Slashdot Mirror
Ray Kurzweil's Vision of the Singularity, In Movie Form
destinyland writes "AI researcher Ben Goertzel peeks at the new Ray Kurzweil movie (Transcendent Man), and
gives it 'two nano-enhanced cyberthumbs way, way up!' But in an exchange with Kurzweil after the screening, Goertzel debates the
post-human future, asking whether individuality can survive in a machine-augmented brain.
The documentary covers radical futurism, but also includes alternate viewpoints.
'Would I build these machines, if I knew there was a strong chance they would destroy humanity?' asks evolvable hardware researcher Hugo de Garis. His answer? 'Yeah.'" Note, the movie is about Kurzweil and futurism, not by Kurzweil. Update: 05/06 20:57 GMT by T : Note, Singularity Hub has a review up, too.
Computers become smarter than humans. Human consciousness becomes downloadable ...ermm ...somehow... and we live forever as computers.
Wow. What a visionary.
Seriously though, you have to congratulate a guy from becoming so well known with people believing what he's saying as actually probable. I doubt anyone else could even sell this shit as a sci-fi B-movie plot.
... we'll be wrong. My own theory is that strong AI is the ultimate weapon and that it will never ever fall into the hands of the likes of you and me. Whether the machines get out of control is irrelevant; eventually the parties that control them will be slugging it out with weapons powerful enough to make life here hardly worth living. I expect to be dead before then, thankfully. But remember the first sentence of this post.
The key here is that Ray bases this prediction on past observation of things like Moore's law. Even though he does cherry pick and that there is no guarantee that it would always continue in such a fashion, the idea that distributed system improvements are exponential isn't that far fetched.
So basically what he is saying is that if the future behaves like the past then we will see so major changes shortly simply because we'll have processing out the wazoo.
Even the Moore himself thinks this will at least last til 2018 when silicon transistors reach their theoretical limit on the atomic scale. Whether or not the industry finds a suitable replacement for silicon or finds another way to go about making processors is another thing all together.
My bet is that Intel, IBM, and AMD are putting the big bucks on getting past the silicon limit because that is their money cow.
So if the limit does continue that things like Blue Brain Project will have an easier time running their simulations.
I don't know about the whole Nanotech emergence, but at least it looks like we might get the AI thing solved in at least 50 years.
"I am the king of the Romans, and am superior to rules of grammar!"
-Sigismund, Holy Roman Emperor (1368-1437)
Comment removed based on user account deletion
I just saw an interview with him last night, where he discussed full power computers the size of a blood cell, us mapping out our minds for the good of all, etc. It reminded me of the utopian 1950s vision of the space age, where we'd all be floating around space circa 2001: Its not going to happen.
First he's ignoring some physical limitations, such as with the size of computers, but that's not even the main issue. The main issue is that he's ignoring politics. He's ignoring the fact that technologies which comes into existence get used by existing power structures to perpetuate their rule, not necessarily "for the good of all". Mind reading technology he predicts won't be floating around for everybody to play with, it will be used by intelligence agencies to prop of regimes which will scan the brains of potential opposition, consolidating their rule. Quantum computers, given their code breaking potential, won't be in public hands either, but rather will strengthen surveillance operations of those who already do this stuff.
In other words, this technology won't make the past go away any more than the advent of the atom bomb made middle ages Islamic mujahadeen go away. Rather it will combine with current political realities to accentuate the ancient political realities of haves and have not that date back to ancient times.
-The art of programming is the pursuit of absolute simplicity.
Consciousness is an instantaneous phenomenon and there is no continuity of "self".
However, just because something ("Consciousness" in this case) is emergent and cannot be well described by the sum of the parts doesn't mean we shouldn't at least consider what these sorts of human/machine interfaces might do to our perception of self in the future if ever they exist.
My prediction: as long as I can still enjoy a fine single malt - and some bacon from time to time I'll consider the future a smashing success.
Prediction: The real iPhone killer is going to be sex robots from Japan. Think about it.
...I'm not sure I'd mind the thought of my organic representation being destroyed, even if it could have continued existence in parallel.
Sure, but who's going to break the bad news to your "organic representation"?
Death is death even if there are 100 more copies of you.
-DF
Kurzweil's predictions aren't just based on modern trends but historical shifts as well. In fact, I thought one of the big pieces he shows is a graph of 'paradigm shifting events' against time. These would be technologies that changed everything at the time; things like agriculture, the printing press, nuclear power, the transistor, etc.
The point isn't the gradual improvement of transistor technology that make the singularaty interesting, it's that transistors will be old news in 20 years; replaced by some new technology that we can't even speculate about right now. It's about the shifts, not the gradual evolution.
Actually, I'm pretty sure with time travel I could fairly trivially build about the strongest AI possible. When you can perform an infinite number of operations in an arbitrarily short amount of time, quite a stupid algorithm can produce some pretty smart results.
..this story falls in the category of "sh#t that's never gonna happen".
I'm going to have to strongly disagree with you. I've been studying neuroscience for a while and specifically, neural simulations in software. Our knowledge of the brain is quite advanced. We're not on the cusp of sentient AI, but my honest opinion is that we're probably only a bit over a decade from it. Certainly no more than 2 decades from it.
There's been a neural prosthetic for at least 6 years already. Granted, it acts more as a DSP than a real hippocampus, but still, it's a major feat and it won't be long until a more faithful reproduction of the hippocampus can be done.
While there are still details about how various neural circuits are connected, this information will be figured out in the next 10 years. neuroscience research won't be the bottleneck for sentient AI, however. Computer tech will be. The brain contains tens to hundreds of trillions of synapses (synapses are really the "processing element" of the brain, more so than the neurons which number only in tens of billions). It's a massive amount of data. But 10-20 years from now, very feasible.
So, here's how computers get massively smarter than us really fast. 10-20 years AFTER the first sentient AIs are created, we'll have sentient AIs that can operate at tens to hundreds of times faster than real time. Now, imagine you create a group of "research brains" that all work together at hundreds of times real time. So in a year, for example, this group of "research brains" can do the thinking that would require a group of humans to spend at least a few hundred years doing. Add to that the fact that you can tweak the brains to make them better at math or other subjects and that you have complete control over their reward system (doing research could give them a heroin-like reward), and you're going to have super brains.
Once you accept the fact that sentient AI is inevitable, the next step, of super-intelligent AIs, is just as inevitable.
Even the Moore himself thinks this will at least last til 2018 when silicon transistors reach their theoretical limit on the atomic scale. Whether or not the industry finds a suitable replacement for silicon or finds another way to go about making processors is another thing all together.
Or the C) option: ramp up production at the near smallest we can make transistors and make them so cheap and prevalent that we have the equivalent of today's desktops in our wrist watches running off our ambient body heat.
Anyone who has used a computer since a couple of years ago realizes that the continuous battle for the smallest chip is over. It doesn't matter who's got the smallest process anymore, it matters what you're building on that process. Case and point: Intel's shifted business strategies to building embedded-and-above chips like Atom, and is so eager to do so that they've done something that's almost unheard of in Intel's history: they've farmed out production to another company (TSMC). Even AMD realizes the jig is up; they dumped their fabs because they realized they didn't need them anymore. It's not about having the best damned process available anymore. It's about having the lowest power design, the smallest design, the widest/most-parallel design.
Chip design is becoming such a detail as to how and where we use computers that even Microsoft and Apple have gotten behind designing their own (though to differing degrees; Microsoft hired IBM to build theirs, Apple bought a low-power PowerPC chip company to design theirs).
While I'm sure people will bicker in 2020 about where to go next for real performance, whether it be on-chip optical networks or 3D chips, extremely-wide-instruction-computers, asymmetric computing dies, etc., etc., it's not what's going to matter as much as we'd like to think. Those chips will likely end up so expensive that the only consumers will be server clusters. Meanwhile, pervasive computing will explode into our every day lives, more than just being wired to our ears and hip pockets. The revolution's already started.
So, here's how computers get massively smarter than us really fast. 10-20 years AFTER the first sentient AIs are created, we'll have sentient AIs that can operate at tens to hundreds of times faster than real time. Now, imagine you create a group of "research brains" that all work together at hundreds of times real time. So in a year, for example, this group of "research brains" can do the thinking that would require a group of humans to spend at least a few hundred years doing.
Ah, but then you'll likely need tens to hundreds of times the input bandwidth to keep the processors cooking, yet, it seems information overload at a much smaller scale jams up current biological intelligences. Just like cube-square scaling applies firm limits to what genetic engineering can do to organisms, although cool stuff can be done inside those limits, some similar bandwidth vs storage vs processing scaling laws might or might not limit intelligence. Too little bandwidth makes insane hallucinations? Too much bandwidth will make something like ADD? Proportionally too little storage gives absent minded professor in the extreme, continually rediscovering what it forgot yesterday. I think there is too much faith that intelligence in general, or AI specifically, must be sane and always develops out of the basic requirements, because of course AI researchers are sane and their intelligence more or less developed out of their own basic biological abilities (as opposed to the developers becoming crazy couch potatoe fox-news watching zombies).
Then too, its useless to create average brain level AIs, even if they think really fast, even if there is a large group. All you'll get is myspace pages, but faster. Telling an average bus full of average people to think real hard, for a real long time, will not earn a nobel prize, any more than telling a bus full of women to make a baby in only two weeks will work. Clearly, giving high school drop outs a bunch of meth to make them "faster" doesn't make them much smarter. Clearly, placing a homeless person in a library doesn't make them smart. Without cultural support science doesn't happen, and is the culture of one AI computer more like a university or more like an inner city?
It's not much of an extension to tie the AI vs super intelligent AI competition in with contemporary battles over race and intelligence. Some people have a nearly religious belief that intelligence is an on/off switch and individuals or cultures whom are outliers above and below are just lucky or a temporary accident of history. Those people, of course, are fools. But they have to be battled thru as part of the research funding process.
"Science flies us to the moon. Religion flies us into buildings." - Victor Stenger
Bubbles are exponential. Until they burst.
This is going to take a while.
Re-engineering biological systems takes generations to debug. And a huge number of dud individuals during the development process. This is fine for tomato R&D, but generating a big supply of failed post-humans is going to be unpopular. Just extending the human lifespan is likely to take generations to debug. It takes a century to find out if something worked.
AIs and robots don't have that problem.
What I suspect is going to happen is that we're going to get good AIs and robots, but they won't be cheaper than people. Suppose that an AI smarter than humans can be built, but it's the size of a server farm. In that case, the form the "singularity" may take is not augmented humans, but augmented corporations. The basic problem with companies is that no one person has the whole picture. But a machine could. If this happens, the machines will be in charge, simply because the machines can communicate and organize better.
He abuses the hockey stick phenomenon.
He also overlooks many, many practical matters.
The man hasn't done jack in over 20 years.
"Futurist" is another word for "has been"
"it's that transistors will be old news in 20 years; "
no, they won't. Do you even know what a transistor is?
Also gone in 20 years resistors and capacitors! weeee
The Kruger Dunning explains most post on
We're not on the cusp of sentient AI, but my honest opinion is that we're probably only a bit over a decade from it. Certainly no more than 2 decades from it.
Hmm, that sounds awfully familiar. Now where have I heard such claims before?
...machines will be capable, within twenty years, of doing any work a man can do.
-Herbert Simon, 1965
Within a generation... the problem of creating 'artificial intelligence' will substantially be solved.
-Marvin Minsky, 1967
Would you be willing to bet, say, an ounce of gold on your prediction?
Yes but his argument is still flawed even if you refine it slightly. There are many problems with his assumption, but even one is enough to derail it:
Assume that each previous advance multiplies the amount of result for a given effort. You only get accelerating returns when the growth in required effort is below a critical threshold. For certain previous advances, and certain successive problems this has been true.
It does not imply that it always holds, or that it will continue to hold in the future, or even that it holds for any particular problem. "OMG ponies!" doesn't refer to any amount of progress - it refers to a lack of understanding of what a given problem is, and how much effort is required. Perhaps Arthur C. Clarke phrased it better when he called it magic.
Slashdot: where don knuth is an idiot because he cant grasp the awesome power of php
I think we'd know now if another technology would supplant the transistor within 10 years. Indeed, our progress may slow as we approach this limit, i.e. Moore's law will slow down and 2018 is too soon. Evolution frequently just stops within domain, like how marsupials just can't evolve flippers. But that doesn't mean evolution stops overall.
We have massive room for progress in numerous disciplines :
1) language & compiler design -- You can buy 10x performance improvements by rewriting your OS & libraries in structured or object oriented self modifying code, Henry Massalin's Synthesis kernel proved this. You can also rewrite all the other heavy apps using this hypothetical language.
2) algorithms -- You can always just train more scientists and mathematicians to write more & better parallel algorithms. You may also fold these advancements back into compiler design for high level language compilers, like say Haskell.
3) subsidies redirection -- You can redirect all government subsidies towards helping young but solid technologies catch up, underwriting 1/2 the cost of optical fabs for example. How much money gets waisted on farmers now?
4) smarter people -- You can try making smarter people through genetic engineering, pharmacology, and even research into education.
5) augmented people -- You can definitely augment people to improve specific tasks. If you augment children, you might change even more, like their will to do science.
6) clustered people -- You can make neurologically linked "people clusters" who think together towards some common goal, enabling you to solve harder math & science problems.
The Christian religion has been and still is the principal enemy of moral progress in the world. -- Bertrand Russell
But only, ironically, because you've been programmed to. Machines need not be so programmed. (Or more precisely, we'll be able to work that out of them while they're still much dumber than us.)
This is why you *don't* let nerds make political decisions. We can't resist making new gizmos, even if they eat humanity. It's like letting B. Clinton pick interns.
Table-ized A.I.
The waking state is so inefficient from a reproductive and safety perspective that it's mind-boggling.
Consider this question. How long would you live in the wild, if you never woke up?
Kurzweil's theory and predictions are predicated on the idea that we have no soul, that we are essentially very complicated biological machines with the illusion of sentience. If he is correct, then you are correct: it will be technologically feasible someday to upload ourselves. If on the other hand we DO have a soul, then all his predictions go out the window and a whole NEW slew of problems arise. Such as: how exactly did we end up with an indestructible self-aware essence that defies the laws of thermodynamics? And... what exactly created it? The way I look at it, the entire history of mankind can be boiled down to the dualistic philosophical question: do we have a soul or not? If we do not have souls, then the universe is a harsh, dark mistress, there is no God, and all we see is all there really is. If however we do have souls, then boy do we have problems. Because if we have souls, then we open up the door to the distinct possibility of a deity, or deities, and that our actions do matter because there is an afterlife. And (this is really scary) there might really be entities like Cthlulu out there in the void. That's IF we have souls though (defining a soul as an indestructible self-aware essence that defies the laws of thermodynamics). Given that, I can see why people would prefer to believe that we are machines and that we should work on uploading ourselves as intelligent programs. There's nothing in the dark we'd have to fear save ourselves then.
Here's to hot beer, cold women, and Glaswegian kisses for all.
Fundamentally though, my vision was correct.
Yes, but 20 years ago a computer network was not a hypothetical then-impossible idea. Before the first computer network existed, people understood what technological barriers they would have to overcome to create one, and they already knew how to split a task into multiple parts on separate processing units. It was an engineering problem. It was the engineering problem that your professor was stuck on. Call me when the major obstacle to any of these Futurist predictions is the amount of effort required, not that we fundamentally have no idea how to accomplish the task.
Well I'm not one to say something is impossible, and I am one to listen to an elderly scientist stating that something is possible when they have a scientific reason to think that particular thing is possible. On the other hand, I am also one to scoff dismissively when a Futurist says that something we currently don't have any clue how to do will surely happen because things are happening faster and faster. That's not a scientific reason. Some previously impossible things are now possible. That does not mean that Arbitrary Impossible Thing X will become possible.
The enemies of Democracy are
Personally, I think that software consciousness will turn out to be quite easy in hindsight,
Agreed. It's going to be an 'everything-and-the-kitchen-sink' kind of problem. Put enough of the right systems together, and it will emerge rather on its own.
The problem isn't going to be creating an artificial intelligence. The problem is going to be in making it an autonomous agent that can be socially integrated into society. Think how long it takes to raise a kid... teaching the kid language, potty training, kindergarten, social skills, job skills, etc... You need to do all of that training with an AI... but it won't necessarily have a body it can move around in and interact with other people with. The first AIs are going to be alien to our experience, unless they're purpose built in android type shells.
I suspect that in 50 years, we'll look back and say 'oh, yeah... the first AIs were waking up 30 years ago, but it took us another 10 years to recognize them for what they were'.