Cookies   I display ads to cover the expenses. See the privacy policy for more information. You can keep or reject the ads.

Video thumbnail
Okay so - Hi, I'm Scott Aaronson. I'm a computer science professor at the
University of Texas at Austin and my main interest is the capabilities and
limits of quantum computers, and more broadly what computer science and
physics have to tell each other. And I got interested in it I guess because it
was hard not to be - because you know as a teenager it just seemed clear to me that
the universe is a giant video game and it just obeys certain rules, and so if I
really wanted to understand the universe you know maybe I could ignore the
details of physics and just think about computation. But then with the birth of
quantum computing and the dramatic discoveries in the mid-1990s
(like Shor's algorithm for factoring huge numbers) it became clear that physics
actually changes the basic rules of computation. So that was something that I
felt like I had to understand and 20 years later we're still trying to
understand it and we may also be able to build some devices that can outperform
classical computers you know namely quantum computers and use them to do
some interesting things. But to me that's that's really just icing on the cake
really I just want to understand how things fit together. Well to tell you the
truth when I first heard about quantum computing (I think from reading some
popular article in the mid 90s about a Shor's algorithm which had only recently
been discovered) you know my first reaction was this sounds like obvious
hogwash - you know this sounds like some physicists who just do not understand
the first thing about computation - and they're just you know inventing some
physics proposal that sounds like it just tries every possible solution in
parallel. But you know none of these things are going to scale right and in
computer science you know there's been decades of experience
of that; of people saying well why don't you build a computer you know using a
bunch of mirrors, or using soap bubbles, or using a folding proteins right.
And there's all kinds of ideas that on paper look like they could you know evaluate
an exponential number of solutions at only a linear amount of time, but they're
always kind of idealizing something right so it's always when you examine
them carefully enough you find that the amount of energy or you know scales
explose up on you exponentially, or the precision with which you would need to
measure you know becomes exponentially precise, or something becomes totally
unrealistic - and I thought the same must be true of quantum computing. But you
know in order to be sure I had to read something about it so I while I was
working over a summer at Bell Labs doing work that had nothing to do with quantum
computing, well my boss was nice enough to let me spend some time learning about
reading up on the basics of quantum computing - and that was really a
revelation for me because I accepted [that] you know quantum mechanics is this real
thing. It is a thing of comparable enormity to the you know the print basic
principles of computation you can say you know the principles of Turing - and
you know it is it is exactly the kind of thing that could modify some of those
principles. But the biggest surprise of all I think was that I despite not being
a physicist you know not not you know having any skill that partial
differential equations or the others tools of the physicists that I could
actually understand something about quantum mechanics so I like to say that
quantum mechanics is much much simpler than most people imagined it could be
after you take the physics out of it - it's a certain generalization of the
rules of probability. And ultimately you know to learn the basic rules of how a
quantum computer would mark and start thinking about you know what they would
be good for - quantum algorithms and things like that -
it's enough to be conversant with vectors and matrices right? So you need
to know a little bit of math but but not that much right? You need to be able to
know linear algebra okay and that's about it. And I feel like this is a kind
of a secret that gets buried in almost all the popular articles; they make it
sound like quantum mechanics is just this endless profusion of
counterintuitive things right? That it's: particles can be in two places at once,
and a cat can be both dead and alive until you look at it you know, and then
you know why is that not just a fancy way of saying well either the cat's
alive or dead and you don't know which one until you look well they they never
quite explained that part right, and particles can have spooky action at a
distance and affect each other you know instantaneously, and particles can tunnel
through walls! You know it all sounds like hopelessly obscure and like you
know there's no hope for anyone who's not a PhD in physics to understand any
of it. But the truth of the matter is you know there's this one counterintuitive
hump that you have to get over which is the certain change to or generalization
of the rules of probability - and once you've gotten that then all the other
things are just different ways of talking about or different
manifestations of that one change right? And you know a quantum computer in
particular is just a computer that tries to take advantage of this one change to
the rules of probability that the physicists discovered in the 1920s was
needed to account for our world. And so that was really a revelation for me that
well you know even you know you're computer scientists are math people
people who are not physicists can actually learn this and start
contributing to it - yeah! [Adam]: So it's interesting like often and when you
try to pursue an idea, the practical gets in the way. Have you noticed that some people try and get
to the ideal without actually considering the practical - and you know
they feel like enemies - is it like that you should be letting the ideal be the
enemy of the practical? [Scott]: Well I think that you know there's
- from the very beginning it was clear that there is a theoretical
branch of quantum computing which is where you just assume you have as many
of these quantum bits (qubits) as you could possibly need, and they're perfect
they stay perfectly isolated from their environment, and you can do you know
whatever local operations on them you might like, and then you know you just
study well how many operations would you need to factor a number? Or you know
solve some other problem of practical importance? And you know and the
theoretical branch is really you know the branch where I started out in this
field and where I've mostly been ever since you know. And then there's the
practical branch which asks well what will it take to actually build a device
that instantiates this theory right - where we have to have qubits you know
that are actually the energy levels of an electron, or the spin States of an
atomic nucleus, or are otherwise somehow you know instantiated in the physical
world. And you know they will be noisy, they will be interacting with their
environment - we will have to take heroic efforts to keep them sufficiently
isolated from their environments you know - which is needed in order to
maintain their superposition state right? How do we do that? Well we're gonna need
some kind of fancy error correcting codes to do that right, and then you know
there are there are theoretical questions there as well but how do you
design those our correcting codes right? But there's also practical questions: how
do you engineer a system you know where the error rates are low enough that
these codes can even be used at all; that they won't you know that if you try to
apply them you won't simply be creating even more error than you're fixing. So
you know and what should be the physical basis for qubits? Should it be
superconducting coils? Should it be ions trapped in a magnetic field?
Should it be photons? Should it be some new topological
state of matter? Actually all four of those proposals and many others are all
being pursued right now! So I would say that until fairly until quite
recently in the field, like five years ago or so, the theoretical and the
practical branches we're pretty disjoint from each other right;
they were never enemies so to speak. I mean we might poke fun at each other you
know sometimes but you know I mean that you know we were we were never enemies I
mean you know the the field always sort of you know rose or fell you know as a
whole and we all knew that. But we just didn't have a whole lot to
scientifically to say to each other because you know the experimentalists
we're just trying to get one or two qubits to work well right, and they
couldn't even do that that much, and we theorists we're
thinking about well suppose you've got a billion cubits, or you know some
arbitrary number, what could you do? And even you know what would still be hard
to do even then right? A lot of my work was has actually been about the
limitations of quantum computers, but you know I also like to say the study of
what you can't do even with computers that you don't have. And only
recently the experimentalists have finally gotten the qubits to work pretty
well in isolation so that now it finally makes sense to start to scale things up -
not yet to a million qubits but maybe 50 qubits, maybe to 60, maybe to a
hundred. This as it happens is what Google and IBM and Intel and a bunch of
startup companies are trying to do right now. And some of them are hoping to have
devices within the next year or two, that you know might or might not do anything
useful but if all goes well we hope will at least be able to do something
interesting - in the sense of something that would be challenging for a
classical computer to simulate, and that at least proves the point that we can
you know do something this way that is you know beyond what classical computers
can do. And so as a result you know the most
nitty-gritty experimentalists are now actually talking to us theorists because
now they need to know not just as a matter of intellectual curiosity but as
a fairly pressing practical matter, well you know once we get 50 or 100 cubits
working what do we do with them? What do we do with them first of all that you
know is hard to simulate classically how sure are you that there's no fast
classical method to do the same thing, you know how do we verify that we've
really done it you know, and is it useful for anything right? And you know and
ideally they would like us to come up with proposals that actually fit the
constraints of the hardware that they're building right, where you could say you
know eventually none of this should matter right, eventually a quantum
programmer should be able to pay as little attention to the hardware as a
classical programmer has to worry about the details of the transistors today
right. But in the near future when we only have 50 or 100 cubits you know each
and every you know you're gonna have to make the maximum use of each and every
qubit that you've got, and the actual details of the hardware are going to
matter, and the result is that even you know we theorists have had to learn
about these details you know in a way that we didn't before. And so there's
been a sort of coming together of the theory and practical branches of the
field just in the last few years that to me has been pretty exciting. [Adam]: So you think we will have something equivalent to functional programming for quantum computing in the near future? [Scott]: Well there
actually has been a fair amount of work on the design of quantum programming
languages. There's actually a bunch of them out there now that you can download
and try out if you'd like there's one called Quipper, there's another one
called a Q# from Microsoft, there's you know and there are
several others. You know of course we don't yet have very good hardware to run
the programs on yet, mostly you can just run them in
classical simulation, which you know naturally only works well for up to
about 30 or 40 cubits, and then it becomes too slow. But if
you would like to get some experience with quantum programming you can try
these things out today, and many of them do try to provide higher level
functionalities, so that you're not just doing the quantum analog of
assembly language programming right, but you can think in higher-level modules, or
you can program functionally. I would say that in quantum algorithms you know
we're you know I mean we're you know mostly we've just been doing theory and
we haven't been implementing anything, but we have had to learn to think that
way right. If we had to think in terms of each individual qubit
each individual operation on one or two qubits well we would never get very far
right? And so we have to think in higher-level terms like there are
certain modules that we know can be done - like one of them is called the Quantum
Fourier Transform and that's actually the heart of Shor's famous algorithm for
factoring numbers (it has other applications as well). Another one is
called Amplitude Amplification that's the heart of Grover's famous algorithm
for searching long long lists of numbers in about the square root of the number
of steps that you would need classically, and that's also like a quantum algorithm
design primitive that we can just kind of plug in as a black box and it has
many applications. So we do think in these higher level terms, but you know
there's a different set of higher level abstractions than there would be for
classical computing - and so you have to you know you have to learn those. But the
basic idea of decomposing a complicated problem by breaking it down into its sub
components that's exactly the same in quantum computing as it is in classical
computing. [Adam]: Are you optimistic with regards to quantum computing in the medium to short term?
[Scott]: You're asking what am i optimistic about - so I am I mean like I feel like the the field
has made amazing progress: both on theory side and on the experimental side.
We're not there yet, but we know a lot more than we did a
decade ago right? Some of the what were my favorite open
problems as a theorist you know a decade ago have now been resolved - some of them
within the last year - actually and the you know the hardware you know the
qubits are not yet good enough to build a scalable quantum computer right - in
that sense you know the skeptics can you know clearly legitimately say well you
know well you know we're not there yet - well you know no duh we're not - okay
but: if you look at the coherence times of the qubits you look at what you can
do with them, and you compare that to where they were 10 years ago or 20 years
ago - you know there's been orders of magnitude type of progress. Right, so the
analogy that I like to make: you know Charles Babbage laid down the basic
principles of classical computing in the 1820s right? I mean not as not with as
much mathematical rigor as Turing would do later, but you know the basic ideas
were there. You know he had what today we would call a design for a universal
computer. So now imagine someone then saying 'well you know so when is this
analytical engine gonna get built? you know will it be in the 1830s or will it
take all the way until the 1840s?' Right well you know in this case you know it
took more than a hundred years for a technology to be invented - namely the
transistor - that really you know fully realized Babbage's vision. Right I mean
the vacuum tube came along earlier, and you could say partially realized that
but you know it was just not reliable enough to really be
scalable in the way that the transistor was, And we are now in
the (optimistically we're in the) very very early vacuum tube era of
quantum computing ok? We don't yet have the quantum computing analog of the
transistor as people don't even agree about which technology is the right one
to scale up yet. Is it superconducting? Is it trapped ion?Iis it photonics? is it a
topological matter? Right so you know so they're pursuing all these different
approaches in parallel. You know the partisans of each approach
have what sounds like compelling arguments why none of the other
approaches could possibly scale. I hope I hope that they're not all correct uh-huh
but you know but people are still just you know they've only just recently
gotten to the stage where one or two qubits work well in isolation, and where
it makes sense to try to scale up to 50 or 100 of them and see if you can get
them working well together you know at that kind of scale. And so so I think the
the big thing to watch for in the next five to ten years is what's been saddled
with the somewhat unfortunate name of Quantum Supremacy (and you know this was
coined before Trump I hasten to say). But so this is just a term to refer to
you know doing something with a quantum computer it's not necessarily useful but
that at least is classically hard. So you know that as I was saying earlier proves
the point that you can do something that would take a lot longer to simulate it
with a classical computer. And you know this is the thing that Google and some
others are going to take their best shot at within the next couple of years so
you know and and what puts that in the realm of possibility is that
you know just a mere 50 or 100 cubits you know if they work well enough should
already be enough to get us this. Right and you know you you in principle you
know you may be able to do this without needing error correction - once you need
error correction then that enormously multiplies the resources that you need
to do even the simplest what's called Fault-Tolerant Computing
might take many thousands of physical qubits, okay, even though you know
everyone agrees that ultimately if you want to scale to you know the realize
the true promise of quantum computing, or let's say to threaten our existing
methods of cryptography - then you're going to need this fault tolerance. But
that I expect we're not gonna see in the next five to ten years. If we do see
it I mean that will be a huge shock I mean but you know - as big a shock as it
would be if you told someone in 1939 that you know there was going to be a
nuclear weapon and in six years okay. You know in that case there was a world war
that sort of accelerated the timeline you could say from what it would
otherwise be you know. In this case I why I hope there won't be a world war that
accelerates this timeline. But you know my guess would
be that you know if all goes well then quantum supremacy might be achievable
within the next decade, and I hope that after that we could start to see some
initial applications of quantum computing which will probably be some
very very specialized ones; some things that we can already get with you know a
hundred or so non-error-corrected qubits okay. And you know by necessity these are
going to be very special things - they might mostly be physics simulations or
you know simulations of some simple chemistry problems. I actually
have a proposed application for near-term quantum computers which is to
generate cryptographically secure random numbers - those random numbers that you
could prove to a skeptic really were generated randomly - turns out that even
like a 50 or 60 qubit quantum computer should already be enough to give us that.
But you know true scalable quantum computing the
kind that could threaten cryptography and that could also speed up you know
optimization problems and things like that - that will probably require our
correction - and you know I could be pleasantly surprised you know. I'm not
optimistic about that part becoming real on the next five to ten years, but you
know I'm I you know since every everyone likes an optimist I guess I'll you know
I try to be optimistic that well you know we will you know take big steps in
that direction and maybe even get there within my lifetime.