Well the laws of physics means it would take some finite time from the initiation of the singularity to the end of all life even if the singularity was extremely hostile to humans.
Well, a huge percentage of people who are actually working on the AI field (unlike Hawking, Gates and others) disagree with you: http://www.nickbostrom.com/papers/survey.pdf (warning: PDF (obviously))
A lot of people fail to even consider the consequences of possible breakthroughs in disciplines that helps discover more breakthroughs.
Cascading discoveries are already hard to think through very far, but discoveries that accelerate the pace at which we make discoveries make the whole thing very chaotic.
This is described in The Singularity is Near (edit: or How to Create a Brain. I dont' remember which) as the Law of Accelerating Returns, and is exactly what this journalist falls prey to. Human brains are accustomed to viewing the world linearly, not exponentially. Cascading technological breakthroughs are exponential, and therefore highly unlikely to be able to be predicted correctly without accounting for that bias.
No, what you are saying is inaccurate. There are plenty of people who've taken both sides of the predictions. Some people have been optimistic about beating the best players in the Go world:
http://www.wired.com/2014/05/the-world-of-computer-go/
As for self-driving cars, they do work under good conditions but we still are at least a decade away. Plenty of people became optimistic about self-driving cars after the 2005 DARPA Grand Challenge 11 years ago.
Agreed. I remember attending a presentation by Google regarding their self-driving cars. What they told us fascinated me. Their last main hurdle is ethics, and they used to following scenario to illustrate:
Two motorcycles are coming towards the self-driving car, and the car is forced to crash into one of them. One of the motorcyclists is not wearing his helmet, and there's a definite chance of fatal injury if crashed into by a car. By not wearing his helmet, he is also breaking the law.
The other motorcyclist is wearing her helmet, and there's less of a chance of fatal injuries.
Should the self-driving car crash into the law-abiding motorcyclist who is doing everything right but with a less chance of fatal injuries, or into the irresponsible non-helmet wearing motorcyclist where injuries could be fatal?
That's the kind of scenarios they have to deal with. The technology however, works.
Or should the Self driving car take into account that the passenger of the self driving car is, hypothetically, a male in his 60s who has raised his children already, and crash himself into the wall or drive off the cliff avoiding both of them, who are younger men with young children to raise?
This is called the "trolley problem". Here's a good read about that: https://backchannel.com/reinventing-the-trolley-problem-85f3...
TLDR: such a scenario will NOT happen. A computer is not likely at all to find itself in such a situation.
My opinion: study of such scenario is there just to give more confidence into self driving cars. Just to show people "See? We thought hard about this. Trust autonomous vehicules".
The answer to almost all of these hypothetical problems is "hit the brakes". Speed is the main killer in automobile accidents, it's a power law. Just a small reduction in velocity can have a dramatic impact on the survivability of an accident.
What about the question of whether to swerve into traffic to avoid a kid who ran into the street from between two parked cars?
Answer: you were going too fast to begin with. If you're travelling anywhere that this is a possibility, your maximum speed should be less than 20mph. A collision at that speed is almost never fatal. That speed also allows almost instant braking.
The fact that autonomous cars will be driving so slow and defensively in the suburbs will perhaps be the biggest cause for the coming backlash against them.
We're talking about exceptional circumstances. With hundreds of millions of autonomous vehicles on the road, 1 in a billion is next Tuesday. We can minimize the need for this decision, but leaving the behavior undefined will just lead to headaches when it does happen.
We have currently ~30,000 fatalities in a year. What portion of these are trolley problems? Less than 1 in a million, I'd bet. What percentage of those trolley problems are entirely avoidable through proper defensive driving techniques?
Neither. Either pick the route most likely to avoid a head on collision if the motor cyclists also swerve, or just pick a path at random if all of them result in collisions. Trying to build in ethical rules is a fool's errand because you'll never get it right and might even create additional legal liability due to the rules chosen.
I was told that driverless cars and household androids would be commonplace by the year 2000. Different people predicted different things depending on the decade you're looking at.
Seems to be a lot of people thinking Moore's Law ending means that innovation in computers is over. It's really just the opposite. We can't rely on simple scaling anymore, so we need to take more innovative steps to push computing forward.
There's a limit to what you can do without "simple" scaling, meaning, without making cheaper, faster, smaller individual gates. That is, an architecture optimized for a given problem domain will give you maybe 1000x efficiency improvement over a more general architecture, but that's it. (Typically it will be <1000x, but let's say you're lucky; the point is, that's your last 1000x, and then you'll see no improvements, ever.)
So if Moore's law is indeed over, progress will be slow, domain-specific, and decelerating due to diminishing returns (specializing for a problem domain is worth it as long as that domain is not too narrow but at some point it's too narrow to justify the investment into building specialized hardware for it; you'll be better off using something less specialized. GPGPU, which does not provide amazing performance in an absolute sense, but does beat CPUs on a large range of problems, is an example to this - any accelerator more efficient than GPGPU needs to be justifiable in the sense that GPGPU is already there wherever there's a need for GPU doing graphics, which is where GPUs do provide amazing performance in an absolute sense [you can totally beat GPGPU with more specialized hardware on most benchmarks, while you can't beat GPUs in graphics.])
I'm a chip & accelerator architect, so it's not like I'm particularly happy about this, I think I'm realistic though. A higher-caliber architect saying the same pessimistic thing is Bob Colwell.
The one nice thing about really stopping at some manufacturing technology and not being able to improve any further is that the cost of using this technology will likely continue dropping for some years. Only when it reaches the bottom will progress have truly stopped.
"it's too narrow to justify the investment into building specialized hardware for it"
FYI, if the only way to improve perf is to build custom hw, the cost and complexity of building custom hw will drop.
Erm... Not necessarily. The only way to travel quickly between 2 cities is a direct flight, but there isn't a direct flight between any 2 cities, because the cost of having such a flight didn't yet drop to the point where it's economical. So it will be with many types of custom hw.
The human brain needs about twenty watts, fits in a compact package and outperforms computers on a variety of tasks. Squishy wetware is pretty far from computronium. I think we still have a long way to go before we hit fundamental limits.
Interesting analysis, thanks. What do you think about the development of more radical redesigns like quanum computers, memristors, or anything else that might show promise to displace the traditional CPU? It seems that there's a lot going on in the more experimental side of computing right now that is reaching enough maturity where experiments might become practical products within the next decade. Even if we don't have pocket-sized quantum computers, the existence of cloud computing might make something like quantum annealing practical in the near future. Is there anything that has you excited?
Displacing CPUs is not the issue (the CPU occupies less than half the area of a modern application processor chip for instance), the problem is displacing CMOS. I don't know much about the alternatives, but I trust Colwell's skepticism.
1. We don't have to understand how it works to make something that works as well. It can always be found by trial and error.
2. That's indeed not an argument.
3. Whatever the human body does, it does not require any of its parts to be smaller than an atom. Transistors are already much smaller than a neuron. When Moore's law ends, companies will focus on making them more efficient so they can use more of them in the same time.
TLDR the singularity hasn't happened yet so it will never happen. Typical low quality technology "journalism".
"For starters, biologists acknowledge that the basic mechanisms for biological intelligence are still not completely understood, and as a result there is not a good model of human intelligence for computers to simulate."
Duh. We don't know how the brain works yet. A big chunk of The Singularity is Near deals with how we build that understanding. Once we know how the brain works, computing will take advantage.
We don't even need to know that - we just need to design a system that can evolve. No single cell knows how a brain works, yet every day thousands of brains are created from single cells.
I don't know about evolution, but most current AI work is done through first principles rather than copying the brain. Neural networks, despite the name, were only vaguely inspired by real neuroscience. And now are entirely rooted in basic calculus and maybe probability theory.
It may very well happen the other way around, that AI researchers figure out intelligence and then use their discoveries to explain neuroscience. Which is already happening a bit, with a well known researcher recently proposing a theory on how the brain might implement a variation of the backpropagation algorithm.
Figuring out the brain is like trying to reverse engineer messy spaghetti code that has gone through an obfuscating compiler. Although it may be possible to figure it out, it's probably faster to just write our own code.
It is if you have the raw computational power. At the moment our best computers are still a few orders of magnitude off the raw computational resources of the human brain.
Of course the downside of doing this is we have really no idea of what the AI is thinking or why. This may not be the wisest way to create an AI.
Consciousness does not have external signs; I have not a slightest idea what it means "act like you're conscious"; a paralyzed person is conscious after all. The only reason to believe that humans (aside of myself, obviously) are conscious is that all humans share structural similarity with me.
A draft version of the singularity is almost just one or two key insights away. With the advent of a different computing model (quantum?), we may knock one of them out.
> Duh. We don't know how the brain works yet. A big chunk of The Singularity is Near deals with how we build that understanding. Once we know how the brain works, computing will take advantage.
That's not what I hear. The singularity-is-near proponents argue that understanding the brain is unnecessary, and that there are short-cuts, whether it be trial-and-error, big data approaches, or neural networks. Otherwise it's going to take a long long time to truly understand how it works and the Singularity-won't-be-so-near.
I was referring specifically to Kurzweil's book, The Singularity is Near. He covers various current and future techniques to analyze the brain, including imaging technologies and extremely thin physical slicing. His view is that this knowledge will inform model development.
The point seems to be that people have been predicting human or super-human intelligence was just around the corner for decades, each time they've been wrong, and there's not much evidence that this time is any different. Granted, this should be common sense, but judging from a lot that gets written it unfortunately isn't.
Current productivity growth is actually somewhat low. But instead of looking at the data, people seem to latch on to anecdotal evidence, posting stories about modern day automats and saying that they're evidence that everyone is going to be replaced by robots soon.
People has been predicting super-human intelligence at the 30's for decades. (That's the original singularity prediction.) Yet, 15 years earlier, we are just having a couple of breakthroughs on AI every year. Yeah, sure, total failure...
You comment almost comes off dismissive - saying that there is only one path to the singularity; by mapping the brain first. If you look at discoveries in several areas, it's happens in leaps and bounds. Think about jet engines, radar, atomic weapons, the computer and mobile phones. All happened in huge leaps, not just by adhering to one rigorous standard and making small inroads here and there.
I'm pretty sure the singularity is a long, long way off, but it will happen. Humans have to evolve, and at some point, we won't be the top of the evolutionary ladder anymore. Whether that means evolving to something more cyborg like, or creating another race of robot humanoids (where you're going to have inherit conflict), it will happen. The dinosaurs didn't last for 65 million years without evolving several times themselves.
I don't think we're going to be able to build a good model of 'intelligence' because our definition of the word is ostensive[0]. We also have this prorblem with other big concepts we're trying to understand, such as life[1].
This article raises literally zero new points that have not been addressed over and over by futurists, "transhumanists", and others optimistic about the singularity.
1.) Moore's law still holds true at the moment. If and when it does stop the argument will begin to have some weight. The claim that Moore's law is about to stop has been made over and over since the 90s. Until then it is empty.
2) Moore's law continuing ad infinitum is not a necessity for the singularity. Distributed computing, neural chips, quantum and biological computing all provide avenues for continued vertical hardware evolution, not counting the Google method of rigging together thousands or millions of average machines to produce incredibly powerful supercomputers.
3) The sheer amount of data we are collecting continues to increase exponentially. (http://techcrunch.com/2010/08/04/schmidt-data/), much of which is applicable to machine learning algorithms which brings us to point 4.
4) The efficiency and adaptability of machine learning algorithms continue to improve year on year. See DeepMind's early videos playing video games., etc., etc. To imagine that we won't see new innovations just as incredible almost every year from here to 2050 is incredibly naive and unrealistically pessimistic.
So considering that in each of the fundamental areas that we know are necessary for an AGI -- ie raw computing power, processable/interpretable data, efficiency/cleverness of algorithms -- we are achieving exponential growth year on year, it is reasonable to conclude we will get a machine that can pass the Turing test in our lifetimes.
This is also ignoring the multitude of other areas that contribute to the likelihood of an intelligence explosion. Brain to computer and brain to brain interfaces are in their early days but already exist. As they become more practical they could lead to exponentially more efficient research. Systems like Watson will speed up scientific research as they evolve. Nootropics and electromagnetic brain stimulation also help in this area.
Capitalism strongly incentivizes innovators to produce technology that automates ever more complex problems, or create tools that improve the efficiency of creating complex problem solving technology. This is an iterative, continuous process that we are all a part of, knowingly or not.
Now that humanity has been connected with a sort of digital nervous system, and is thoroughly incentivized to all aim towards this intelligence explosion, one way or another, it is naive to think we won't continue to find novel ways of improving the efficiency of every single system we utilize no matter how macro- or microscopic, which creates an intelligence creating feedback loop. The singularity has already happened, its just not running fast enough yet for it to 'feel' magical and miraculous the way it will once human level intelligence is shown across multiple fields by integrated computer systems.
Depends who you ask. Until we clearly go a reasonable amount of time in which it proves false we should assume that any momentary hold ups/falterings are just that.
I didn't see anything in the first HPC Wire article about Moore's law not being dead. All they really said was that HPC was a useful tool to continue improving processor performance:
"Without supercomputers, we wouldn’t be able to understand what it takes to continue the march of Moore’s Law, and without this understanding, we wouldn’t be able to create more powerful supercomputers. This symbiosis is at the heart of the relationship between Moore’s Law and HPC."
As for the second HPC Wire article, this sums it up fairly well:
"What you see is that the performance per core has taken a dramatic hit around 2005-2006, but it was compensated by our ability to put more and more cores on a single chip"
Increasing the number of cores on a CPU is not the same as keeping Moore's Law. We could keep doubling transistor count every two years if we could keep doubling the silicon wafer size at the same time. Moore's Law only really makes sense if you consider it as doubling the number of transistors within the same wafer size.
IMHO the “Singularity” will be achieved when AI will invent new things. We have only reached the point where AI can only do improvements of something humans have created...
Innovation is something hard for humans, so I don't know if we will see AI innovations anytime soon.
I think your standard needs some work. Humans also only improve something other humans have created. And in niche applications AIs already outperform humans.
so the whole idea of the "singularity" has been overblown by fiction and media. The reason we call it a "singularity" is because it is part of the "acceleration" which is the observation that the rate of human progress in technology is increasing faster over time.
The "singularity" is the point beyond which the rate of change is faster than humans can internalize and react to it. Super AIs and being uploaded to the great server in the sky are just ways we imagined an event literally defined by our inability to imagine it.
The singularity could have a few outcomes:
- The classical "humanity gets left behind by it's creations" doomsday
- The rate of change caps out simply because humanity cannot drive change much faster than humanity can react to it.
- The rate of change asymptotically approaches our capacity to handle it.
Interestingly, even after the "singularity" happens these three states may be indistinguishable to us.
Right now I am leaning towards the second outcome, that the singularity has come, and now we are simply crap at predicting the future because of it. I saw this a few months ago and considered it a potential symptom of that outcome https://www.youtube.com/watch?v=aEIPfpxFrlg.
For me, the tl;dr on the entire subject of the Singularity can be summarized as:
"...the field of artificial intelligence has a long history of over-promising and under-delivering..."
That being said, I think we are finally at the dawn of "useful AI," like when my iPhone correctly guesses where I am headed and gives me a time estimate. For normal humans such as myself, this is WAY cooler and more exciting.
>That being said, I think we are finally at the dawn of "useful AI"
Were AI systems that outperformed humans in solving complex differential and integral equations useless? Were systems that diagnosed diseases better than doctors useless? Were systems that handled logistics at scale never possible before useless?
All of those were created in the past (60s, 70s, 80s).
The problem with AI was never the lack of results. It was the hype that far exceeded the results, however useful and practical they were.
The real problem with people saying "not in your lifetime" is that they fail to understand the exponential nature of a self-improving AI entity. That's the whole reason it's the singularity. Lets say we create an AI that can rewrite and improve itself. The first iteration may take a year to actually be better. But then the next version is only a month away. Then the next is only a day.
Suddenly the AI has self-improved beyond our understanding before we even realize or understand what happened.
The true singularity is dangerous precisely because we won't know when it actually happens, and it will outpace us before we can respond.
Preemptive drone strikes against AI programmers don't sound too far fetched given the direction things are going.
Of course, the singularity is a zero-sum game, in the sense that the first entity there wins. Which is why I plan on programming the singularity AI as a copy of my consciousness so that I will be the singularity god.
But the important question is whether Moore's law is a necessary part. There's a big difference between "a redundant part, if it fails, the whole thing continues" and "a necessary but not sufficient component".
Is that part even necessary? Even linear growth with a higher constant factor than humans have due to faster iterations could rapidly outgrow human capabilities if we take a hypothetical human-level AI as starting point.
No, the article is wrong on that. Moore's Law is merely the simplest and most obvious example for our current generation to understand the occasional exponential scaling of technologies.
While I concede it is unlikely anything resembling a Vingeian singularity is going to occur in my lifetime, I am actively working towards building the technology to bootstrap the singularity. It seems far more useful to spend time building new technology that people look at and consider real and surprising accomplishments (for example, AlphaGo), or to understand the neural correlates of intelligence, than to spend time pontificating about whether the singularity will occur.
Your view on singularity largely depends on your relationships with John Searle Chinese room argument; I accept it, and do not find Kurzweil views convincing.
I read the Chinese room argument, it seems really naive. It's like saying "my neurons aren't conscious therefore I am not conscious", which is clearly not the case.
We're made of matter, computers are made of matter, where is the difference?
> It's like saying "my neurons aren't conscious therefore I am not conscious", which is clearly not the case.
You need to read Searle's works again; his point is actually the opposite - we are machines made out of biological neurons, which seem to have an externally unobservable property of being conscious; we do not know what it is caused by, but we can see today that it is possible to create a simulation of human behavior by means of logical gates. We have no idea if it will have mind or not.
> We're made of matter, computers are made of matter, where is the difference?
That's an odd argument. And actually does not contradict Searle at all.
80 comments
[ 2.9 ms ] story [ 148 ms ] threadCascading discoveries are already hard to think through very far, but discoveries that accelerate the pace at which we make discoveries make the whole thing very chaotic.
As for self-driving cars, they do work under good conditions but we still are at least a decade away. Plenty of people became optimistic about self-driving cars after the 2005 DARPA Grand Challenge 11 years ago.
https://en.m.wikipedia.org/wiki/DARPA_Grand_Challenge_(2005)
Two motorcycles are coming towards the self-driving car, and the car is forced to crash into one of them. One of the motorcyclists is not wearing his helmet, and there's a definite chance of fatal injury if crashed into by a car. By not wearing his helmet, he is also breaking the law.
The other motorcyclist is wearing her helmet, and there's less of a chance of fatal injuries.
Should the self-driving car crash into the law-abiding motorcyclist who is doing everything right but with a less chance of fatal injuries, or into the irresponsible non-helmet wearing motorcyclist where injuries could be fatal?
That's the kind of scenarios they have to deal with. The technology however, works.
Just drive safely. Keep your distance and be very good at reducing your own speed as fast and as safe as possible.
Even in this hypothetical it will certainly help reducing the speed as the motorcycles will have a better chance to react.
What about the question of whether to swerve into traffic to avoid a kid who ran into the street from between two parked cars?
Answer: you were going too fast to begin with. If you're travelling anywhere that this is a possibility, your maximum speed should be less than 20mph. A collision at that speed is almost never fatal. That speed also allows almost instant braking.
The fact that autonomous cars will be driving so slow and defensively in the suburbs will perhaps be the biggest cause for the coming backlash against them.
He seems to dismiss the singularity with 3 points
1. We don't understand human intelligence
2. "We've" been wrong about it before
3. Moores law is ending
1. Who says it has to be human intelligence? 2. That's not an argument 3. Moores law -might- be ending but datacenters are getting bigger and bigger
So if Moore's law is indeed over, progress will be slow, domain-specific, and decelerating due to diminishing returns (specializing for a problem domain is worth it as long as that domain is not too narrow but at some point it's too narrow to justify the investment into building specialized hardware for it; you'll be better off using something less specialized. GPGPU, which does not provide amazing performance in an absolute sense, but does beat CPUs on a large range of problems, is an example to this - any accelerator more efficient than GPGPU needs to be justifiable in the sense that GPGPU is already there wherever there's a need for GPU doing graphics, which is where GPUs do provide amazing performance in an absolute sense [you can totally beat GPGPU with more specialized hardware on most benchmarks, while you can't beat GPUs in graphics.])
I'm a chip & accelerator architect, so it's not like I'm particularly happy about this, I think I'm realistic though. A higher-caliber architect saying the same pessimistic thing is Bob Colwell.
The one nice thing about really stopping at some manufacturing technology and not being able to improve any further is that the cost of using this technology will likely continue dropping for some years. Only when it reaches the bottom will progress have truly stopped.
2. That's indeed not an argument.
3. Whatever the human body does, it does not require any of its parts to be smaller than an atom. Transistors are already much smaller than a neuron. When Moore's law ends, companies will focus on making them more efficient so they can use more of them in the same time.
"For starters, biologists acknowledge that the basic mechanisms for biological intelligence are still not completely understood, and as a result there is not a good model of human intelligence for computers to simulate."
Duh. We don't know how the brain works yet. A big chunk of The Singularity is Near deals with how we build that understanding. Once we know how the brain works, computing will take advantage.
It may very well happen the other way around, that AI researchers figure out intelligence and then use their discoveries to explain neuroscience. Which is already happening a bit, with a well known researcher recently proposing a theory on how the brain might implement a variation of the backpropagation algorithm.
Figuring out the brain is like trying to reverse engineer messy spaghetti code that has gone through an obfuscating compiler. Although it may be possible to figure it out, it's probably faster to just write our own code.
Of course the downside of doing this is we have really no idea of what the AI is thinking or why. This may not be the wisest way to create an AI.
That's not what I hear. The singularity-is-near proponents argue that understanding the brain is unnecessary, and that there are short-cuts, whether it be trial-and-error, big data approaches, or neural networks. Otherwise it's going to take a long long time to truly understand how it works and the Singularity-won't-be-so-near.
Current productivity growth is actually somewhat low. But instead of looking at the data, people seem to latch on to anecdotal evidence, posting stories about modern day automats and saying that they're evidence that everyone is going to be replaced by robots soon.
Gotta love people that can't read a date.
I'm pretty sure the singularity is a long, long way off, but it will happen. Humans have to evolve, and at some point, we won't be the top of the evolutionary ladder anymore. Whether that means evolving to something more cyborg like, or creating another race of robot humanoids (where you're going to have inherit conflict), it will happen. The dinosaurs didn't last for 65 million years without evolving several times themselves.
No, we don't. We are in power of our own destiny. We don't have to create a race of robot humanoids. There is nothing inevitable about any of this.
[0] https://en.wikipedia.org/wiki/Ostensive_definition
[1] https://en.wikipedia.org/wiki/Life
1.) Moore's law still holds true at the moment. If and when it does stop the argument will begin to have some weight. The claim that Moore's law is about to stop has been made over and over since the 90s. Until then it is empty.
2) Moore's law continuing ad infinitum is not a necessity for the singularity. Distributed computing, neural chips, quantum and biological computing all provide avenues for continued vertical hardware evolution, not counting the Google method of rigging together thousands or millions of average machines to produce incredibly powerful supercomputers.
3) The sheer amount of data we are collecting continues to increase exponentially. (http://techcrunch.com/2010/08/04/schmidt-data/), much of which is applicable to machine learning algorithms which brings us to point 4.
4) The efficiency and adaptability of machine learning algorithms continue to improve year on year. See DeepMind's early videos playing video games., etc., etc. To imagine that we won't see new innovations just as incredible almost every year from here to 2050 is incredibly naive and unrealistically pessimistic.
So considering that in each of the fundamental areas that we know are necessary for an AGI -- ie raw computing power, processable/interpretable data, efficiency/cleverness of algorithms -- we are achieving exponential growth year on year, it is reasonable to conclude we will get a machine that can pass the Turing test in our lifetimes.
This is also ignoring the multitude of other areas that contribute to the likelihood of an intelligence explosion. Brain to computer and brain to brain interfaces are in their early days but already exist. As they become more practical they could lead to exponentially more efficient research. Systems like Watson will speed up scientific research as they evolve. Nootropics and electromagnetic brain stimulation also help in this area.
Capitalism strongly incentivizes innovators to produce technology that automates ever more complex problems, or create tools that improve the efficiency of creating complex problem solving technology. This is an iterative, continuous process that we are all a part of, knowingly or not.
Now that humanity has been connected with a sort of digital nervous system, and is thoroughly incentivized to all aim towards this intelligence explosion, one way or another, it is naive to think we won't continue to find novel ways of improving the efficiency of every single system we utilize no matter how macro- or microscopic, which creates an intelligence creating feedback loop. The singularity has already happened, its just not running fast enough yet for it to 'feel' magical and miraculous the way it will once human level intelligence is shown across multiple fields by integrated computer systems.
No it doesn't, it's already broken. Consider:
http://arstechnica.co.uk/gadgets/2015/07/intel-confirms-tick...
http://www.hpcwire.com/2016/01/11/moores-law-not-dead-and-in...
http://www.hpcwire.com/2015/11/20/top500/
"Without supercomputers, we wouldn’t be able to understand what it takes to continue the march of Moore’s Law, and without this understanding, we wouldn’t be able to create more powerful supercomputers. This symbiosis is at the heart of the relationship between Moore’s Law and HPC."
As for the second HPC Wire article, this sums it up fairly well:
"What you see is that the performance per core has taken a dramatic hit around 2005-2006, but it was compensated by our ability to put more and more cores on a single chip"
Increasing the number of cores on a CPU is not the same as keeping Moore's Law. We could keep doubling transistor count every two years if we could keep doubling the silicon wafer size at the same time. Moore's Law only really makes sense if you consider it as doubling the number of transistors within the same wafer size.
http://www.damninteresting.com/on-the-origin-of-circuits/
Ctrl+F for "baffling"
The "singularity" is the point beyond which the rate of change is faster than humans can internalize and react to it. Super AIs and being uploaded to the great server in the sky are just ways we imagined an event literally defined by our inability to imagine it.
The singularity could have a few outcomes:
- The classical "humanity gets left behind by it's creations" doomsday
- The rate of change caps out simply because humanity cannot drive change much faster than humanity can react to it.
- The rate of change asymptotically approaches our capacity to handle it.
Interestingly, even after the "singularity" happens these three states may be indistinguishable to us.
Right now I am leaning towards the second outcome, that the singularity has come, and now we are simply crap at predicting the future because of it. I saw this a few months ago and considered it a potential symptom of that outcome https://www.youtube.com/watch?v=aEIPfpxFrlg.
"...the field of artificial intelligence has a long history of over-promising and under-delivering..."
That being said, I think we are finally at the dawn of "useful AI," like when my iPhone correctly guesses where I am headed and gives me a time estimate. For normal humans such as myself, this is WAY cooler and more exciting.
Were AI systems that outperformed humans in solving complex differential and integral equations useless? Were systems that diagnosed diseases better than doctors useless? Were systems that handled logistics at scale never possible before useless?
All of those were created in the past (60s, 70s, 80s).
The problem with AI was never the lack of results. It was the hype that far exceeded the results, however useful and practical they were.
Suddenly the AI has self-improved beyond our understanding before we even realize or understand what happened.
The true singularity is dangerous precisely because we won't know when it actually happens, and it will outpace us before we can respond.
Preemptive drone strikes against AI programmers don't sound too far fetched given the direction things are going.
Of course, the singularity is a zero-sum game, in the sense that the first entity there wins. Which is why I plan on programming the singularity AI as a copy of my consciousness so that I will be the singularity god.
Read the two Wait But Why posts on this: http://waitbutwhy.com/2015/01/artificial-intelligence-revolu...
http://waitbutwhy.com/2015/01/artificial-intelligence-revolu...
Is that part even necessary? Even linear growth with a higher constant factor than humans have due to faster iterations could rapidly outgrow human capabilities if we take a hypothetical human-level AI as starting point.
We're made of matter, computers are made of matter, where is the difference?
You need to read Searle's works again; his point is actually the opposite - we are machines made out of biological neurons, which seem to have an externally unobservable property of being conscious; we do not know what it is caused by, but we can see today that it is possible to create a simulation of human behavior by means of logical gates. We have no idea if it will have mind or not.
> We're made of matter, computers are made of matter, where is the difference?
That's an odd argument. And actually does not contradict Searle at all.