The Singularity is (not) Near
I’m not a big reader on issues surrounding The Singularity, but sometimes I find knowing less about the concept actually helps to critique it better, because my brain isn’t overloaded with other peoples’ arguments and I can actually find my own opinion and draw my own conclusions.
The concept of The Singularity revolves around the idea that one day, the human race will lose control of its own evolution to superior machines (artificial intelligences) which, combined with a range of other technologies, will combine to make human beings essentially obsolete. Some versions of the theory say that humans will be disposed of, others speculate that we’ll be transformed into something post-human.
This is dopey.
Okay, I’ll concede that the nature of humanosphere many hundreds or thousands of years from now is a hard thing to predict, but a lot of singularity proponents are throwing around dates like 2030. Bollocks.
There’s a couple of things going on here. There’s a strand of science fiction that has always attracted writers who aren’t overly fond of human beings. This is more about deep-seated geekish insecurity complexes wishing for techno-revenge upon the bullies who flushed their heads down the toilet in highschool than it is about the future. Or rather, it’s looking at the future from within that prism. Lots of science fiction writers have underestimated human biology, to say nothing of human intelligence. The old stereotype of the nerdish kid who falls in love with computers because he can’t make heads or tails of people is so enduring because it’s true.
When Arthur C Clarke created thinking computer HAL in 2001, that seemed a reasonable timeline -- Clarke wrote the original short story “The Sentinel” in the early 1950s, and obviously, 2001 was a long way away. Surely we’d have sentient computers by then, right? The truth is that for all the advances since then, today in 2005, we’re really not much closer. Intelligence may be comprehensible, but consciousness is baffling. It takes armies of neuroscientists to approach even a vague (and usually controversial) understanding of even basic neuromechanical processes. How can you replicate the creation of software you don’t understand (consciousness) on completely different hardware (computers)? We don't even know how it works on human hardware yet. I’ve no doubt it will happen someday, but thirty years? I’ll be astonished to see it in my lifetime, and I’m usually very optimistic about these things.
Next assumption; in the ‘humans will be disposed of’ version of the theory, it is assumed that when we create artificial intelligence, it will be cold, emotionless, and will have no need of people. The assumption here is that emotion and sentiment are redundant... which once again, is more a reflection on the people doing the assuming than anything else. Humans evolved emotions because they’re useful. Emotions give us purpose, and direction. Technical skills are all about ‘how’, but emotions are about ‘why’. Sentiment is nothing if not a higher intellectual function -- it’s exactly this prejudice about future AIs that inspired me in part to create Cassandra Kresnov. It may turn out it’s not even possible to have higher intelligence without emotions -- they may be a natural side-effect. Or they may not be, but such higher AIs may wander aimlessly, with no sense of direction to guide their existence. Or they may become self-destructive, narcissistic psychopaths.
Think about it. The first question all higher intelligences ask is ‘why?’ Understanding cause and effect is the reason evolution reckoned higher intelligence was a good idea in the first place. Cause and effect is a vital thing to understand, if any lifeform is going to last long beyond its metaphorical cradle. Any intelligence continually asking ‘why?’, in an attempt to understand cause and effect, will at some point become philosophical -- it’s unavoidable, because as anyone with any interest in philosophy knows, once you’ve started asking ‘why?’, you only arrive at even more questions, followed by the conclusion that many things are simply unknowable. From philosophy comes questions of morality, which are also pragmatic -- morality is truly about social organisation, and this would be just as vital for groups of AIs, even those operating outside of human purview, as it is for humans beings. But SF singularity-speculators assume superior AIs will just be intelligent toasters, with no imagination.
How on earth are they going to be able to create their own future if they can’t first imagine it? A society open to the exploration of imaginative concepts is an open society. This is how new knowledge is acquired -- no intelligence can have instinctive knowledge hardwired into its circuits, knowledge must be gained. That’s called learning. To learn, one must be open to new ideas -- any new ideas, since to prejudge is to make the mistake of narrowing your options. Imaginative, open societies will therefore learn faster than closed ones. What a lot of singularity-theorists are suggesting is a kind of AI-dictatorship. But dictatorships collapse in time when confronted by open societies because dictatorships don’t learn as efficiently. Would these super-intelligent AIs be stupid enough to ignore this fact from human history? If so, would they in fact be as good at learning as we regular humans? If not, do we then have anything to fear?
Furthermore, AIs will only cease to be dependent on humans if we give them control of all means of production. When they can run their own economy, produce their own power, and completely govern their physical environment, then they won’t need us... but the prospect of them acquiring all of this at the same moment seems like another of those ‘magic-wand’ imaginings that some futurists are so fond of. The ultimate ‘magic-wand’, of course, is nanotechnology, that miraculous ability of machines to produce unlimited quantities of whatever they want from mid-air. Now, I can’t say this won’t happen... which of course is the wonder of these self-perpetuating arguments -- they can’t be contradicted because no one really knows what the hell they’re about. It’s like religion in that sense. But my guess is that nanotech will have its limits, like everything has its limits, and we humans won’t be quite so stupid as to hand it all over for AIs to control anyway.
In other words, the evolution of AI will be gradual, progressive, and will emerge from within existing human systems. AIs, forced by necessity to work WITH humans, will need to arrive at understandings of how complex human systems are organised. This means culture. It also means emotion, and sentiment, and morality... which as I’ve said, are not entirely useless, and exist for a reason. I can just imagine the shock of some computer-geek types that their wonderful new, coldly rational AIs, instead of wanting to play chess and discuss quantum physics all day, actually become big soap opera fans and demand realistic arms and hands be made for their physical bodies so they can play guitar like Eddie Van Halen.
It’s only rational, after all, that if an AI, with an advanced learning-capacity, wanted to learn quickly about the key organising social principles, it would first draw upon the most extensive database of social organisation available -- human society. That’s the important bit, and the thing that would save humanity from a nasty fate -- if we raised AIs like we raised kids, giving them good reason to appreciate and even love (yes, another higher intellectual function) their parents, we wouldn’t have so much to be scared of. Of course, there’s also the Battlestar Galactica scenario, where we create them, then decide we don’t like them and dump them somewhere on their own, and allow them not only create their own society free from human input from the ground up, but also give them a reason to hold a grudge. If we mess that up, we could well be screwed. But then, if we raised our kids that badly, we’d probably deserve it.
Singularity proponents also talk about the exponentially-increasing pace of technological advancement. That would be nice, if it existed. Yes, technology is moving much faster now, and we’re jumping from techno-age to techno-age far faster than, say, the gap between stone age and bronze age. So what?
In many fields, the pace of advancement is clearly slowing. Take aviation. In 1905, the Wright Brothers made the first powered flight. By 1915, airplanes were actually recognisable as such by today’s standards, and not just powered gliders. By 1945, there were flying some of the most powerful, high-performance prop-driven aircraft ever made, which remain impressive even today. Also in 1945, the first military jets were flying in combat. Forty years, to go from not being able to do it at all, to having almost mastered it. That’s extraordinary change. In the subsequent sixty years, however, we’ve gone a bit faster, gone a bit further, and gone a big higher. The envelope has expanded, but the basic principles remain pretty similar. It slowed down fast, and these days, it takes ten years or more to build a new high-tech aeroplane. Back in WW2, many completely new models would come out every year.
We see it today with computers, consumer electronics, cars, you name it. Technologies move fast when they’re young, but the more they mature, the harder and more expensive it becomes to make even minor improvements. There's a narrow-focus view here of computer technology, and the oft-recited figure that computing power doubles every 18 months, which of course means in absolute terms, an exponential increase in computing power. But computing power itself is only one axis of development. Power and complexity alone don't make consciousness, nor even basic intelligence. Today's super computers are vastly more complex in raw computing power than, say, the brain of an insect, but most insects can perform tasks even supercomputer powered robots would be baffled by. It's the way that computing power is organised that makes the computer 'intelligent'. And we simply don't understand that structural organisation yet. It's a problem of learning and problem-solving that we face, not one of industrial-powered technology growth. And learning is never exponential, it's the same long, laborious process of research and experimentation it's always been.
The point is that, like in any physics equation, for every action there’s an equal and opposite reaction. The harder you accelerate, the more drag you encounter. The further you go, the harder it gets. The faster you develop a new technology, the more problems you'll encounter. That’s a universal law that seems to apply pretty much to everything. That’s why ‘magic-wand’ singularity arguments should always be (in my opinion) treated with healthy suspicion. Technological progress, like any progress, is a process, not an event. In that light, even the name, ‘singularity’ seems unlikely. Nothing in the universe is ever that simple, or that singular.
In truth, The Singularity is not actually an imaginative concept -- rather it represents a failure of imagination. It represents the idea that all this technology is somehow going to rupture the very fabric of existence as we know it. Which is essentially luddite argument, the inability of people to comprehend that life as we know it can possibly survive impending change. When the internet arrived it was ‘the death of the office’, because everyone can work at home. When computers arrived, it was the death of paper. Many environmentalist-types continue to insist that modernity itself represents the death of nature, and technologists have been proclaiming everything from the death of TV, radio, books, cinema, even culture itself, for as long as I can remember. Y2K was proclaimed by some to be the death of technology, or even civilisation. But they’re all still here, only the form changes.
Don’t underestimate the universe, guys. It’s complex, and it doesn’t give up its secrets easily. Dominating even small portions of it remains tough to do, and the kind of Godlike powers that The Singularity is supposed to bestow upon us or our artificial children remain a long, long way off yet.
The concept of The Singularity revolves around the idea that one day, the human race will lose control of its own evolution to superior machines (artificial intelligences) which, combined with a range of other technologies, will combine to make human beings essentially obsolete. Some versions of the theory say that humans will be disposed of, others speculate that we’ll be transformed into something post-human.
This is dopey.
Okay, I’ll concede that the nature of humanosphere many hundreds or thousands of years from now is a hard thing to predict, but a lot of singularity proponents are throwing around dates like 2030. Bollocks.
There’s a couple of things going on here. There’s a strand of science fiction that has always attracted writers who aren’t overly fond of human beings. This is more about deep-seated geekish insecurity complexes wishing for techno-revenge upon the bullies who flushed their heads down the toilet in highschool than it is about the future. Or rather, it’s looking at the future from within that prism. Lots of science fiction writers have underestimated human biology, to say nothing of human intelligence. The old stereotype of the nerdish kid who falls in love with computers because he can’t make heads or tails of people is so enduring because it’s true.
When Arthur C Clarke created thinking computer HAL in 2001, that seemed a reasonable timeline -- Clarke wrote the original short story “The Sentinel” in the early 1950s, and obviously, 2001 was a long way away. Surely we’d have sentient computers by then, right? The truth is that for all the advances since then, today in 2005, we’re really not much closer. Intelligence may be comprehensible, but consciousness is baffling. It takes armies of neuroscientists to approach even a vague (and usually controversial) understanding of even basic neuromechanical processes. How can you replicate the creation of software you don’t understand (consciousness) on completely different hardware (computers)? We don't even know how it works on human hardware yet. I’ve no doubt it will happen someday, but thirty years? I’ll be astonished to see it in my lifetime, and I’m usually very optimistic about these things.
Next assumption; in the ‘humans will be disposed of’ version of the theory, it is assumed that when we create artificial intelligence, it will be cold, emotionless, and will have no need of people. The assumption here is that emotion and sentiment are redundant... which once again, is more a reflection on the people doing the assuming than anything else. Humans evolved emotions because they’re useful. Emotions give us purpose, and direction. Technical skills are all about ‘how’, but emotions are about ‘why’. Sentiment is nothing if not a higher intellectual function -- it’s exactly this prejudice about future AIs that inspired me in part to create Cassandra Kresnov. It may turn out it’s not even possible to have higher intelligence without emotions -- they may be a natural side-effect. Or they may not be, but such higher AIs may wander aimlessly, with no sense of direction to guide their existence. Or they may become self-destructive, narcissistic psychopaths.
Think about it. The first question all higher intelligences ask is ‘why?’ Understanding cause and effect is the reason evolution reckoned higher intelligence was a good idea in the first place. Cause and effect is a vital thing to understand, if any lifeform is going to last long beyond its metaphorical cradle. Any intelligence continually asking ‘why?’, in an attempt to understand cause and effect, will at some point become philosophical -- it’s unavoidable, because as anyone with any interest in philosophy knows, once you’ve started asking ‘why?’, you only arrive at even more questions, followed by the conclusion that many things are simply unknowable. From philosophy comes questions of morality, which are also pragmatic -- morality is truly about social organisation, and this would be just as vital for groups of AIs, even those operating outside of human purview, as it is for humans beings. But SF singularity-speculators assume superior AIs will just be intelligent toasters, with no imagination.
How on earth are they going to be able to create their own future if they can’t first imagine it? A society open to the exploration of imaginative concepts is an open society. This is how new knowledge is acquired -- no intelligence can have instinctive knowledge hardwired into its circuits, knowledge must be gained. That’s called learning. To learn, one must be open to new ideas -- any new ideas, since to prejudge is to make the mistake of narrowing your options. Imaginative, open societies will therefore learn faster than closed ones. What a lot of singularity-theorists are suggesting is a kind of AI-dictatorship. But dictatorships collapse in time when confronted by open societies because dictatorships don’t learn as efficiently. Would these super-intelligent AIs be stupid enough to ignore this fact from human history? If so, would they in fact be as good at learning as we regular humans? If not, do we then have anything to fear?
Furthermore, AIs will only cease to be dependent on humans if we give them control of all means of production. When they can run their own economy, produce their own power, and completely govern their physical environment, then they won’t need us... but the prospect of them acquiring all of this at the same moment seems like another of those ‘magic-wand’ imaginings that some futurists are so fond of. The ultimate ‘magic-wand’, of course, is nanotechnology, that miraculous ability of machines to produce unlimited quantities of whatever they want from mid-air. Now, I can’t say this won’t happen... which of course is the wonder of these self-perpetuating arguments -- they can’t be contradicted because no one really knows what the hell they’re about. It’s like religion in that sense. But my guess is that nanotech will have its limits, like everything has its limits, and we humans won’t be quite so stupid as to hand it all over for AIs to control anyway.
In other words, the evolution of AI will be gradual, progressive, and will emerge from within existing human systems. AIs, forced by necessity to work WITH humans, will need to arrive at understandings of how complex human systems are organised. This means culture. It also means emotion, and sentiment, and morality... which as I’ve said, are not entirely useless, and exist for a reason. I can just imagine the shock of some computer-geek types that their wonderful new, coldly rational AIs, instead of wanting to play chess and discuss quantum physics all day, actually become big soap opera fans and demand realistic arms and hands be made for their physical bodies so they can play guitar like Eddie Van Halen.
It’s only rational, after all, that if an AI, with an advanced learning-capacity, wanted to learn quickly about the key organising social principles, it would first draw upon the most extensive database of social organisation available -- human society. That’s the important bit, and the thing that would save humanity from a nasty fate -- if we raised AIs like we raised kids, giving them good reason to appreciate and even love (yes, another higher intellectual function) their parents, we wouldn’t have so much to be scared of. Of course, there’s also the Battlestar Galactica scenario, where we create them, then decide we don’t like them and dump them somewhere on their own, and allow them not only create their own society free from human input from the ground up, but also give them a reason to hold a grudge. If we mess that up, we could well be screwed. But then, if we raised our kids that badly, we’d probably deserve it.
Singularity proponents also talk about the exponentially-increasing pace of technological advancement. That would be nice, if it existed. Yes, technology is moving much faster now, and we’re jumping from techno-age to techno-age far faster than, say, the gap between stone age and bronze age. So what?
In many fields, the pace of advancement is clearly slowing. Take aviation. In 1905, the Wright Brothers made the first powered flight. By 1915, airplanes were actually recognisable as such by today’s standards, and not just powered gliders. By 1945, there were flying some of the most powerful, high-performance prop-driven aircraft ever made, which remain impressive even today. Also in 1945, the first military jets were flying in combat. Forty years, to go from not being able to do it at all, to having almost mastered it. That’s extraordinary change. In the subsequent sixty years, however, we’ve gone a bit faster, gone a bit further, and gone a big higher. The envelope has expanded, but the basic principles remain pretty similar. It slowed down fast, and these days, it takes ten years or more to build a new high-tech aeroplane. Back in WW2, many completely new models would come out every year.
We see it today with computers, consumer electronics, cars, you name it. Technologies move fast when they’re young, but the more they mature, the harder and more expensive it becomes to make even minor improvements. There's a narrow-focus view here of computer technology, and the oft-recited figure that computing power doubles every 18 months, which of course means in absolute terms, an exponential increase in computing power. But computing power itself is only one axis of development. Power and complexity alone don't make consciousness, nor even basic intelligence. Today's super computers are vastly more complex in raw computing power than, say, the brain of an insect, but most insects can perform tasks even supercomputer powered robots would be baffled by. It's the way that computing power is organised that makes the computer 'intelligent'. And we simply don't understand that structural organisation yet. It's a problem of learning and problem-solving that we face, not one of industrial-powered technology growth. And learning is never exponential, it's the same long, laborious process of research and experimentation it's always been.
The point is that, like in any physics equation, for every action there’s an equal and opposite reaction. The harder you accelerate, the more drag you encounter. The further you go, the harder it gets. The faster you develop a new technology, the more problems you'll encounter. That’s a universal law that seems to apply pretty much to everything. That’s why ‘magic-wand’ singularity arguments should always be (in my opinion) treated with healthy suspicion. Technological progress, like any progress, is a process, not an event. In that light, even the name, ‘singularity’ seems unlikely. Nothing in the universe is ever that simple, or that singular.
In truth, The Singularity is not actually an imaginative concept -- rather it represents a failure of imagination. It represents the idea that all this technology is somehow going to rupture the very fabric of existence as we know it. Which is essentially luddite argument, the inability of people to comprehend that life as we know it can possibly survive impending change. When the internet arrived it was ‘the death of the office’, because everyone can work at home. When computers arrived, it was the death of paper. Many environmentalist-types continue to insist that modernity itself represents the death of nature, and technologists have been proclaiming everything from the death of TV, radio, books, cinema, even culture itself, for as long as I can remember. Y2K was proclaimed by some to be the death of technology, or even civilisation. But they’re all still here, only the form changes.
Don’t underestimate the universe, guys. It’s complex, and it doesn’t give up its secrets easily. Dominating even small portions of it remains tough to do, and the kind of Godlike powers that The Singularity is supposed to bestow upon us or our artificial children remain a long, long way off yet.
6 Comments:
While we may be a ways off programming consciousness, we are not that far off from simulating behavior. So, we may create something fully interactive, that isn't necessarily self-aware, but can make decisions and function autonomously as if it were self aware. I'm a big fan of Greg Egan, and love scenes where he'll have the protagonist ask an artificial intelligence, with whom he's had an interesting and complex conversation, "Excuse me for asking, but are you sentient?" To which it will reply something like, "Oh, no sir. Sentience isn't required for my job description."
Hi Lou
It's really the ultimate question in semantics, isn't it? Who's sentient, and who isn't? You could be having a conversation with your ATM, and your ATM insists that it's sentient. And you say, 'well prove it'.
And it replies, 'No, you prove you're sentient first'.
How? I think it takes a lot more than conversation to establish sentience one way or the other, which is why the Turing test turns out to be so silly. And the fact that we struggle with these definitions so much only goes to demnstrate, I think, just how far away we are from understanding what sentience actually is.
Agreed,
but I'd rather grant the possibility of sentience to an AI claiming that it had it and be wrong, than deny an AI sentience and be wrong! If consciousnes is an emerging property of complexity, then our creations may give rise to sentience without the requirement that we understand the process whereby it arises.
Absolutely. Which is why abritrary definitions of any kind are so problematic. If there's one thing we know about the one existing model of sentience we're aware of, it's that it's extremely unpredictable. AIs could turn out to be anything, both structurally, and in terms of personality. But the most common predictions of The Singularity I hear are predicated on certain (I think) fairly narrow assumptions of what AIs will be.
What cognitive tasks can an insect perform that cannot be done by a computer?
It can perform tasks truly randomly
Post a Comment
<< Home