I feel like there's a small minority of people who enjoy elevating kurzweil into a quasi-religious genius figure because of his speculations on the singularity, but he's really just an above average software engineer who happened upon some ideas which captured the public imagination. Easy to forget he's human too sometimes.
And there's also a small minority of people like me who think that Kurzweil is mostly a crackpot who gave the technological singularity a bad name because of his baseless claims on exponential acceleration of paradigm shifts.
We're also some who regard the whole concept of "the Singularity" as being a quasi-religious crackpot theory no matter how you slice it.
Mind you, it's a powerful meme. Religious tradition abounds with references to some future event that will forcibly clean up the mess and propel us back to Eden. Some of the most prominent -isms of the 20th century subscribed to similar ideas. I guess our brains are a willing host to that kind of stuff.
On top of that, the technological version of the end-of-history cult has the added attraction of whispering into our ears that maybe we .. as programmers and masters of the machine .. may just be able place ourselves on the right side of history and become immortal citizens in future tech utopia. A pantheon of gods, for all practical purposes. That's a seducing idea.
I don't come from a protestant country, "the end is nigh" does not really ring a bell in catholic countries.
The technological singularity, originally, comes from a very practical observation by Vernor Vinge, who was explaining that as a SF author, once humanity reaches the point where it can make intelligent machines that improve themselves, he hits a "horizon event" that he is not intellectually equipped to look beyond.
I have a hard time finding a good counterargument to it: the emergence of self-improving AIs of human or super-human capabilities will be an extremely strong and fast change in human progress. That much is true.
So yes, the hopes one can place in it look similar to the hopes of the religious eschatologists, but using this similarity to reject the argument is just an association fallacy.
Going a bit off topic here but your remarks on religion makes me curious. I'm from a firmly -- albeit lazily -- protestant country and don't describe myself as religious in any meaningful sense, but still an interesting subject.
My reference to religious "singularities" was to the apocalypse, or however you want to name it. The End of Days. The Second Coming. Stuff like that. I thought these ideas played a similar role in all branches of Christianity, although ignored by most people for most of the time. Is Catholicism and Protestantism so different in that regard?
More on topic, I see the evolution of (better) AI as a logical extension to the computer revolution, and I see the computer revolution as equivalent in scope to the industrial revolution. So I'm not saying it's a small deal. I just don't believe a whole lot in self-improving AI as a concept all that different from what has been going on in, say, processor design, for decades, and I certainly don't believe in the idea of a runaway, self-improving AI.
To become rapidly and exponentially more intelligent, it would have to either work smarter or work harder (to use an overused phrase). Working smarter would imply that if you arranged your CPU instructions in increasingly clever ways, you could extract exponentially rising performance from your programs. That's not how I remember the assembler programming that I did as a kid; you could be clever but there were some pretty hard limits. The strategy of working harder, on the other hand, would imply that someone would be willing to supply exponentially rising amounts of energy. Again, even if some Zuckerberg or some shady government agency would be willing to play this game, you'd quickly run into hard limits.
So yeah, AI is not the first game changing phenomenon in the history of humankind but it will change stuff. Like did bronze, railroads, or nukes.
About the religious thing, I think the catholic mindset is that the end of the world can't be predicted and anticipated. That you should worry less about it than the afterlife. Most protestant actually think the same way but protestantism is more diverse. If 9 out of 10 groups do not care about the apocalypse, you will not hear from them. You will hear from the JW-style preacher saying the end of the world comes next week.
Also, the notion of the rapture is pretty much a protestant interpretation. Most catholics see nothing to rejoice about in the apocalypse, even for good believers.
Back on the AI. Take this scenario: we make a software that's can be as smart as a human engineer, runs in 100% realtime speed on the fastest CPU we can produce. We know it is physically possible because our brain does that.
Buy all the CPU you can buy, compared to human researchers it is cheap. Make teams of hundreds of units (or whatever number is ideal). Put some team on the task of improving the parallelization of the AI program, some on the task of designing a faster CPU, some on the single-core optimization of the code, some on the theoretical research to improve the underlying algorithms. You'll quickly gain exponential results. Actually you will get a fast exponential rise of global intelligence.
The number of researchers will not show an exponential function but a steep step: almost overnight the number of minds working in R&D will change its order of magnitude. This is disruptive. All the currently researched and underfunded problems will be solved orders of magnitude faster. The small problems not worth solving nowadays will see teams of hundreds of top engineers devoted to them.
> The strategy of working harder, on the other hand, would imply that someone would be willing to supply exponentially rising amounts of energy. Again, even if some Zuckerberg or some shady government agency would be willing to play this game, you'd quickly run into hard limits.
Science fiction authors searched for that hard limit but do not see a roadblock until all of earth's matter has been converted to computronium and in a matrioshka brain, making us reach type II on the Kardashev scale. And I agree with them.
Yes, more energy will be required. So what? With enough research to improve generation and automation of construction and deployment of power plants, we can have an exponential rise in energy for quite a while. And if we mastered fusion, we could even go beyond type II civilizations.
> So yeah, AI is not the first game changing phenomenon in the history of humankind but it will change stuff. Like did bronze, railroads, or nukes.
I don't think humanity ever had an event of the same potential. Maybe the birth of civilization, when they started gathering in cities, but even these changes were pretty slow, and their effects happened at the time scale of human reproduction, that has not changed since millions of years. The changes we are talking about rely on our speed of CPU power production, which we improve constantly and that this process will accelerate.
Some aspect of the singularity is kind of undeniable. There's some information thing that's happening right now, and it's growing exponentially, and it just looks like it (for all intents and purposes) asymptotes vertical at some point, approximately 20 years from now.
What actually happens at that point, I think that's up for debate. Are we going to become immortal? Probably not. Are we going to have AIs who are better at everything than expert humans? Probably not.
But I think it's strange if someone denies that it's even happening. Just look at the amount of information that's flowing around, by any metric. It's exponential, and you can fit the curve and plot where it goes vertical. If we were talking about physical processes, I'd understand the skepticism. Maybe that exponential levels off. But I don't see why there is any physical limitation on how much information can flow. Or at least: if we hit that wall, things will be VERY different.
A minor point- but exponential curves never actually asymptote or go vertical, they just keep getting closer to vertical forever. But with a big enough level of growth it might as well be vertical.
True. Making any exponential curve look like a hockey stick is mostly a matter of picking the scale. I could probably create a chart showing how the population of the earth went vertical during the reign of the Caesars. Of course, a logarithmic plot solves that issue.
Exponential functions do not have singularities. There would have to be some additional property of the curve in order to conclude the presence of a singularity.
I agree that "the Singularity" is not a singularity in the mathematical sense of divergence, but it is a singularity in the sense of a singular (i.e. special or unique) event. (If it happens at all, that is.)
Look twenty years out and it appears vertical. But 10 years from now it will look much flatter, and what's then 20 years out (30 years from today) will look vertical.
The grass is always greener 20 years in the future.
As I understand it, "the Singularity" refers to the point in time where AGI becomes better at improving itself than the humans that created it. Then the existing exponential growth gets replaced by another exponential, but with stronger feedback.
It's the difference between a differential equation like x' = x and x' = 10x . In either case you can say "meh, if you normalize on today's value, the increase in 10 years is just a constant multiple", but the transition from one phase to the other will still be noticable, even though things won't "go vertical".
It seems much more likely to me that we're somewhere in the middle of an S-curve, before the limiting factors have kicked in. In terms of processing power, it's already happened to performance per thread, and it doesn't seem unreasonable to assume that there's a plateau to total throughput as well.
Even then, we're undoubtedly going to see some really amazing stuff come out of machine learning in the decades to come, just maybe not the beginning of Ragnarok.
An S-curve still goes "vertical" at some point. That's the idea: that some aspect of reality will be qualitatively weirder when information goes vertical-ish.
I kind of feel the same way. I read "Age of Spiritual Machines" 15 years ago and thought it was interesting. But now I don't believe most of what he says or predicts. Some of the stuff will come true, but we won't be Matrix-like entities. I also don't like how he downplays (or completely neglects) social backpressure to technology adoption.
Edit: His health regimen is also completely ridiculous.
> baseless claims on exponential acceleration of paradigm shifts
I'm not a fan of Kurzweil, but isn't this really one of his claims with the most data behind it? His presentations on exponential trends is always accompanied with endless citations of data points that went into it. He has (or had prior to Google?) a research team whose job was to find this data.
and see the kind of bogus things he puts together to fit his narrative of accelerating shifts. He puts no paradigm shift between the apparition of city states the invention of the printing press but he separates the inventions of computers and the invention of personal computers as two paradigm shifts.
I did not check all the sources he uses in the graph page 19 but I suspect the same bias: we ascribe more importance to events closer to us than to events further in the past.
In the context of information technology that's perhaps somewhat accurate though, right? City states centralized information into religious and political power structures. That pretty much stayed the same until the invention of the printing press. Maybe you could add the invention of double entry accounting and the banking industry it created, which allowed cooperation between disparate city states. But I'm not sure what else might be an epoch in between.
Gee, I don't know... Paper? Books? Libraries? Big empires with a uniform language? Roads networks? Trade routes? Accurate copying of books? Schools? University? Logic?
Can't we just get rid of the whole singularity term altogether and just realize it is standard exponential growth that people are marking points on a logarithmic graph as the "singularity"?
cmon now. This is a good attempt at pandering to the hackernews audience by being skeptical and cynical, but Quantum Supremacy is a big deal. From the same article:
If I have any policy advice, it’s this: recognize that a clear demonstration of quantum supremacy is at least as big a deal as (say) the discovery of the Higgs boson. After this scientific milestone is achieved, I predict that the whole discussion of commercial applications of quantum computing will shift to a new plane, much like the Manhattan Project shifted to a new plane after Fermi built his pile under the Chicago stadium in 1942.
the documentation here is quite sparse, so correct me if i'm wrong. This seems like a dressed up theorem proving system like PROLOG dressed up as something new.
No, not really. The wikipedia page of Maude has a good explanation of what it is like: https://en.wikipedia.org/wiki/Maude_system. You can also see the Maude book if you're interested [1].
I'm a bit out of my league here, but how does Maude compare to say Idris?
I remember dabbling with Pure, which used term rewriting, and lent itself to expressing mathematical equations in a familiar format, but it was not a prover.
I downloaded the PDF for Maude, and I will certainly take a look in the meantime. My interests are solely for learning, and I thought using something like Idris to work my way through Sussman's Structure and Interpretation of Classical Mechanics, might drive it home for me in ways that my physics classes didn't.
one big dimension here, the "implementation error" can be easily be debugged. Gradients can be checked numerically. The model can be checked to work by looking at the optimality conditions (not just the loss function go down). This shouldn't be an issue for anyone from a traditional coding background.
I don't think it was pretentious, and your analogy is quite off. You don't need to understand any of those fields to grok bubblesort (algorithmics is enough). You do need to know some statistics, calculus and algorithmics to understand stochastic gradient descent.
you don't need to understand statitstics. calculus (calculus is a high school subject and is hardly deep), and what is "algorithmics"? If you mean the study of any and all algorithms, then sure,
a rather awful essay not worth reading. This sounds like a rather large sum of money you're paying for few minutes of internet attention on your half baked ideas about the current american political system.
You may disagree with Peter Thiel's philosophy (as I do), but you cannot deny he is a thoughtful and intelligent man who has some interesting conjectures about the future which are (in my assessment) likely to be wrong. Furthermore Thiel himself hasn't personally made any investments (apart from the trump donation) which I find personally objectionable. If you had specific quarrels with his ethics, apart form his association with trump, it'd be a different story. Your essay betrays none of that.
quasi newton methods are not square in the dimension of the problem (think limited memory L-BFGS), and can be run in linear time. In my experience, however, they're 2-3 times slower than regular methods like ADAM.
The most infuriating argument against GMO's (2) might the "unknown unknowns" argument, because it sounds so reasonable. Hey, the just say the Jury's still out, so its best to be cautious yes?
But the truth be said, by not eating GMO's, you are too making a decision. And we don't know what we're doing eating good old fashioned food either. And the argument from historical data "we've been doing this for thousands of years, so it must be ok" is a pretty weak one. We've been smoking for over 7000 years now. We've filled our guts with carcinogens, practiced sham medicines and doing plenty of stupid things for a long time now. We did it not because it was "healthy", or out of some ancient wisdom, we did it because it was we knew, and we were making the best of a strange situation.
