Google announces it's nearing quantum supremacy

Sorry but I won’t be satisfied until we have self aware AI improving itself.

barack black panther just isn’t impressive for me dude!

the real future of AI is gonna be dystopian but in the dumbest ways possible.

we need ai that can make convincing watmm simulations

We already have AI that has created music WATMM thought sounded like a new emptyset album, so we’re at least part way there.

  On 10/26/2019 at 8:50 AM, sheatheman said:

Sorry but I won’t be satisfied until we have self aware AI improving itself.

barack black panther just isn’t impressive for me dude!

the real future of AI is gonna be dystopian but in the dumbest ways possible.

Expand  

i predict that one day you'll wake up in sweat realising your own ideas were blocking you to see the truth and that you missed the boat

  On 10/26/2019 at 10:49 AM, goDel said:

i predict that one day you'll wake up in sweat realising your own ideas were blocking you to see the truth and that you missed the boat

Expand  

missed what boat?

  On 10/26/2019 at 6:58 AM, goDel said:

Ehm, people have been modelling stock markets and the weather for years. As long as science has existed, id argue. That's what science is all about. Climate science? Models! Cosmology? Models!

As long as you disregard the expected quality of those models, that is.

Expand  

That's kind of a big problem though, weather models are never going to be solved by classical computation, they will always fail eventually. The same is probably true for economic modelling too, though this hasn't been completely proven as of yet. Quantum Computers will not solve either of these problems. The best they could do in these situations is improve the efficiency of the computation, but all that would mean is they would fail faster.

Google: we made the best quantum computer ever, why don’t you?

watmm: read the rules!

  On 10/26/2019 at 12:00 PM, caze said:

That's kind of a big problem though, weather models are never going to be solved by classical computation, they will always fail eventually. The same is probably true for economic modelling too, though this hasn't been completely proven as of yet. Quantum Computers will not solve either of these problems. The best they could do in these situations is improve the efficiency of the computation, but all that would mean is they would fail faster.

Expand  

the 'solvability' of weather models has nothing to do with computability. and i disagree with this idea of solvability to begin with. or rather, you need to explain what you mean with solvability.

reason is that it's about chaotic systems. this means they're provably unpredictable even if you have the "right" model. meaning, you have modelled all causal relationships relevant to what is being modelled. and you basically completely figured out the science behind some real world phenomena (like weather). or in other words, you've solved the science and modelled a phenomenon to a tee, but you still end up with unpredictability in the long run.

not saying we've completely figured out weather. but we're currently able to define perfectly understood chaotic systems and still end up with unpredictability. that's inherent to chaotic systems and has nothing to do with computational power or having enough scientific knowledge about what is going on.

quantum computers indeed will do nothing to change this. 

and so we're back at the original point: what can you reasonably expect from models. my point was/is that it's unreasonable to expect 100% accuracy all the time. similar argument holds for AI btw

    Quote

the 'solvability' of weather models has nothing to do with computability. 

Expand  

not quite. there are two separate problems here: computational complexity (in the P-NP sense), and chaotic nonlinear determinism. if a problem can't be solved in polynomial time, then a quantum algorithm isn't going to be any help. and you're right that if a system is chaotic it will also be no help to use a quantum computer either. there are certain ideal models of complex systems which can be solved in polynomial time, so that's not a problem, but that still doesn't solve the chaos issue, so if the model is chaotic, which weather and economic models are, then a quantum computer won't help anyway. but such models are probably also not correct, and the correct versions are probably not solvable in polynomial time. so either way it doesn't work. there's also the issue that the first question is a problem in discrete mathematics, and the latter usually deals with real numbers (though there are chaotic systems which use integers or even booleans, like cellular automata). it could be that both things are actually hinting at the same fundamental problem (in fact it looks like there are some mathematicians who have tried to look at P/NP from a dynamical point of view).

  On 10/26/2019 at 1:00 PM, goDel said:

my point was/is that it's unreasonable to expect 100% accuracy all the time. similar argument holds for AI btw

Expand  

it's not just that you can't expect 100% accuracy, it's that as time increases you can be certain of 100% inaccuracy eventually.

  On 10/26/2019 at 1:56 PM, caze said:

not quite. there are two separate problems here: computational complexity (in the P-NP sense), and chaotic nonlinear determinism. if a problem can't be solved in polynomial time, then a quantum algorithm isn't going to be any help. and you're right that if a system is chaotic it will also be no help to use a quantum computer either. there are certain ideal models of complex systems which can be solved in polynomial time, so that's not a problem, but that still doesn't solve the chaos issue, so if the model is chaotic, which weather and economic models are, then a quantum computer won't help anyway. but such models are probably also not correct, and the correct versions are probably not solvable in polynomial time. so either way it doesn't work. there's also the issue that the first question is a problem in discrete mathematics, and the latter usually deals with real numbers (though there are chaotic systems which use integers or even booleans, like cellular automata). it could be that both things are actually hinting at the same fundamental problem (in fact it looks like there are some mathematicians who have tried to look at P/NP from a dynamical point of view).

Expand  

i think we're going off the deep end here, but why the hell not..

one thing: we currently don't know what the relationship is between the classical "NP" problems and the problems a quantum computer can solve in polynomial time with a certain degree of certainty.

so your argument about polynomial time is currently not something which is supported by some proof as far as i can tell. unless you assume a problem isn't solvable in polynomial time for quantum computers. haven't read up on it, so i might be off. it's been a while.  but my guess is, there currently is no general rule to prove some problem is non-polynomial for quantum computers. unless you introduce a problem which is provably unsolvable. and i'd argue, a chaotic system is not (or should not) be part of the set of unsolvable problems. it would surprise me if complexity theory currently suggests otherwise.

as far as i'm concerned, at this point in time we don't know what quantum computers can do in the space of (current) NP. with respect to chaotic systems this is irrelevant however, as these carry unpredictability regardless of the amount computational power the model requires. liek you also argued (and i already did in my post).

so i think we're just pissing in the same pond here. academic bickering on unimportant details which - to me seem - irrelevant to the original point.

  On 10/26/2019 at 10:49 AM, goDel said:

i predict that one day you'll wake up in sweat realising your own ideas were blocking you to see the truth and that you missed the boat

Expand  

Can’t wake up if I never sleep

*taps head with finger*

seriously though what a dumb comment. The truth about AI is capitalists are developing it in order to:

employ a fraction of the workforce

make each generation incrementally more reliant on AI

basically zuck consumer society in ways no one ever dreamed it could be zucked

...

so I will happily miss the boat, because the boat is heading towards a giant whirlpool

haha sure. that boat heading towards a giant whirlpool is a different boat though. that's planet earth. you can't miss that one. even if you want to

you're mistaking capitalists are the only people developing AI btw. but please continue believing your *woke* nonsense.

  On 10/26/2019 at 1:58 PM, caze said:

it's not just that you can't expect 100% accuracy, it's that as time increases you can be certain of 100% inaccuracy eventually.

Expand  

Wait ... what? That’s not how this works!

The way it *does* work is that 90% (or whatever) accurate turns out to be good enough in a *lot* of cases. As computing resources increase, there are more and more tasks we can perform automatically with this 90% accuracy and if quantum computing is ever really going to come to pass it’ll be orders of magnitudes more of processes you can run by doing a lot of linear algebra and statistics really fast (what is known as “AI”)

  On 10/26/2019 at 4:18 PM, rhmilo said:

Wait ... what? That’s not how this works!

The way it *does* work is that 90% (or whatever) accurate turns out to be good enough in a *lot* of cases. As computing resources increase, there are more and more tasks we can perform automatically with this 90% accuracy and if quantum computing is ever really going to come to pass it’ll be orders of magnitudes more of processes you can run by doing a lot of linear algebra and statistics really fast (what is known as “AI”)

Expand  

I assumed he was referring to the accuracy of models of nonlinear systems, so not AI, but weather prediction or resource allocation in an economy or something like that. In which case the errors blow up as time increases. It's true a quantum computer running a DNN could probably do it's vector/matrix computations faster (almost all linear maths), maybe that would help allow it do human level intelligence or better (though there's no good reason to think classical computers are incapable of this either), it's not going to allow it to predict the weather or do resource allocation reliably though, as I said, at best it would speed up the time it takes for the predictions to turn to shit, create a more efficient wrong-answer-producer.

What it's most likely going to be used for is cryptography, certain classes of maths things (but not others), simulations of quantum mechanical and chemical systems (with big impacts for biochemistry and materials physics).

  On 10/26/2019 at 3:41 PM, goDel said:

you're mistaking capitalists are the only people developing AI btw. but please continue believing your *woke* nonsense.

Expand  

yeah, the Chinese are spending more on AI research than anyone else.

  On 10/26/2019 at 6:49 PM, caze said:

yeah, the Chinese are spending more on AI research than anyone else.

Expand  

it's kind of splitting hairs though.. china is capitalist they're also totalitarian.  maybe it's their own form of capitalism but it's capitalism none the less. 

  On 10/26/2019 at 6:54 PM, ignatius said:

it's kind of splitting hairs though.. china is capitalist they're also totalitarian.  maybe it's their own form of capitalism but it's capitalism none the less. 

Expand  

it's very much not free market capitalism, this is not splitting hairs. it's not just that they're authoritarian on social issues and liberal on economic ones, they also have significant government control of their economy, and how their economy interacts with the world economy. they only allow a limited amount of free market operation internally. it's worked well for them, but it's not 'capitalism' in the normal sense of how everyone talks about it.

in other words: there's plenty development on the non-commercial side of things. (you can read between the lines that the usual use of capitalism is more or less equivalent to anything what commercial businesses do. and there's plenty of development outside commercial businesses when it comes to AI. do you really think that countries don't invest in their own AI technology? come on.... capitalism schmapitalism. this is about basic power of countries on the world stage. some knowledge they rather have in their own hands, as opposed to the hands of commercial parties)

edit: in other words 2 - it's not just china either

  On 10/26/2019 at 7:32 PM, goDel said:

edit: in other words 2 - it's not just china either

Expand  

Yup. I know for a fact the Dutch tax service is working on this (and running into some pretty spectacular SNAFU’s recently), I’ve heard of agricultural companies using AI (such as it is) to improve the circumstances of livestock and that’s just off the top of my head.

This whole “eating through large amounts of data in more or less automated fashion” is really booming.

And, to get back on topic, if you can make all that math happen on quantum computers you can do so much more of it, which is why Google and IBM are so interested in it.

  On 10/26/2019 at 3:26 PM, goDel said:

one thing: we currently don't know what the relationship is between the classical "NP" problems and the problems a quantum computer can solve in polynomial time with a certain degree of certainty.

so your argument about polynomial time is currently not something which is supported by some proof as far as i can tell. unless you assume a problem isn't solvable in polynomial time for quantum computers. haven't read up on it, so i might be off. it's been a while.  but my guess is, there currently is no general rule to prove some problem is non-polynomial for quantum computers. unless you introduce a problem which is provably unsolvable. and i'd argue, a chaotic system is not (or should not) be part of the set of unsolvable problems. it would surprise me if complexity theory currently suggests otherwise.

Expand  

reading my original post I see I said 'if a problem can't be solved in polynomial time', this is imprecise, what I meant was 'if a problem in a nonlinear system can't be solved in polynomial time', thought this would have been obvious from context, apologies if not. it's not been proven generally for all problems obviously, factorisation or quantum simulation show an exponential speedup, but it has been proven for some of the other algorithms that have been discovered (e.g. Grover's algorithm can't turn an NP-Complete problem into a P). and nothing has been discovered which can provide such speedups for the class of problems we've been talking about here. maybe it's imprecise to say we'll certainly never be able to get an exponential speedup for any of these problems, but it does appear that exponential speedups don't exist in general for nonlinear systems (which are also in non-polynomial time on a classical computer), it will have to be a limited class of them if at all, and also that it's only a limited set of linear problems where there are exponential speed ups too (mostly they are sqrt speedups or similar).

as to whether nonlinear systems are in the unsolvable part (whether classical or not), it depends on which models you're talking about. it's the 'spherical cows' problem. certain idealised versions might be solvable, e.g. in a nonlinear price calculation system of perfect efficiency (i.e. where the new prices can be perfectly calculated from all prior information), then you're fine, but the real world is empirically not perfectly efficient and far more complicated than our models. not sure whether this applies to weather modeling too, but I would presume so.

  On 10/26/2019 at 7:32 PM, goDel said:

in other words: there's plenty development on the non-commercial side of things. (you can read between the lines that the usual use of capitalism is more or less equivalent to anything what commercial businesses do. and there's plenty of development outside commercial businesses when it comes to AI. do you really think that countries don't invest in their own AI technology? come on.... capitalism schmapitalism. this is about basic power of countries on the world stage. some knowledge they rather have in their own hands, as opposed to the hands of commercial parties)

 

edit: in other words 2 - it's not just china either

Expand  

spending from the Chinese government dwarfs both other countries national and commercial spending, and commercial spending dwarfs non-Chinese national spending (e.g. Google and Microsoft outspend the US government). China is still a ways back though, they rely a lot on everyone else's software and hardware, so much of this spending is them trying to catch up. 

one last comment. regardless of the amount of money spent, it's other resources which might prove to be more important. like data, for instance. google might have the wildest ambitions when it comes to AI in the field of medical diagnostics, but if they don't have actual access to the data, they're basically teens dry humping the wall saying they're good at having sex. in the context of china, you could most definately argue that they have plenty data and cheap access to knowledge (chinese people/engineers work for less money). so, i'm not too keen on blindly following the amount of money.

and i'm not even go into the development of open source algorithms. which makes it even more cheaper. read: the algorithms themselves are relatively free. it all depends on access to data. and before we jump into discussions about the amounts of data company X, Y and Z have access to, I say sure, companies might have lots of data. but do they have access to the data needed to be able to this or that? it's mostly language and image stuff they're doing. and perhaps a couple of games that are being developed. but, come on, AI playing starcraft doesn't lead to world domination.

  On 10/26/2019 at 8:14 PM, goDel said:

they're basically teens dry humping the wall saying they're good at having sex

Expand  

Huhuhuh ...

But seriously: well put.

In the end Google and Facebook and all these other companies are merely playing around. 
 

  On 10/26/2019 at 8:14 PM, goDel said:

the algorithms themselves are relatively free. it all depends on access to data

Expand  

Again, well put. I’ve got students that are doing more relevant work than Google simply because they have significant data to work with and Google doesn’t. The simplest possible algorithm (it’s their first machine learning assignment!) is more than enough to extract meaningful insights from it.

most IDM 2019