Kurzweil continues to pitch his prognostications for strong AI so that he can sell more books and speaking engagements. Despite experts in the fields of biology [1] and cognitive science telling him that we just don't understand these things very well, he professes that the key discoveries lie just around the corner and that we should just wait. You know, there's probably a business in selling some of his followers rapture insurance...
I'm a systems biologist who is at an intersection of some of the various fields where Kurzweil makes his boldest predictions, and I'm willing to place some bets. Before we see this ludicrous explosion in technology and AI that he describes, I think we're much more likely to see all of these:
* A cure for AIDS.
* Whole-genome sequencing for under $100.
* A scale up and widespread use of tissue bioreactors.
* Identified causes and cures for neurodegeneration.
* Effective treatments and diagnostic tools for many or most forms of cancer.
* Artificial hematopoietic cells approved for human use (no more blood donations, etc.)
* Designer babies with hundreds if not thousands of selected markers.
These problems are easier than human-surpassing machine intelligence. While I don't have a problem with believing machines will one day be smarter than us, I find it offensive that this guy continues to sell us on it happening within our lifetimes. It almost certainly won't.
Kurzweil has almost no appreciation (and I would argue he has little understanding) for biological systems and their complexity. [2] We can barely understand a single cell, and yet we're supposed to understand the brain in only a few years!
The best part is, when is is called out on this argument, Kurzweil admits it. But he then proceeds to tell us silly biologists in the errors of our ways--that we are missing his point entirely. Strong AI doesn't need understanding from the field of biology in order to happen. Consciousness can't be that hard.
One of Kurzweil's important points is that humans consistently underestimate the speed of exponential growth. We have a natural bias to assume rates of growth are constant. But with exponential growth, the rate of growth is exponential, and the rate of growth of the rate of growth is exponential, etc. So although it seems like his predictions are pretty wild, they are merely based on extrapolating out the present exponential curves to the future. Keep in mind your own bias about this when criticising his conclusions. To refute him, you'd need to argue that the exponential growth was going to stop, or that his calculations about the rate of growth are wrong.
I believe that there is a large difference between the magnitude of error in my natural bias for estimating exponential rates of growth and the naive application of a mathematical model to the fields of biotechnology and cognitive science by someone untrained in either.
Computing hardware and software may be developing rapidly due to the commercial sector, but my guess is that the developments in academia (algorithms, etc.) are hardly exponential. What about the fields where happenstance discoveries drive improvement?
Innovation occurs readily in fields that have money--mobile computing is huge right now, isn't it? The question is whether this distribution of reward correlates with the reverse salients of technological advancement.
In my original post I offered several predictions for the growth of biotechnology. I think I'm placing important and valid markers to be able to identify the trend. When I see them happen I'll be happy to conclude that we're on the path to exponential growth in our field. (Note: I follow this area well enough to see where the reverse salients lie and what discoveries may inform us to better techniques.)
I'm tired of the Kurzweil meme because I feel that it gives people a comfort that we'll reach some kind of Utopia without much effort. It's wrong, and quite frankly damaging. If the people who believe in him so strongly really want to see his vision happen, perhaps they would lend their hands in developing automation tools for us biologists.
> I'm tired of the Kurzweil meme because I feel that it gives people a comfort that we'll reach some kind of Utopia without much effort.
Sounds like you're arguing we shouldn't hold comforting beliefs because, even if they are true, they might make us less productive. However, I think it is wiser to hold beliefs that are true, always. The reason is that it is, in general, far more valuable to hold correct beliefs than false ones. The fact that there are some special cases where holding false beliefs is better is useless, because there is no way to know when those cases are without first having correct beliefs.
Further, Kurzweil has (obviously) never argued that it wouldn't take any effort to bring these technologies about. What he has argued is that we will put forth the effort to make them happen. If anything, it is rather discomforting to think about how much work is going to have to go into this.
Is there anything observed in the real world that is actually of exponential growth and not in fact logistic - affected by negative feedback effects? In an effectively finite universe continued exponential growth is impossible.
I'm not sure if PZ Meyers is a reliable source to cite when he is regarded a "confrontationalist" and has been proven wrong on many of his criticisms against Kurzweil. There are always going to be examples of heavy critics, and Kurzweil has openly responded to many of these, including PZ Meyers himself in this link:
I sympathize with your position. Kurzweil seems to regularly makes strong remarks that shock me in their assertive naivety.
However, our species is incredibly poor at intuitively understanding exponential trends. Kurzweil's essential point is that our gut-feeling predictions of the future will be off substantially. Specifically, our intuitive sense of future progress significantly underestimates actual progress and the gap systematically widens as time progresses.
Neither you or I could have intuitively predicted twenty years ago the level of today's technological advancement. And we'd be even more wrong in the predicting the next twenty.
It's far too easy to dismiss Kurzweil's work as that of a superficial nutjob who wants to "sell more books and speaking engagements." Kuzweil's crunched the numbers. While he might not be right on everything, his predictions are based on data, but yours are based on human intuition that has historically been wrong.
I would be very careful of the knee-jerk reaction you display to Kurzweil's work. Keep in mind we are cognitively disabled at intuitively predicting exponential progress.
The problem is not all tasks are of linear difficulty. If building a 1% more intelligent AI takes 1% more computational power then he might have a point. However if building a 1% more intelligent AI takes 2x the computational power then you don't end up with a singularity because you don't get a strong feedback loop.
Consider starting at the low end of the scale booting windows 7 on a 2KHz CPU is about twice as fast as a 1KHz CPU and all things being equal faster processors let you get more things done but having a 1GHz vs 2GHz CPU has little effect on boot times. We might be able to build a 10x human level intelligence but if that can only build a 11x human intelligence and that can only build a 11.1x human intelligence you quickly hit a wall. Or for a more natural looking progression (10x, 15x, 18.75x, 21.09x, 22.4x). x sub 1 = 10; x(sub n) = x(n-1) * (2^n+1)/ (2^n)
The point of TS is that once you get to 1.0001x smarter than human you get to wall really quickly. If the wall is 2x or 11x, self-improving AI will reach it in no time. And we'll still have tremendous benefits from it, no matter how close the wall is to human intelligence. AI only need be a tiny bit smarter than humans to achieve things such as interstellar travel, eternal life, etc.
Slightly more intelligent than human is not the same as magic. Self improving AI is a nice theory but it assumes you can get better at everything at the same time without tradeoffs and that takes better hardware not just software. Science fiction is full of magic masquerading as technology, but it ignores the possibility that reality may not support that capability. Predicting the future requires more than simply extrapolating trends it also requires understanding physical limitations, early engines saw rapid increases in both HP and efficiency but that trend slowed down for a reason.
EX: Literal eternal life requires a universe that is capable of lasting that long. If the fundamental laws of physics limit either the age of the universe or the amount of useful work that can be done then you at some point everything dies. So no we are down from eternal to just really long life.
Also, 1.0001x a smart as a human is not really significant. 1,000 AI's 1.0001x smarter than a human is not going to be able to say beat 1 grand master at go without training or hundreds of years of preparation. I have no problem saying AI get's to quickly build better software, but building hardware at scale takes time.
Data can be manipulated to confirm biases - hence the flaw of averages and jokes about 3 types of lies. And human intuition has also led to discoveries. Consider that Poincare felt logic was inferior to intuition. Point being just because someone uses data and another uses intuition doesn't inherently make the intuitive wrong. The intuition could be built from a large experiential base with a model whose generative process is much more accurate than a naive application of data.
I was skeptic at first too. Read the first few dozen pages of The Singularity Is Near and you'll be impressed the quality of the data. By almost all accounts, technological progress datapoints match very closely an exponential curve.
I've been familiar with Kurzweil and his critics since I was in middle school. He's consistently been derided as wildly over-optimistic, but he's been right far more often than not. People originally thought his predictions of world champ-beating chess AIs by the late 90's were insane, but it actually happened about a year before he predicted. Similarly, his support and predictions for the human genome project were the subject of a great deal of ridicule from both the public and many biologists. He was right in that case, too.
I would be very careful about betting against Kurzweil's predictions given his multi-decade track record.
You bring up some good points but you ruined it from the get go by starting out with poisoning the well[1]. Right or wrong, Kurzweil is not doing it for the speaking engagements, he's not doing it for the money. He believes what he says. And where he might be wrong is because he believes it too much; he wants to live forever and for that to happen his predictions have to right.
Kurzweil's arguments about biology are definitely his weakest. He does arbitrarily chart advances as that PZ piece points out. I find that biologists get more enraged about Kurzweil's predictions for this reason, and because biologists tend to have a "my field of study is the most important in the world" attitude in general. It's hilarious that PZ says that Kurzweil's book is terrible because he got the biology wrong, then says he has some points about Moore's Law and AI.... it shows the self-importance PZ places on himself.
The only thing that will decide whether we reach a TS is not how fast medical technology advances. The only thing that matters is reaching strong AI. If we reach strong AI it's game-over for everything else. We'll have all of the bets you place virtually overnight, if our new masters decide to keep us around that is.
"These problems are easier than human-surpassing machine intelligence. While I don't have a problem with believing machines will one day be smarter than us, I find it offensive that this guy continues to sell us on it happening within our lifetimes. It almost certainly won't."
So, I am assuming you think the EU's Human Brain Project is a waste of $1.6B?
From my point of view, our ability to engineer, augment and replicate intelligence has very little to do with our understanding of biology. I really don't see your expertise in the complexity of biology as lending any insight into the ability to predict the evolution of intelligence capacity of our species. Strong AI based on emulating biological design is only one of many, many approaches to engineering intelligence..
Additionally, both your links and the basis of your arguments attempt to make a strawman of Kurzweil by attacking auxiliary points that are not critical to his primary empirical insights - rather you chose to attack choice speculative predictions.
I'm no Kurzweil fanboy, but I think his empirical insights have done nothing less than frame the discussion of the technology singularity, and I enjoy his predictions for what they are - speculation. If you are presenting yourself as a scientist looking at this argument critically, I would expect 2 things: 1) an understanding of where your expertise has, and does not have relevance, 2) a critique of Kurzweil's empirical observations, not his speculative predictions.
Consider the following: We built working airplanes over a century ago, but powered ornithopter are still quite the engineering challenge, to put it lightly.
A working understanding of biology is not necessarily a prerequisite for "machine intelligence", or more generally, a technological singularity.
The thing I don't like about Kurzweil's singularity prediction is it kind of sounds like predictions in the 1960's that we'd all be living in space by now. Yet, here we are 50 years later still burning gas in our cars but we have phenominal GPS technology in our phones. People (even scientists) tend to overestimate the "deepness" of technological advancement and underestimate the "wideness". I predict that by the time a singularity occurs that it'll be much farther out in time and may not even be a distinguishable event.
"People (even scientists) tend to overestimate the "deepness" of technological advancement and underestimate the "wideness"."
Very interesting thought, that our predictions have a bias toward "vertical" advancement -- vast improvements in areas addressed by tech that we already have -- rather than the "horizontal" creation and filling of new niches in demand.
A great example of this is Star Trek. They can travel at warp speed, but they can't use a communicator to send a video signal back to the ship, the captain has to beam down to see it himself...
In fairness, that's more of an "interesting TV" issue than a "failure of imagination" issue. A show about a bunch of people who sit around in a room teleoperating a robot probably would be a lot less interesting.
And he seems to believe it will or has to occur during his lifetime not five or 500 years from now. To me that self-centered arrogance is a sure sign someone is a nut.
One only needs to look at history to see our capacity for rapid improvement in retrospect. One of my favorite metrics is life expectancy. In 1800, most people wouldn’t expect to live past 30.
It's hard to take someone seriously writing about science who makes such a dumb mistake. The life expectancy for a 20 year old in 1800 was 64. I don't see how it can even pass the smell test for an intelligent person to assume that a couple hundred years ago people systematically keeled over at 30.
Yes, as cited it's dumb, and a good example of why averages can be so misleading (if he had been looking at median life expectancy charts, would he have made the same claim?)
But on the other hand, the improvement is very real - it's not fake data. Where did this massive improvement come from? Orders of magnitude improvements in infant/child mortality!
You are deliberately disregarding the improvements in modern medicine that caused such a dramatic drop in child/infant death. This seems like a pretty classic example of people not realizing how wildly spectacular modern society is, because they're living in it.
I'm skeptical about TS because it seems to good to be true (or too terrible), but every time I hear about it I can't help but think of this clip [1] from Neil deGrasse Tyson about how aliens would only have to be less than a percent smarter than us for us to look like peons to them. So it makes it seem inevitable that if strong AI comes to pass and AI is measures more intelligent than us, it's the end of human civilization and the begin of the era of machines. Our only hope is they decide to keep us around as pets, and are good owners. If so we'll be able to kick our feet back and enjoy eternity (should we choose to live it) debating iPhone vs. Android on Hacker News.
It's an interesting thought experiment, but the 1% is an oversimplification - necessary, perhaps, in a general audience. The 1% difference he talks about is not phenotype (reasoning ability) but genotype (DNA). It's not hard to imagine a few critical changes such as more cortical folds and improved myelination leading to most of the phenotypic cognitive difference. These could be controlled in some key fractions of a % of the genome.
While it works as a metaphor, it's flawed as a general metric. A species in which Hawking-level ability is - if not average - at least commonplace, would require the average IQ to be 2-3 standard deviations higher (relatively) on the human scale. IQ being, itself, a rather flawed metric of course :)
It sort of annoys me that whenever people discuss technological singularities its always through Kurzweil's ideas. I feel that Verner Vinge, the guy who invented the term, had a much more interesting take on it, and there are other schools of thought as well.
http://singinst.org/blog/2007/09/30/three-major-singularity-...
We've already passed the technological singularity, we just don't realize it because we assume machine intelligence is going to act like human intelligence.
Look around, computers have been more intelligent than humans in many different domains for decades. Computers can solve complex equations billions of times faster than humans, they can play chess better than people that have dedicated their entire lives to mastering the game, they can sift through the information of billions of websites in hundredths of a second, they can answer obscure and complex questions better than the best Jeopardy players, they can accurately model extremely complex systems like the world's weather, and the list goes on and on and on.
But guess what, your calculator isn't plotting to kill you because unlike humans it hasn't been programmed to do that. There may come a time when someone is evil or careless enough to program these traits into a machine capable of acting on them, but this is a game humans (and life in general) have been playing for a very long time, and we're very good at it. Good luck programming a billion years of evolution based learning into your calculator.
It's amazing how similar Kurzweil's observations about technology are to Stafford Beer's same observations in the early 1970s. Beer even draws an exponential graph and talks about a succession of S-curve shaped technology paradigms.
34 comments
[ 2.7 ms ] story [ 83.7 ms ] threadKurzweil continues to pitch his prognostications for strong AI so that he can sell more books and speaking engagements. Despite experts in the fields of biology [1] and cognitive science telling him that we just don't understand these things very well, he professes that the key discoveries lie just around the corner and that we should just wait. You know, there's probably a business in selling some of his followers rapture insurance...
I'm a systems biologist who is at an intersection of some of the various fields where Kurzweil makes his boldest predictions, and I'm willing to place some bets. Before we see this ludicrous explosion in technology and AI that he describes, I think we're much more likely to see all of these:
* A cure for AIDS.
* Whole-genome sequencing for under $100.
* A scale up and widespread use of tissue bioreactors.
* Identified causes and cures for neurodegeneration.
* Effective treatments and diagnostic tools for many or most forms of cancer.
* Artificial hematopoietic cells approved for human use (no more blood donations, etc.)
* Designer babies with hundreds if not thousands of selected markers.
These problems are easier than human-surpassing machine intelligence. While I don't have a problem with believing machines will one day be smarter than us, I find it offensive that this guy continues to sell us on it happening within our lifetimes. It almost certainly won't.
Kurzweil has almost no appreciation (and I would argue he has little understanding) for biological systems and their complexity. [2] We can barely understand a single cell, and yet we're supposed to understand the brain in only a few years!
The best part is, when is is called out on this argument, Kurzweil admits it. But he then proceeds to tell us silly biologists in the errors of our ways--that we are missing his point entirely. Strong AI doesn't need understanding from the field of biology in order to happen. Consciousness can't be that hard.
In closing, SMBC's take: http://www.smbc-comics.com/comics/20100114.gif
[1] PZ Meyers vs. Kurzweil - http://scienceblogs.com/pharyngula/2009/02/singularly_silly_...
[2] http://pipeline.corante.com/archives/2010/08/18/reverseengin...
Computing hardware and software may be developing rapidly due to the commercial sector, but my guess is that the developments in academia (algorithms, etc.) are hardly exponential. What about the fields where happenstance discoveries drive improvement?
Innovation occurs readily in fields that have money--mobile computing is huge right now, isn't it? The question is whether this distribution of reward correlates with the reverse salients of technological advancement.
In my original post I offered several predictions for the growth of biotechnology. I think I'm placing important and valid markers to be able to identify the trend. When I see them happen I'll be happy to conclude that we're on the path to exponential growth in our field. (Note: I follow this area well enough to see where the reverse salients lie and what discoveries may inform us to better techniques.)
I'm tired of the Kurzweil meme because I feel that it gives people a comfort that we'll reach some kind of Utopia without much effort. It's wrong, and quite frankly damaging. If the people who believe in him so strongly really want to see his vision happen, perhaps they would lend their hands in developing automation tools for us biologists.
Sounds like you're arguing we shouldn't hold comforting beliefs because, even if they are true, they might make us less productive. However, I think it is wiser to hold beliefs that are true, always. The reason is that it is, in general, far more valuable to hold correct beliefs than false ones. The fact that there are some special cases where holding false beliefs is better is useless, because there is no way to know when those cases are without first having correct beliefs.
Further, Kurzweil has (obviously) never argued that it wouldn't take any effort to bring these technologies about. What he has argued is that we will put forth the effort to make them happen. If anything, it is rather discomforting to think about how much work is going to have to go into this.
http://www.kurzweilai.net/ray-kurzweil-responds-to-ray-kurzw...
Another critic response here: http://nextbigfuture.com/2010/01/ray-kurzweil-responds-to-is...
However, our species is incredibly poor at intuitively understanding exponential trends. Kurzweil's essential point is that our gut-feeling predictions of the future will be off substantially. Specifically, our intuitive sense of future progress significantly underestimates actual progress and the gap systematically widens as time progresses.
Neither you or I could have intuitively predicted twenty years ago the level of today's technological advancement. And we'd be even more wrong in the predicting the next twenty.
It's far too easy to dismiss Kurzweil's work as that of a superficial nutjob who wants to "sell more books and speaking engagements." Kuzweil's crunched the numbers. While he might not be right on everything, his predictions are based on data, but yours are based on human intuition that has historically been wrong.
I would be very careful of the knee-jerk reaction you display to Kurzweil's work. Keep in mind we are cognitively disabled at intuitively predicting exponential progress.
Consider starting at the low end of the scale booting windows 7 on a 2KHz CPU is about twice as fast as a 1KHz CPU and all things being equal faster processors let you get more things done but having a 1GHz vs 2GHz CPU has little effect on boot times. We might be able to build a 10x human level intelligence but if that can only build a 11x human intelligence and that can only build a 11.1x human intelligence you quickly hit a wall. Or for a more natural looking progression (10x, 15x, 18.75x, 21.09x, 22.4x). x sub 1 = 10; x(sub n) = x(n-1) * (2^n+1)/ (2^n)
EX: Literal eternal life requires a universe that is capable of lasting that long. If the fundamental laws of physics limit either the age of the universe or the amount of useful work that can be done then you at some point everything dies. So no we are down from eternal to just really long life.
Also, 1.0001x a smart as a human is not really significant. 1,000 AI's 1.0001x smarter than a human is not going to be able to say beat 1 grand master at go without training or hundreds of years of preparation. I have no problem saying AI get's to quickly build better software, but building hardware at scale takes time.
Also see http://en.wikipedia.org/wiki/Technological_singularity#Criti...
From what I have seen only in the most superficial and optimistic sense - long term technological prediction is hardly a recognized scientific field.
Here's a good mashup from a number of third party sources: http://en.wikipedia.org/wiki/File:ParadigmShiftsFrr15Events....
This is hardly a wishy washy thesis by a history major. The data are pretty robust.
I would be very careful about betting against Kurzweil's predictions given his multi-decade track record.
Kurzweil's arguments about biology are definitely his weakest. He does arbitrarily chart advances as that PZ piece points out. I find that biologists get more enraged about Kurzweil's predictions for this reason, and because biologists tend to have a "my field of study is the most important in the world" attitude in general. It's hilarious that PZ says that Kurzweil's book is terrible because he got the biology wrong, then says he has some points about Moore's Law and AI.... it shows the self-importance PZ places on himself.
The only thing that will decide whether we reach a TS is not how fast medical technology advances. The only thing that matters is reaching strong AI. If we reach strong AI it's game-over for everything else. We'll have all of the bets you place virtually overnight, if our new masters decide to keep us around that is.
[1] http://en.wikipedia.org/wiki/Poisoning_the_well
So, I am assuming you think the EU's Human Brain Project is a waste of $1.6B?
From my point of view, our ability to engineer, augment and replicate intelligence has very little to do with our understanding of biology. I really don't see your expertise in the complexity of biology as lending any insight into the ability to predict the evolution of intelligence capacity of our species. Strong AI based on emulating biological design is only one of many, many approaches to engineering intelligence..
Additionally, both your links and the basis of your arguments attempt to make a strawman of Kurzweil by attacking auxiliary points that are not critical to his primary empirical insights - rather you chose to attack choice speculative predictions.
I'm no Kurzweil fanboy, but I think his empirical insights have done nothing less than frame the discussion of the technology singularity, and I enjoy his predictions for what they are - speculation. If you are presenting yourself as a scientist looking at this argument critically, I would expect 2 things: 1) an understanding of where your expertise has, and does not have relevance, 2) a critique of Kurzweil's empirical observations, not his speculative predictions.
A working understanding of biology is not necessarily a prerequisite for "machine intelligence", or more generally, a technological singularity.
Very interesting thought, that our predictions have a bias toward "vertical" advancement -- vast improvements in areas addressed by tech that we already have -- rather than the "horizontal" creation and filling of new niches in demand.
It's hard to take someone seriously writing about science who makes such a dumb mistake. The life expectancy for a 20 year old in 1800 was 64. I don't see how it can even pass the smell test for an intelligent person to assume that a couple hundred years ago people systematically keeled over at 30.
But on the other hand, the improvement is very real - it's not fake data. Where did this massive improvement come from? Orders of magnitude improvements in infant/child mortality!
[1] http://www.youtube.com/watch?v=w-uZZ7RdL5E
While it works as a metaphor, it's flawed as a general metric. A species in which Hawking-level ability is - if not average - at least commonplace, would require the average IQ to be 2-3 standard deviations higher (relatively) on the human scale. IQ being, itself, a rather flawed metric of course :)
Look around, computers have been more intelligent than humans in many different domains for decades. Computers can solve complex equations billions of times faster than humans, they can play chess better than people that have dedicated their entire lives to mastering the game, they can sift through the information of billions of websites in hundredths of a second, they can answer obscure and complex questions better than the best Jeopardy players, they can accurately model extremely complex systems like the world's weather, and the list goes on and on and on.
But guess what, your calculator isn't plotting to kill you because unlike humans it hasn't been programmed to do that. There may come a time when someone is evil or careless enough to program these traits into a machine capable of acting on them, but this is a game humans (and life in general) have been playing for a very long time, and we're very good at it. Good luck programming a billion years of evolution based learning into your calculator.