (a) such scaling is not required for human-level AI (I believe it isn't), or that (b) other technologies such as graphene may replace silicon.
The assertion in the title is nonsense because "human intelligence" is not properly understood at this point. There is still much debate as to what drives the brain, and whether there is more to human intelligence than just the physical brain. Even if we could completely duplicate a human brain biologically, would it just start working automatically? Would it have feelings? Would it be able to learn if we could attach sensory inputs? This is not known. It's quite possible that there is much more to human intelligence than massive computing power.
> There is still much debate as to what drives the brain
There is no debate. The short answer is: sugar.
> whether there is more to human intelligence than just the physical brain
No.
> Even if we could completely duplicate a human brain biologically, would it just start working automatically
Not unless you teach it. Humans who grow up in the wild on their own are quite ape-like -- clearly, human brain doesn't start working automatically on its own (at least working well). Also, as demonstrated by machine learning, you can teach a non-biological system just as efficiently as you can teach a biological one, which means that the fist AI will reside (or already resides) on a run-of-the-mill silicon chip.
> Would it be able to learn if we could attach sensory inputs?
It would have to, otherwise it would not be human-level AI by definition.
> It's quite possible that there is much more to human intelligence than massive computing power.
Yes, and that is years of learning and good enough (just good enough) learning algorithms.
Since you seem to know this stuff, I had a couple of questions.
> Not unless you teach it. Humans who grow up in the wild on their own are quite ape-like -- clearly, human brain doesn't start working automatically on its own
I find this very interesting. Are there some articles/papers that I can look in to?
> also, as demonstrated by machine learning, you can teach a non-biological system just as efficiently as you can teach a biological one
Is there something you can refer me to that compares the efficiency of learning in humans vs machines?
> Are there some articles/papers that I can look in to?
There was little real science done because "real" science is when you do experiments, and a study of this kind would involve the so-called forbidden experiment, which would create a huge ethical stir. The most often-cited report is this one:
http://academic.research.microsoft.com/Publication/2053957/g...
While it is often said that it is hard to draw conclusions from the few reports that exist -- I think that the one conclusion that can be drawn is that nearly everything that makes us "human-like" is a product of learning and social interaction. No confirmed "feral" child has ever been reported to have acquired language in the full sense of the term.
> Is there something you can refer me to that compares the efficiency of learning in humans vs machines?
Just look up any introduction on machine learning. Nearly everything we do every day (recognizing objects, thinking in terms of goals, etc.) involves a classification problem, and every classification problem can be reduced to a regression on data. So there is nothing more fundamental to learning than fitting a model to data. Humans are pretty efficient learners, but they have their limits (your Bayesian spam filter is probably more efficient than you on recognizing spam, even though it may fail in a few cases because it doesn't quite "live" in the world you live in, meaning that its world model is not as deep/rich as yours). To acquire a deep/rich world model, you would need more training (provided you have a correctly balanced algorithm), which would ultimately be as expensive as raising a child. There is no reason though to think that it won't be done one day.
Don't see an argument here. One could also use your theorem to arrive to the exactly opposite conclusion: AI will happen and will eat your lunch, therefore there will be no free lunch.
Will strong AI ever enjoy a gin and tonic at the end of work on a summer's day? Will it ever feel the endorphin release from a strong run on the treadmill? Of course these are rhetorical questions. AI, however you define it, is a machine, a tool. It may become arbitrarily powerful, but that's all it will ever be; and it will always be a creation of man.
Your rhetorical questions of course have rhetorical answers: yes, if it is programmed that way. "AI" isn't a singular mind design. The rest of your comment is weird, I'll refrain from guessing what you're implying (if anything).
And when the AI designs a machine that is superior to it, and superior to anything that man has designed - then what? Or are you suggesting that humans are 'by definition' the only beings that can be said to have created something?
The AI (or its progeny) might disagree - and be able to argue pretty convincingly... After all, somewhere down the line they'll be much smarter than any human.
Or are you also arguing that human intelligence is some kind of asymptotic limit? Tell that to a kid that's smarter than its parents, or teachers.
Mitch Kapor and Ray Kurzweil made a public bet on this back in 2002. Mitch challenged that "By 2029 no computer - or 'machine intelligence' - will have passed the Turing Test." Stakes: $20k to the EFF or Kurzweil Foundation depending on the winner. See: http://longbets.org/1/
I understand the cynicism of many people here, and I totally agree that there is a lot more to intelligence than mere computational ability.
Then I type "adress of resurunt in adelide", it autocorrects to "address of restaurant in adelaide" and shows me a list of restaurants in my city.
I tell my phone to navigate to one, and it understands me and displays a map with route directions from my current location.
I wonder if anyone in my city has deployed SceneTap yet, and laugh to myself at how weak it is. I mean - it can only count the number of people in a bar and identify if they are male or female. I wonder why they aren't rating the attractiveness of the people there instead? [2]
Driving my car, I swear at the bad drivers ahead of me, and wish the cars here were driverless. It's crazy it has taken so long! Ok, I understand back in 2004 no cars managed to finish the DARPA challenge across the desert, and the winner in 2005 said "The impossible has been achieved" when his team's car finished[3] - but that was 6 years ago! Anyone would think this was complicated or something.
Then I come on HN, and see everyone bitching about how ignorant it is to think computers could ever be intelligent. I agree of course - it's obvious that chess is no test of intelligence, nor is spelling correction, nor is contextual instructions, nor is machine vision, nor is driving a car, nor is breaking CAPTCHAs[4]!!!
The Turing test - now that's a true test of intelligence. At least until it is passed - then obviously it's a flawed test.
Machine intelligence will soon diverge from Kurzweil's hyper-optimistic predictions and timetables so dramatically that he won't be able to gloss over it any longer.
As for where AI stands today, I'd say we're not even at the level of simulating a jellyfish, or even a single cell. Simulated protein folding is still a pipe-dream for our fastest supercomputers, nevermind simulating the intelligence of a sentient being.
On top of that, algorithmic systems as we understand them may be a shallow model of what it means to have 'human intelligence'. How do you design a circuit that feels pain, or know when you've succeeded?
39 comments
[ 1.6 ms ] story [ 94.0 ms ] thread"In line with my earlier predictions, supercomputers will achieve one human brain capacity by 2010, and personal computers will do so by around 2020. By 2030, it will take a village of human brains (around a thousand) to match $1000 of computing."
So, have super computers achieved one human brain's worth of computing? I believe we're only in the petaflop range, which is a factor of 10^10 behind his prediction.
At any rate, I believe AI is logically impossible.
I'm going to go ahead and assume you've misunderstood what this person was talking about, as the alternative is to say that you're sitting on proof of the impossibility of a very quickly moving field with extremely deep implications for the future of humanity, and I've never once heard of it.
Unless you'd like to explain it :)
Currently, the fastest supercomputer does abt 8 petaflops. This might be already 80 times more than the power of a human brain. And it could be less.. but where is your 10^10 factor from?
Sweet, I should start seeing human type sentience any day now!
The amount of computational resources needed to combat information overload on one sense organ is immense. To be able to pull that off with 5 organs, regulate body function, make inferences with ease, apply vast swaths of info to fields completely unrelated to the underlying information is just immense. Although, I can see a budding brain peripherals market in about 2050+. It is just that we need a lot more than computational resources to emulate the brain.
One interesting quote I took away from a cognition class (the only one I considered legit in our Psych dept.) was that we tend to compare our brain to the most advanced tech of the moment. My gut feeling says that the current computational models we use are not adequate representations of the human mind and potentially a socratic approach also might fall short.
But I'll definitely agree that true AI is a difficult target to aim at, since AFAIK no-one knows what the important magical bit of intelligence is.
"Machine intelligence will surpass human intelligence and a singularity will happen. Why? Technology gets more advanced."
Broken record...
http://dl.dropbox.com/u/8464115/graph_test.html
And even then, maybe the AI level of technology is below the level to which we might be tending towards?
Incientally, "a few decades" means "just after we get fusion power" right?
Until we have a reasonable definition for intelligence and consciousness (which I would argue are related), we'll always be moving the goalposts, and machine intelligence will always be a "few decades" away.
1. At each time T, we believe that "machine intelligence", according to the definition in use at time T, is a few decades away;
2. At each time T, we're right.
In other words, the fact that in thirty years we'll still be talking about how we'll have machine intelligence one day, has nothing to do with the fact that by that point we'll already have machine intelligence as we currently conceive it.
They're not incompatible, but the second one isn't true.
It's true that a few decades ago some folks thought that a computer capable of playing good chess would be true "machine intelligence". But they also thought such a computer would be able to carry on a sensible conversation and generally act like a person.
It turns out that playing a decent game of chess was easier than we thought, but building a HAL-like computer (man, I just realised that movie is nearly fifty years old!) is still incredibly difficult.
Still, I shouldn't even be talking about "building" an intelligent computer. The hardware is the easy part, the software is the hard part. Surely by now, all the supercomputers in the world are quite capable of simulating a rat's brain. But are they? Hell no, nobody has ever written a rat-brain simulation and nobody has the foggiest idea how to start.
There's no telling really. It also depends on how you define intelligence. It doesn't necessarily have to be (explicitly simulated) human intelligence.
(Mind that intelligence in humans wasn't created on demand either. It arose as an emergent property of the mammalian brain, due to evolutionary pressures)
We now have that body of knowledge, audio, visual, and instantaneous worldwide communications.
I have to agree with Kurzweil that, surely machines that broadly exceed humans cannot be far away, though it may be their poetry will leave something to be desired.
When nuclear power was new people predicted the future would be shaped by massive amounts of clean/cheap power. When air travel was new we predicted that everyone would fly everywhere.
The singularity is our version of the fusion powered car, or the personal jetpack.
See for example http://en.wikipedia.org/wiki/Mind-body_problem
There is no debate. The short answer is: sugar.
> whether there is more to human intelligence than just the physical brain
No.
> Even if we could completely duplicate a human brain biologically, would it just start working automatically
Not unless you teach it. Humans who grow up in the wild on their own are quite ape-like -- clearly, human brain doesn't start working automatically on its own (at least working well). Also, as demonstrated by machine learning, you can teach a non-biological system just as efficiently as you can teach a biological one, which means that the fist AI will reside (or already resides) on a run-of-the-mill silicon chip.
> Would it be able to learn if we could attach sensory inputs?
It would have to, otherwise it would not be human-level AI by definition.
> It's quite possible that there is much more to human intelligence than massive computing power.
Yes, and that is years of learning and good enough (just good enough) learning algorithms.
> See for example http://en.wikipedia.org/wiki/Mind-body_problem
My foot tells me it doesn't have that problem.
> Not unless you teach it. Humans who grow up in the wild on their own are quite ape-like -- clearly, human brain doesn't start working automatically on its own
I find this very interesting. Are there some articles/papers that I can look in to?
> also, as demonstrated by machine learning, you can teach a non-biological system just as efficiently as you can teach a biological one
Is there something you can refer me to that compares the efficiency of learning in humans vs machines?
There was little real science done because "real" science is when you do experiments, and a study of this kind would involve the so-called forbidden experiment, which would create a huge ethical stir. The most often-cited report is this one: http://academic.research.microsoft.com/Publication/2053957/g...
Here is a brief Powerpoint with an overview of interesting cases: http://www.tulane.edu/~howard/LING411/ppt/d39-FeralChildren....
While it is often said that it is hard to draw conclusions from the few reports that exist -- I think that the one conclusion that can be drawn is that nearly everything that makes us "human-like" is a product of learning and social interaction. No confirmed "feral" child has ever been reported to have acquired language in the full sense of the term.
> Is there something you can refer me to that compares the efficiency of learning in humans vs machines?
Just look up any introduction on machine learning. Nearly everything we do every day (recognizing objects, thinking in terms of goals, etc.) involves a classification problem, and every classification problem can be reduced to a regression on data. So there is nothing more fundamental to learning than fitting a model to data. Humans are pretty efficient learners, but they have their limits (your Bayesian spam filter is probably more efficient than you on recognizing spam, even though it may fail in a few cases because it doesn't quite "live" in the world you live in, meaning that its world model is not as deep/rich as yours). To acquire a deep/rich world model, you would need more training (provided you have a correctly balanced algorithm), which would ultimately be as expensive as raising a child. There is no reason though to think that it won't be done one day.
The AI (or its progeny) might disagree - and be able to argue pretty convincingly... After all, somewhere down the line they'll be much smarter than any human.
Or are you also arguing that human intelligence is some kind of asymptotic limit? Tell that to a kid that's smarter than its parents, or teachers.
Then I type "adress of resurunt in adelide", it autocorrects to "address of restaurant in adelaide" and shows me a list of restaurants in my city.
I tell my phone to navigate to one, and it understands me and displays a map with route directions from my current location.
I wonder if anyone in my city has deployed SceneTap yet, and laugh to myself at how weak it is. I mean - it can only count the number of people in a bar and identify if they are male or female. I wonder why they aren't rating the attractiveness of the people there instead? [2]
Driving my car, I swear at the bad drivers ahead of me, and wish the cars here were driverless. It's crazy it has taken so long! Ok, I understand back in 2004 no cars managed to finish the DARPA challenge across the desert, and the winner in 2005 said "The impossible has been achieved" when his team's car finished[3] - but that was 6 years ago! Anyone would think this was complicated or something.
Then I come on HN, and see everyone bitching about how ignorant it is to think computers could ever be intelligent. I agree of course - it's obvious that chess is no test of intelligence, nor is spelling correction, nor is contextual instructions, nor is machine vision, nor is driving a car, nor is breaking CAPTCHAs[4]!!!
The Turing test - now that's a true test of intelligence. At least until it is passed - then obviously it's a flawed test.
[1] http://blogs.forbes.com/kashmirhill/2011/06/28/using-facial-...
[2] http://www.springerlink.com/content/t75552811t449746/
[3] http://www.foxnews.com/story/0,2933,171673,00.html
[4] http://news.ycombinator.com/item?id=1897932
As for where AI stands today, I'd say we're not even at the level of simulating a jellyfish, or even a single cell. Simulated protein folding is still a pipe-dream for our fastest supercomputers, nevermind simulating the intelligence of a sentient being.
On top of that, algorithmic systems as we understand them may be a shallow model of what it means to have 'human intelligence'. How do you design a circuit that feels pain, or know when you've succeeded?