My shorthand for the singularity is that it's whenever AI becomes as relevant as we thought it was going to be. It is really a brilliant rhetorical device to discuss technology.
Similarly though, you could say that pneumatic tubes are still going to be a major technology, but first we have to reach the "tubularity." When is football going to be as massively popular in the US as the rest of the world? Well, after we reach The Soccularity. Now that you presume it's a future event, you can bracket the conversation around that.
Note that I'm not doubting the real singularity's existence, just marveling at this act of rhetorical jujitsu.
The definition in his first paragraph is unrelated to the criticism in the second, which is that using the phrase "The Singularity" already frames things in favor of the idea that it is a real future event.
Personally, the concept has always struck me as the childish fantasy of a certain personality type. But I think it's worth pointing out that in another current article on HN (http://news.ycombinator.com/item?id=205869), Freeman Dyson has some interesting things to say about it:
Q: Does any of that [concept of "The Singularity"] resonate with you?
Dyson: Not at all. The technical tricks these people are talking about are only a small part of the human experience. They vastly overestimate their own importance. I look at the world in a very different way. It's partly a matter of being old, but I look at the subway networks in cities, for instance. They also have the N-squared law. If you have a subway network with N routes, its value to the passenger is N squared. That's fine. But once you get to a certain number of routes, like 20 or so, there's very rapid growth, followed by saturation. This will also happen with chips. To some extent, it already has. It's true that the price per megaflop is going down according to Moore's Law, but what you can do with the processing power isn't increasing at the same rate.
I remember doing a study on the cost of nuclear power in the 1950s, when people thought it would be very cheap. We studied what the economic effect would be if the cost of electricity were zero. The answer is, "Not much." It costs far more to use electricity than it does to make it. There's about a 5 percent drop in the GNP if electricity is free. So cheap energy is all it takes. The same is true of computing power.
"Personally, the concept has always struck me as the childish fantasy of a certain personality type."
Do you have some factual basis for considering the claims to be "childish fantasy"? Or is the fact that it conforms to your stereotypes is enough to dismiss it without investigating the claims?
Well its certainly fantasy. What factual basis do you need for categorising speculation?
The problem with singularity is its unchecked use of extrapolation. With any kind of prediction you ought to factor in a certain level of uncertainty based on how much you've observed about a phenomena. Considering we don't have yet what we'd call real artificial intelligence, let alone artificial intelligences creating other artificial intelligences - our level of uncertainty renders any prediction absurd.
AI is, by definition, the set of problems that are not yet solvable by computers. Ergo, all AI is fantasy. Nevermind that AI has beaten the world chess champ, is used every day for pattern recognition, stock trading, and a myriad other uses.
Those who do not prepare for the coming Singularity are going to be blindsided by it when it arrives.
AI is, by definition, the set of problems that are not yet solvable by computers.
No it's not. At least, that's not how leading textbooks define it and it's not how McCarthy defined it. Moreover you promptly contradict it: beating the chess champ is hardly "not yet solvable by computers". Did it cease to be AI the moment it succeeded?
Those who do not prepare for the coming Singularity are going to be blindsided by it when it arrives.
This seems to me a religious sentiment expressed in religious language. Prepare ye the way of the Lord!
I think it would be more accurate to say that "AI is, in practice, the set of problems that are not yet solvable by computers". It's pretty cynical to call that the definition of AI, but I don't think that's an inaccurate characterization if you look at the domain of problems that people who consider themselves "AI researchers" has shifted over the years.
There's a sound argument that goes from "AI is feasible" to "AI can be recursively self-improving". So it's not just a random conclusion as you imply.
I'm not going to go into detail explaining it here because I'd rather spend my energy on doing that in a way that can communicate to more people. (I'm working on that.)
"More interestingly, what's the response to Dyson's claim that growth in processing power is not matched by growth in people's capacity to use it?"
Dyson seems to be referring to the dumb "extrapolated exponential growth = 'Singularity'" hypothesis. I agree that that idea is completely stupid, I don't see how they can conclude anything from that extrapolation. My argument rests on two postulates: 1. AI is feasible. 2. When AI arrives on the scene, there will be large changes.
There are still limits on what an infinitely intelligent creature can do in the real world. AI is not going to result in magic levels of technology. For example you might build an IC engine that is 40, 50, or 60+% efficient but it's never going to be 100.00001%. And as we can already build 40+% efficient power plants so it's not possible to build something that extracts 3x as much energy from burning fuel.
I fully agree, and I don't see why you think I'm implying that there will be "magic" levels of technology, or that something can be "infinitely intelligent".
All I'm saying is that there's still a lot of room for improvement, and AI will be a hugely beneficial tool toward that end.
I was more talking about the "Event Horizon" from your link. You can look at the limits on what can be designed without the ability to design it yourself. We already use computers to look though each variation in boat hulls to find the "best" design. Adding 2x more processing power is useful but there are rapid diminishing returns. In the same way "Intelligence Explosion" assumes x IQ would let you build an X * n level of IQ but it's more reasonable to expect that at some point x IQ let's you build an x + (n / x) IQ.
In the US we talk about what level of battery technology will finally let EV's take off, but adding an electric rail to all roads is the limit on how useful battery technology would ever get. AKA you endup removing the limit how much energy you can move from the power plant to car but that does not help you generate energy.
Edit: I still think AI would be useful but I don't think it's going to change the world as much as people might think. It's like the internet it's new and useful but it's not all that much better than POTS + library + a good local collage.
That blog post I referenced is making the point that these three schools of thought are mutually incompatible. I agree that the "Event Horizon" idea is not making any interesting claims, although it's a good intuition pump. I think the "Intelligence Explosion" idea is worth investigating further.
Also, none of the arguments in that blog post are arguing that there will be unbounded/infinite growth.
> But once you get to a certain number of routes, like 20 or so, there's very rapid growth, followed by saturation. This will also happen with chips. To some extent, it already has.
The human brain has billions of routes, and uses them somewhat efficiently...
That's not the point.
It's the massive amount of neurons and (finely tuned) synapses that allows human intelligence, and the vast amount of possible mental "states".
Functional organization is of course of prime importance, but AFAIK, it's the sheer cortical surfaces that allow us to be smarter than apes (who are able to use symbolic (sign) language, teach it to their offspring, use tools, and recognize themselves in the mirror, all feats that were once thought to be proper to the human species).
We might have reached a local maximum , though(related to the neurotransmission speed?). Neanderthals had a bigger brain, but were seemingly dumber and less culturally advanced than contemporaneous Homo Sapiens.
19 comments
[ 2.0 ms ] story [ 62.9 ms ] threadSimilarly though, you could say that pneumatic tubes are still going to be a major technology, but first we have to reach the "tubularity." When is football going to be as massively popular in the US as the rest of the world? Well, after we reach The Soccularity. Now that you presume it's a future event, you can bracket the conversation around that.
Note that I'm not doubting the real singularity's existence, just marveling at this act of rhetorical jujitsu.
Personally, the concept has always struck me as the childish fantasy of a certain personality type. But I think it's worth pointing out that in another current article on HN (http://news.ycombinator.com/item?id=205869), Freeman Dyson has some interesting things to say about it:
Q: Does any of that [concept of "The Singularity"] resonate with you?
Dyson: Not at all. The technical tricks these people are talking about are only a small part of the human experience. They vastly overestimate their own importance. I look at the world in a very different way. It's partly a matter of being old, but I look at the subway networks in cities, for instance. They also have the N-squared law. If you have a subway network with N routes, its value to the passenger is N squared. That's fine. But once you get to a certain number of routes, like 20 or so, there's very rapid growth, followed by saturation. This will also happen with chips. To some extent, it already has. It's true that the price per megaflop is going down according to Moore's Law, but what you can do with the processing power isn't increasing at the same rate. I remember doing a study on the cost of nuclear power in the 1950s, when people thought it would be very cheap. We studied what the economic effect would be if the cost of electricity were zero. The answer is, "Not much." It costs far more to use electricity than it does to make it. There's about a 5 percent drop in the GNP if electricity is free. So cheap energy is all it takes. The same is true of computing power.
Do you have some factual basis for considering the claims to be "childish fantasy"? Or is the fact that it conforms to your stereotypes is enough to dismiss it without investigating the claims?
The problem with singularity is its unchecked use of extrapolation. With any kind of prediction you ought to factor in a certain level of uncertainty based on how much you've observed about a phenomena. Considering we don't have yet what we'd call real artificial intelligence, let alone artificial intelligences creating other artificial intelligences - our level of uncertainty renders any prediction absurd.
Those who do not prepare for the coming Singularity are going to be blindsided by it when it arrives.
No it's not. At least, that's not how leading textbooks define it and it's not how McCarthy defined it. Moreover you promptly contradict it: beating the chess champ is hardly "not yet solvable by computers". Did it cease to be AI the moment it succeeded?
Those who do not prepare for the coming Singularity are going to be blindsided by it when it arrives.
This seems to me a religious sentiment expressed in religious language. Prepare ye the way of the Lord!
I'm not going to go into detail explaining it here because I'd rather spend my energy on doing that in a way that can communicate to more people. (I'm working on that.)
How do you know what I've investigated?
Which part of my claim about how it strikes me are you taking issue with?
More interestingly, what's the response to Dyson's claim that growth in processing power is not matched by growth in people's capacity to use it?
I don't. That's why I'm asking.
"More interestingly, what's the response to Dyson's claim that growth in processing power is not matched by growth in people's capacity to use it?"
Dyson seems to be referring to the dumb "extrapolated exponential growth = 'Singularity'" hypothesis. I agree that that idea is completely stupid, I don't see how they can conclude anything from that extrapolation. My argument rests on two postulates: 1. AI is feasible. 2. When AI arrives on the scene, there will be large changes.
See this for more info: http://www.singinst.org/blog/2007/09/30/three-major-singular...
All I'm saying is that there's still a lot of room for improvement, and AI will be a hugely beneficial tool toward that end.
In the US we talk about what level of battery technology will finally let EV's take off, but adding an electric rail to all roads is the limit on how useful battery technology would ever get. AKA you endup removing the limit how much energy you can move from the power plant to car but that does not help you generate energy.
Edit: I still think AI would be useful but I don't think it's going to change the world as much as people might think. It's like the internet it's new and useful but it's not all that much better than POTS + library + a good local collage.
Also, none of the arguments in that blog post are arguing that there will be unbounded/infinite growth.
The human brain has billions of routes, and uses them somewhat efficiently...
Edit: The brain uses most links but you use a tiny fraction of your overall memories at any one time.
Functional organization is of course of prime importance, but AFAIK, it's the sheer cortical surfaces that allow us to be smarter than apes (who are able to use symbolic (sign) language, teach it to their offspring, use tools, and recognize themselves in the mirror, all feats that were once thought to be proper to the human species).
We might have reached a local maximum , though(related to the neurotransmission speed?). Neanderthals had a bigger brain, but were seemingly dumber and less culturally advanced than contemporaneous Homo Sapiens.