So our plan today is to talk about what quantum
computers are. How people are building them. What they can do. What they can't do. They're not all powerful god like devices
so they do have limitations that we'll get into as well. OK. Without further ado let me introduce our esteemed
panelists. Our first panelist is a mathematician, computer
scientist, physicist, expert on quantum information theory. She was a professor at Tel Aviv University
and a researcher at the CNRS in Paris. Pease welcome Julia Kempe. Our next participant is the professor of quantum
mechanical engineering, that's a thing, at MIT. Director of the Keck center for extreme quantum
information theory and hopefully will explain why it's extreme to us. Please welcome Seth Lloyd. Next up, our next guest comes to us from the
Air Force Research Laboratory in Rome in upstate New York. She's a senior research scientist there and
the primary investigator for the Trapped-Ion Quantum Networking Group. Please welcome Kathy-Anne Soderbergh. And finally coming to us from IBM in Yorktown
Heights, just north of the city, is the Manager of the Experimental Quantum Computing Group,
distinguished research staff member. Please welcome Jerry Chow. So Seth, let me turn to you first. Quantum mechanics: weird or just unfamiliar? Definitely weird. It's, I prefer the word funky actually. OK you heard its. This is the official terminology. It's the James Brown of sciences. How so? I mean why and particularly in terms of computation
what is it? What's the special quality of quantum physics? So in quantum mechanics, things that we think
of being like particles, like basketballs or soccer balls, have waves that are attached
to them. And so you know I have a ball over here and
it's got a wave and then I have a ball over here and it's got a wave. But the funky thing about that is that the
waves can add up so I can have a ball that's both here and there at the same time. And if you map this to a bit in a quantum
computer, so this is zero, ball over here, and ball over there is one, then I can have
a quantum bit or qubit that is zero and one at the same time. So Julia, why would that be something you
would actually want? Why would this, I want to actually comment
on the weird first if I may. Funky please. Funky. Because I think it becomes a, it depends on
your point of view coming at it as a physicist with a lot of training in classical physics
it's indeed probably very weird, but if you look at it as a computer scientist it maybe
becomes less weird because we are not spoiled. Our intuition is still, that it's not, how
can I say, biased in any way. And I view it as something which is like probability
theory except the probabilities can be negative or they can be even complex that that is not
so essential. And so in that sense it's not very weird. It's just gets, requires a bit of getting
used to but it's pretty natural. And so for the qubit, why is it useful to
have qubits in being both zero and one at the same time? Cause if we have many of those, we can think
of having various states at the same time and we can think of computing all these possibilities
at the same time. So that leads us to this massive what's called
quantum parallelism. So Jerry, can you walk me through like just
a quick numeric example, you have a certain number of qubits, what that means in terms
of parallelism? Yeah. So I mean one of the, as has been mentioned
with regard to these qubits, you have these superpositions. And so you might have some number of bits
but instead of bits now you have qubits. Right. And in terms of this parallelism what you
can actually have is access to a much larger space of possibilities. So if you have n qubits you actually have,
using these principles of superposition and entanglement, you have access to a space of
up to two to the n possibilities. And so that type of, that type of exponential
space gets really, really large for rather modest numbers of qubits. In fact if you get to around n of three hundred
cubits you actually have some state-space that is greater than the total number of particles
in our universe. Kathy-Anne, walk me through an extremely simple
numerical example…suppose I, in fact we do have, but to take an example two of these
qubits. What does that mean in terms of the computational
space that we can work with? So if you have two qubits you have a four
qubit state space because you can have each one stored as a zero plus one and the other
stored as a zero plus one, so you get 00011011. Whereas classically you can only have zero
or one. You can't have anything in between. So the computer's in a sense an all possible
computational state space. Yes that's right. Now Seth, on the face of it that doesn't sound
like such a good idea because you want to get an answer from the computer and it's just
basically telling you everything. Yeah and indeed it's kind of dangerous. If I have a quantum bit that's zero and one
at the same time and I say yo are you zero or are you one? Well you know you could, OK, so the electron
is over here and I bring up a very sensitive electrometer that says, yo, are you here or
there? Well it's either going to show up here, say
with fifty percent probability or there were fifty percent probability. So it's just going to behave like something
that's generating a random number which is kind of useful but it's not, if you actually
want certainty for your answers that's not so great. So the kind of the way that these quantum
computations work is you set up all these waves and they're wiggling on top of each
other and they're, they're performing multiple computations simultaneously. So you can think of an individual wave, a
wave of say electron here and not there, that's kind of like a pure tone like ahh. An electron here and not there is like ahh. An electron here and there at the same time
is does somebody want to supply the other tone? It's a chord. So it's you, you get the computational power
from the interference from the kind of symphonic nature of this. So the idea of a quantum computation: you
set up all these waves, they make this beautiful music together in this symphonic way. But at the end of the day you actually want
to have an answer that says yes or no, or zero or one, and all the trickery and talent
goes into making that happen. So Julia, is that, any problem I care to pull
out of a hat or that my professors give me as a homework assignment. Can I turn it into a problem that's amenable
to this kind of chord pattern that then reduces to a single pure tone? That might depend on your artistry but in
general quantum computers are good at certain things and we would leave, you know, a lot
of other things to our normal classical computer. And of course you probably all heard of the
problem that a quantum computer can solve very well. And that's factoring numbers, large, very
large numbers into primes. And the way it's done, just like Seth was
saying and I was saying, you can think of these amplitudes of a quantum computer as
positive and negative numbers. We try to arrange these waves in a way that
all the bad answers cancel out because you can, you know ,you can have several ways to
arrive at the wrong answer. And what you'll try is have some with a positive
amplitude and some with a negative amplitude and they cancel out and then you arrive at
the right answer and you, voila you get a factor for a very big number. And that's one of the first and most remarkable
things that was discovered a quantum computer can do. And I can elaborate on why it's important. Most of you, all of you, I assume, are using
credit cards at the machine or over the Internet and you rely on the fact that they're encryption,
that they are encrypted. And it so happens that the modern day encryption
is based on assumptions of hardness of factoring or problems of that type. And it's these type of problems that a quantum
computer would be able to easily break. Interesting. Kathy, let me ask as you to drill down a little
bit more here. Sure So when I think of a computer I think of it
has, does arithmetic logical operations. How do you, I mean without going into details,
we'll get to some of the details later, but does the quantum computer also have those
same elements of it can add, it can subtract, it can do logical comparisons of things? Yes, yes you need all of the same similar
components to a classical computer to do quantum computing. But the way that you make the gates looks
very different than in a conventional computer. And that somewhat depends on the underlying
qubit technology. OK I'm going to definitely come back to the
details on that later. I just want to establish that it's a computer. Yes It has, you look at it and though it's configured
differently, does have the recognizable qualities of a computer. Yes You can program it in C or Java, and it has
one additional instruction which says take this quantum bit and put it here and there
at the same time, put it in the state of zero and one at the same time. So you supplemented ordinary computer language
with just one additional quantum instruction and you're good to go. If you can build the thing of course. Which we'll see in a little bit. Jerry, what are we up to just in general state
of the art about how many qubits we have and what does that mean in terms of vis a vis
a classical machine. Yeah. So there's many different physical implementations
of building a quantum computer. The underlying core of a quantum computer
we call a quantum processor. And what you basically need to build the quantum
processor is something that follows the laws of quantum mechanics and can have this quantum
mechanical zero and one. And in terms of where we are in experiments
we're looking at building universal based quantum computers of order of ten to twenty
qubits at the moment. And that's, that's kind of where the state
of the art is in the field. And just for reference sake, twenty bits doesn't
sound like a lot actually, but how does that what kind of power does that endow a machine
with? Well it's actually very interesting because
although you might have twenty qubits you can actually then have the state space up
to two to the twenty possibilities, right? But how much you're actually able to access
that then determines, is determined by your coherence time. So there's a metric known as coherence time
which says how good of a quantum state can you actually keep in your quantum processor. And different types of technologies varying
from superconducting to trapped-ions, like the kind Kathy-Anne works on, have different
amounts of coherence time. So overall this type of time times the number
of qubits, we try to we try to at least my colleagues that we've only started thinking
about a metric for this called quantum volume to kind of describe what is the power of a
quantum computer. So walk me through. What's a quantum volume? You can basically think of it as how many
steps of these logical operations, or these, these gate operations that you can do the
superposition or entanglement steps in the amount of time that before all the quantum
information is gone, becomes just classical. And so you have a certain number of steps
and then you have a certain amount of depth in terms of the total number of qubits that
are connected to one another. So Julia, let me ask this of you. If I just have a phone, a classical ordinary
computer and I store information in it, one hopes at least it will be able to retain that
for a long period of time. Is Jerry saying that actually it decays away? And why would it do that inside a quantum
computer? So the big challenge for a quantum computer
is indeed to maintain these coherences or these waves that are spread over, not just
you know one qubit but over a collection of, in this case, perhaps twenty qubits. And what we call entanglement these, these
correlations at a distance of these qubits. And it is true that this is what nature does
on a very small scale when we describe electrons and so on. But it's also true that we don't observe this
in our everyday life. I mean when you have a bit you have a big
it's either zero or one and the reason we don't observe it is that once we start interacting
with the environment, once this very fragile superposition is being subject to the surroundings,
it's being subjected to noise. And these very fragile superpositions will
start to what we call decohere, so just disappear. And the point is of course that we need to
be able to address this quantum computer. We need to be able to talk to it. We need to be able to manipulate it. So it has to be exposed to us, to our you
know to the world in that sense. And so we're living in this tradeoff situation
where on one hand we need to protect the state. So we would like to just put it in fridge
and never touch it. On the other hand we have to touch it in order
to manipulate it. And this is the big challenge that, you know,
these experimentalists are facing. To battle this decoherence invariably comes
along with the fact that we're exposed to the, you know, to the environment. Kathy-Anne, walking into your laboratory,
what would we see and then walk me through what that represents. Sure. So I'll talk about trapped-ion technology
which is what we work on and I believe Jerry will talk about superconducting qubits in
a few minutes. They're very different technologies but they're
both very advanced right now in the field. So for ions we track single atoms and we hold
them, they're they're charged so you can hold them using electric fields. So first we prepare a vacuum chamber, because
as you just heard these systems are very fragile, so you need to protect them from the environment. So these trapped-ions operate at room temperature
but we hold them in a vacuum chamber roughly ten to the minus twelve Torr - it's the same
vacuum as outer space. So the only thing in there is the atoms that
you want to manipulate. And you have a neutral atom source which is
just a piece of metal in a stainless steel oven that you heat up and then that creates
a beam of neutral atoms which you can put in your, what's called an ion trap, which
is just a collection of metal electrodes. Because as I said we trap these using electric
fields since they're charged. So you put an oscillating electric field on
that trap. What that looks like to the atom is a rotating
saddle. So then that looks like a bowl. And if you drop a marble in a bowl eventually
it'll come to rest at the bottom of the bowl. The trapped ions do the exact same thing and
these potentials. And so we shine the neutral atom beam near
the trap and then we have a laser that actually rips one of the electrons off the neutral
atom and that makes our ion. So that leaves it charged. And at the same time we have to shine a different
color of light in, because coming straight out of an oven the atoms are essentially screaming
hot and the trap potential just can't catch an atom it's going that fast. So you have to cool it down a little bit with
a laser, its laser cooling. And then that allows you to trap it in this
bowl like potential and then we shine yet another color of light on the atom and all
these different colors of light create different transitions within the atomic structure. So if you could look in the atom you would
see different energy levels inside and each color of light is resonant with a different
energy level. And so the detection light when we shine it
on. It hits a very strong transition in the atom,
which excites it from it its ground state, so the qubits themselves are held in the ground
states of these atoms. There's two ground states. And it sends it to an excited state. It's a very short lived state. And when it emits it emits a photon and it
does sense that this strong transition it does that hundreds of thousands of times and
then we collect those photons on a camera. The ion we're using is ytterbium so it emits
a UV. If it was a visible color you could actually
see it with your eye, it would be a tiny speck. Hang on. You can see atoms. You can see atoms. See atoms. What do they look like? Just like when you shine a flashlight on a
ball in a dark room, right, it scatters light and then your eyes can see it and you say
‘oh there's a ball sitting there’. The atoms do the same thing. They emit, they emit photons that if your
eye was visible through the UV you could see them. It would just be a tiny dot on a very dark
background. You can see a single atom fluorescing. The darkest, you know that the brightest star
on a very dark sky. Unfortunately we can't see in the UV with
our eyes so we have to use a camera. But if you were to walk into our lab you would
see large optical tables that are about six feet long by four feet wide filled with lasers. Because to do all the different operations
you need different frequencies of light. And then you'd see another vacuum chamber
that holds or another optical table excuse me the holds our vacuum chambers. OK. So let me just see if I follow. You load your system Yes With ytterbium ions? Yes. And to perform a computation just, for example,
how do you clear the memory? What would be the first step in your computation? Sure. So usually you start with some number let's
say between two and five ions is what you'd want. So you'd load two or five ions, let's say
two for this example. So we can turn the oven on for a set amount
of time then we shine the laser that takes that rips the electron off to create the ions
and we wait till we get two ions. And then we can see them on the camera. And so we initialize the system to a zero
state just like in conventional computing you have to initialize your computer to zero
state, and then if we wanted to put those two ions in a superposition we could shine
either a laser or a microwave at them and that would create a superposition. So you could think of a qubit on what's called
a block sphere, which is just a unit sphere where the up z axis could be your one qubit
state and then down z axis could be your zero qubit state. So you're prepared it in a zero, and then
we shine these microwaves or lasers on the atoms and it causes the population to rotate,
basically. And so you just stop when you get to the upstate
and you can look at that, it’s a trace on a scope. So for example I mean a standard operation,
simplest possible operation you might have in a computer system is a ‘not’. That's right. So how would you do a ‘not’? So you prepare, you would prepare your qubit
in the zero state then you would shine a microwave a laser beam on it for a set amount of time
and it would cause the population to evolve to the up state, and that's a ‘not’ gate
if it go from zero to one that's a ‘not’ gate. If you let that light or microwave interaction
on and it would go back down to the zero state and it would just keep rotating. So suppose I want to do something a bit more
sophisticated like an ‘and’ or something that actually combines two qubits. How would you how would you do that? So if you had two qubits in trapped-ions,
the nice thing is that because they're charged they want to repel each other, but because
there's a trapping potential on them they get pushed together so they find a happy medium
where they sit. But they have a shared motional mode due to
this interaction. So there are a lot like a Newton's cradle. If you pull a ball in a Newton's cradle you
see all the balls move together. That’s right. That’s right. And so the trapped ions do the same thing. If you start to shine a laser beam on one
and you excite some motion and actually excites motion in in both ions. And so then you have a databus that you can
get the ions to talk to each other. And if you had five ions you could actually
use this databus to get one in five to talk to each other directly. So you're not limited to your nearest neighbor
interactions in a trapped ion system. And you can use that to combine them. You can use that combined motional mode to
get the qubit states to talk to each other and create things like controlled ‘not’
gates, say. Great. So the Jerry, can you kind of repeat that
kind of virtuoso performance for your own lab? Yeah, I'll do my best there. But what's it like in there actually? So our lab looks a lot different from what
Kathy-Anne described. And that, that the reason for that is because
the underlying qubit is very different. One difference that, the main difference is
that instead of actually having physical, naturally occurring qubits, in this case ytterbium
ions, that you can you know that all of this work is based off of having a really, really
stable atomic clocks. What we're doing with superconducting qubits
is to actually engineer and build them on a chip. So it's a little more integrated. You're actually using lithographed techniques
that you know and love today with your silicon processors. And instead of the materials that are in your,
in your in your chipset or in your phone or your laptop we're using slightly different
materials to build superconducting circuits. So superconducting refers to materials that,
that when they're cold they have basically no resistance. And by using the right kind of superconductors
you can actually build quantum effects into circuit elements. So with Kathy-Anne, I have a good picture
for what the bit is. The ion is either pointing up, or you know
rotating that direction that corresponds up, or the other way so what's the corresponding? Yeah, so the way to think about it here is
that you're actually building an oscillator circuit. So if you if you go back to your electrical
engineering days think about the circuits that you might build with resistors or capacitors
or inductors, these are varied circuit elements. In the case of a superconducting circuit you
actually could use an element known as a Josephson junction. And a Josephson junction is basically a sandwich
of aluminum, aluminum oxide, aluminum. And what's phenomenal about this this element
is that you can combine it with it with a standard capacitor and you can make it oscillate
in the microwave regimes, so around five gigahertz and choosing the right parameters of the capacitance
in the Josephson injunction you can isolate it to build a qubit state, so zero and one,
that that resonates at around five gigahertz. So in your case can you walk through an example. You load your computer… Right. So in this case, in this case where we have
a silicon fabrication facility that builds these circuits, we, they come out in large
wafer form and then we have to cut them up into smaller chips. These chips are packaged into a printed circuit
board like what you might see in inside your phone. But this printed circuit board carries microwave
signals and so those the printed circuit board then needs to be cooled down to really, really
low temperatures to basically have the qubits function properly. So I said that we use the superconducting
materials. And so the materials are niobium and aluminum. And for them to superconduct and for there
to be so little noise that we can actually see these quantum mechanical effects at five
gigahertz, we need to cool down to fifteen millikelvin. So that's, since you already brought up the
space analogy, it's colder than outer space as well. And in fact you know with the microwave background
space is around it's a little under four kelvin there, but we're getting down to fifteen millikelvin. Wow And so the refrigeration systems that we built,
that we use they're commercially available but it is phenomenal that you can just turn
the turn hit a button turn a key and cool down to these these devices to such a low
temperature. So in the example of the trapped ions, if
you want to execute a ‘not’ operation you hit it with lasers or microwaves. What do you do in your case? Yes. In this case it's more electrically controlled. So you you're placing this chip inside this
printed circuit boards, it's inside of the refrigerator, but then you have all these
wires that come down through the refrigerator and those carry electrical signals. And so to do say, a ‘not ‘operation, what
we're doing is basically applying a shaped microwave pulse that's generated at room temperature,
so on the set of electronics that sits outside of the refrigerator, we generate a five gigahertz
signal for a certain amount of time say maybe twenty nanoseconds or thirty nanoseconds. That pulse gets sent down into the refrigerator,
applies just enough energy to flip your qubit state from zero to one. And then you could do. How would you do an ‘and’ or ‘nand.’ And then with regard to two qubit gates, so
our particular architecture connects qubits on the chip. So there's there's other microwave circuitry
that is used to, to define particularly interactions between qubits on a chip. But then those, those interactions are again
activated using microwaves so just the way that we do the not we might send it we might
send pulses at a slightly different frequency down into the refrigerator to induce a two
qubit operation such as a controlled ‘not’ gate. So everything you've described is acting on
the system. So how does this system act on us to return
its information and the result of the computation? Yeah, so in the case of Kathy-Anne they're
sending another laser beam to do the detection, and you can see it with the camera, but us
what we actually have to do is send a another microwave pulse which is resonant with a detection
cavity, so there's actually a resonator on the chip that oscillates at a slightly different
frequency depending on if the qubit is zero or if the qubit is one. And so we interrogate this cavity with a microwave
pulse and at a very low energy levels, so single photon energy levels at say six gigahertz,
and so that that signal goes down into the fridge, gets amplified through various stages
and then we basically have to digitize it to determine whether the qubit was a zero
or one. Cool. Seth, just to kind of bring some perspective
on the technical discussion here. What would be some pros and cons of the different
techniques? Why would you use trapped-ions in some cases,
superconducting qubits in others? Well so pretty much anything at the microscopic
level will compute if you shine light on it in the right way, via either lasers or microwaves. But some things compute better than others. So what's been happening over the last decade
and a half or so is that the technologies for instance superconducting quantum computing
have really advanced by a lot. I mean I was participating in the early experiments
to build I think the second superconducting qubit around 2000. It was a so-called ‘flux qubit’ these
super currents you have a little loop interrupted by Josephson junction. And so super current going around forever
that way you call it zero and super current going around forever that way, counter-clockwise,
sorry, clockwise for you then you call it one. And then you know super current going around
both ways simultaneously both clockwise and counterclockwise simultaneously, that's 0
and 1 at the same time. So that's how you get quantum bit in these
things. But you would let them sit for a little while
and then, you know, they'd get kind of completely randomized very, very rapidly. And so these originals superconducting qubits
were, well they sucked let's face it. That's the technical term like funky, right? So but then there was this great innovation
actually which Greg participated in, I think this was part of your PhD thesis, this was
developing - people thought oh the materials are bad, something's wrong with how we're
building these things. But it turned out that it was really much
more of a design issue and by being really sneaky about how you design these systems
you can make that much, much much, much more coherent so that they could you know you could
have, they could oscillate around or you could perform ten thousand logic operations before
these things got messed up. And so with superconducting systems I think
that what you did in your PhD thesis and afterwards was a really amazing innovation. And then which also allowed, because you building
them of these all these chips you can put many of them together, so there's a clear
path towards scalability. Similarly with ion traps, the first ion trap
experiments were done in the in the late 90s in the mid 1990s but they were you know two
qubit experiments, sometimes two qubits you can still do interesting things with two cubits,
right, you know. You can search your data space with four possibilities
and you can find is it here or here or here or here by only looking once. Like how can that be, classically? But quantum mechanically. We're about to find out actually in a little
while. Yeah. So, so what's happening is that there ,is
that there is a really there's been also with ion traps there have been all these advances
in integration and making ion traps larger and larger, integrating them with quantum
communication lines. So there's been a steady advance in constructing
more and more elaborate and complex quantum information processors. Ion traps and superconducting systems are
the two technologies that are furthest along the way. But there are a whole bunch of other technologies
like nitrogen vacancies and diamond topological systems and all kinds of crazy things because
again pretty much anything will compute. And even though as Jerry was saying that the
twenty qubits, OK that doesn't sound like a lot, but two to the twenty is about a million. Thirty cubits…two to the thirty is about
a billion. Forty cubits that's a trillion. Well you know now you're starting to try to
manipulate these, a trillion numbers, a billion or a trillion numbers and actually that becomes
very difficult classically. So the devices that are being built right
now are just at the threshold where we actually can't understand what's going on inside them
classically. Previously we were able to simulate what was
happening on a hugemungous classical computer and try to figure out what's going on. Now we're kind of on our own and sort of exploring
this quantum frontier and we, you know, we we are going to be able to try to figure out
what's going on. And then the hope is that when we build these
devices we can use them to build ever-larger devices and build quantum computers that have
a thousand qubits or a million qubits or a billion qubits. So Julia-Ann, the machines, and it's a great
moment to be in historically to be in, the machines are now crossing over and exceeding
the power of our most powerful classical computer. But then how do we know that they're working
properly if we can't even compare the result of the calculation of the quantum computer
to a classical computer any more? And working properly. I mean Because I don't know about you but my computer
crashes sometimes. I mean how can we ensure that they're working
properly is maybe a question I can answer because as you ask what, what architecture
will eventually you know win or be the best one and of course the question we need to
answer is which ones scales best for a large number of qubits? And in theory, I'm a theoretician so I you
know I mean I'm in a position where I write my papers saying let's assume we have a quantum
computer of ten thousand cubits and then but there is a lot of theory developed, a theory
of say quantum error correction for instance, where we face the fact that no matter how
well Kathy-Anne and Jerry perform their jobs, the elements out of which they build their
quantum computers will be faulty at some level. There will be a probability that they'll fail,
that they'll lose their coherence and so on. And there actually is a very beautiful theory
of quantum error correction that once we are above a certain threshold with the noise in
their system, so once the noise is small enough, then we can actually build in redundancy into
these qubits, in a way that the computation will flow flawlessly. And that's a very nice theory that will then
allow us to make the quantum computer work at a larger scale. So to fix the mistakes You can fix the mistakes. Yes. So I think this might be a good point to talk
about how they actually work. And Jerry I know you've got a demonstration
you'd like to present to us. Well first, the way that I want to actually
motivate this is based off of this search, a search algorithm and Seth already alluded
to this. But let's say you have four cards, right and
you play the game of monte or you might go to a street corner somewhere. Not that we're advocating you do that. Don’t do that. And so out of these four cards you've got
one of them which is different, one of them is the queen. And now, now we're going to flip them over
and when you play this game you're randomly going to try to find where that where that
queen is right. You're going to try once and you're going
to flip over a card and see whether or not it's the queen. And so on, on in playing this game you really
only have a one in four chance of getting it right on your on your first try. But now what's interesting about this type
of game is we can also ask well how would we how do we do this if we had a computer? What what what does a classic computer do
with this game and what does a quantum computer do with this game? OK. So in the case of a classical processor what
we're doing is when we when we flip over these cards you can think of this as as storing
a database. In this case we can also call it an oracle. So you store the database with the hidden
set of cards where the queen is properly located. With a classical computer what you're going
to do is in order to to find where it is you're going to look at all the possible arrangements. Right. So you're going to start with one particular
arrangement. Let's, let's start with placing the queen
in the first slot. And we're going to take that entry, use it
as an input, we're going to do some processing, in this case the green box where you're gonna
do some comparison with what's in the database and you're going to make a decision at the
end of it whether or not it correct or not. In this case it was not correct and you get
a zero. OK. And so now then you can try the next one. And again you're going to get to get a zero
and then the next one and this time you get it right. But of course classically you're going to
go through all four of them. And so after you go through all four you see
that on average you would have gotten this correct, basically you would get a correct
after querying this database about two and a quarter times. Right so this problem of search in this case
with a classical computer you can only kind of do this sequential, sequentially or by
choosing at random. But with a quantum processor, this is where
a lot of the ideas of quantum mechanics can come through. And so in the next slide here with the quantum
processor you have access to superposition. And so just like we talked about how you can
be in zero and one at the same time, what you can do with two qubit system is to make
a superposition of all four of the possibilities: 00, 01, 10, 11, to represent all four of the
possible arrangements of this hidden queen. And so we can take that superposition state,
use it as as basically as an input, call the database just once. Perform some processing step, in this case
the processing step involves entanglement and it involves this quantum interference
of adding adding together the waves. And it'll amplify the answer for exactly the
right answer. And so every time, no matter what you what
you place into the database, wherever you hid that card, you use only one call to the
database, you get the right answer using this algorithm. And this particular algorithm is known as
Grover's algorithm. It is a simple case that gives you the sense
for what is done differently in terms of processing information on a classic computer versus a
quantum computer. And so on the next slide what would you actually
would do when you want to program an actual quantum, quantum processor is to use this
language of quantum gates. And so what you see here is actually a quantum
circuit. And as Seth alluded to the idea of music,
we call this actually a score because it kind of looks like a musical score. And the concept of time really in time with
gates really has a strong analogy here because it's like you're playing different notes on
these different qubits. What you see here is really just, just two
of the qubits being populated with these different operations, which realizes Grover's algorithm. And to break it down a little bit further
in the next slide what you, what you see is that these various steps of superposition,
the stored database and the actual post-processing steps are all, are all encoded into these
various gate operations that you can apply. And in this case we can actually run it through
and get , get a result. And I can actually launch this live if you? Please by all means. While we are you switching over? Kathy-Anne, you were really one of the first
people to actually do this for real in your dissertation work. Can you describe what you accomplished? Yes sure. So we did it with trapped ions, we had two
trapped ions and as Jerry just showed you that'll give you the four element database. And what we did in practice was we had the
computer mark a state and then we would run the algorithm similar to the diagram that
Jerry just showed, it looks very similar and then at the end we would see what the probability
was that we recovered that marked state. And at the time the untangling gate that we
used we were just starting to learn to use it, it had just been demonstrated. And so we found the mark with a probability
of about sixty percent. But Jerry just told you it should be one percent
and that's because the fidelity, that's one way you can measure how good a quantum gate
is, the fidelity of our gate wasn't as high as we would have liked it to be, mostly due
to technical difficulties. This is a fundamental limit of trapped ions. They've since repeated this experiment with
three qubits recently, Chris Monroe's group at the University of Maryland, and they did
quite a bit better because the technology has progressed. Now at this point at these gates are at very
high fidelity near the fault tolerance level that Julia was saying earlier that you need
to run these computers. You were telling me earlier that these are
so delicate that you can so much as look at the laser wrong and it would it would give
you the, it wouldn't work. Yes so we used cadmium ions in my graduate
work and they, they need laser frequencies that are about two hundred fifty nanometers,
which is an incredibly difficult color of light to generate. You basically have to quadruple a laser to
get there. So you take a laser, you double it and then
you double it again. And doubling's hard and the efficiency is
low. And one of the people in our lab just had
the right acoustic sound to his voice that he would unlock are doubling cavities and
so sometimes when he came in the room he would start to talk and we were trying to run our
experiment and our laser would shut off. They're a very fragile system. So Jerry are you ready to go with thIs? Yeah. This is actually a live quantum computer. I wanted to start by just showing a little
bit about the interface of what we have. So this is the IBM ‘Q Experience’ and
what we actually have is a lot of content on there for anybody to get started with learning
how to program and actually use a quantum computer. So anybody can do this? Anybody can log in and sign up for an account. We have this library with various user guides
for beginners, if you're more familiar with some mathematics like linear algebra and even
another other guide which actually leads you to our Github developer repository. But through this, through this portal you
have access to learning about the basics of a qubit superposition, entanglement, simple
algorithms such as this Grover’s algorithm. And we also have a community board feature
where we have the ability for anyone to ask questions and to, and our IBM researchers
are more than happy to answer. Julia I wanted to go back to what you were
saying about the factorizing problem, that's in addition to the search algorithm the other
use that people we talk about with with quantum computers, so what's what's kind of state
of the art in that when you factor the number fifteen or, or get that high even? You can, I mean Jerry would probably be a
better person, but with twenty qubits you can imagine that you can factor perhaps with
some overhead I would guess you can maybe factor numbers up to a hundred? Which of course you can do in your head. So at this stage we are really at the level
where we demonstrate things when it comes to factoring. The cryptographic systems that I was talking
about that your credit cards rely on these usually have something maybe up to a thousand
bits? So I think once we get a quantum computer
to the order of perhaps one or several thousand qubits then you better stop using your credit
cards with the current encryption. So… One thing I would I would like to mention
though with regards to the Shor's algorithm though is that because of the error rates
that we end up having with the physical qubits, sure if you have a thousand perfect qubits
you might start thinking about Shor's algorithm for a thousand, thousand-digit numbers but
with, with needing quantum error-correction and a lot of the best known encodings you
have an overhead that significantly pushes that threshold further. So I think that in terms of Shor's algorithm
and a realistic Grover's search you're thinking about probably needing millions, of tens of
millions qubits. So it's it's a bit further off but it's still
there will come a day where There will come a day when you've got to worry
about your bank accounts but it's on, the horizon for that is a bit further beyond where
we are currently at. So millions of physical Millions of physical qubits, yeah so that's
only maybe around a few thousand logical qubits but the, the encoding that's what's going
to matter there. So Kathy-Anne you were describing also to
me earlier that on the one hand quantum computers take away our privacy by breaking these codes
but they might also restore. Can you describe some work, work you've done
for the restoration process? Yes so there's people working now on, in addition
to quantum, computing quantum networking and what you can do with quantum mechanics for
networking communication, the easiest example to explain, and it's been around for a while,
is called quantum key distribution, where you send a message between two parties say
Alice and Bob using single photons. And because the photons are encoded using
quantum mechanics you can actually make protocols that are ultra secure. And by that we mean they're tamper-proof,
meaning that even if an, an eavesdropper can't get between Alice and Bob to get the signal
because they don't have the corresponding information that was encoded in the quantum
mechanics. But even if an eavesdropper were to try and
grab the signal the protocols are tamper evident so Alice and Bob would see that immediately
when they started to talk about the results that they'd gotten and would abandon the protocol. So yes it can do things like break, break
encryption but it can also provide ultra secure protocols too. So what are you doing in your lab now to kind
of bring that into fruition? So we're working on quantum networking where
instead of sending key information, which that just sends information to generate a
key and then you would use the key for something else. We, and a lot of people in the field, are
moving towards quantum networking where you actually send quantum information directly
over some longer distance link. And so this allows you to do things like ultra
secure communication protocols or people are also looking at it for distributing computing. Where you don't just have one computer sitting
there with millions, let's say, of qubits but you distribute these qubits over a larger
space and you have smaller banks of qubits. Seth, you once told me, this is a couple of
years ago now, I don’t know if you remember the anecdote, the NSA funded some early quantum
computing work to show this wasn't possible because they didn't want to have unbreakable
codes. Can you walk through that? Oh yeah, so I was at the, so back in 1993
I wrote the first paper showing how you build a quantum computer using these methods of
zapping stuff with microwaves and lasers and things like that. And then we started to work with people to
build them. In 1994, I think the first U.S. government
meeting to fund, to discuss funding for quantum computing took place at DARPA in Arlington,
Virginia. And during this meeting there were a bunch
of people, including Peter Shor, they were there talking about stuff and a fellow named
stood up and he said I'm Keith Miller from the NSA and I am authorized to tell you that
the NSA is interested in quantum computing and then he sat down again. And everyone went, oh my God! Some people actually told us something. That's incredible! But it caused such a stir that he stood up
again and he said well I believe I'm also authorized to tell you this, of course the
NSA is interested in quantum computing because our primary mission is to protect the secrets
of the country, up to thirty years for top secrets. We have a whole bunch of information that's
out there that's already encrypted which if someone could build a quantum computer could
be decrypted. And that would be bad. So really what we would really prefer is that
it not be possible to build a quantum computer. By the way this is a good person to have funding
you, it's like they call up and they say how's it going? And we say oh it's terrible, the qubits aren't
working. Great great! That's wonderful. Here's your money. That didn't last very long. So then he said. But because of our secondary mission, if it
is possible, we want to have the first one, so. To bring this back down to more quotidian
kind of applications Jerry, you once described to me some of the molecular calculations you
were doing. Can you walk through what you're doing with
these molecules? Yeah. So, I think one of the more kind of near-term
areas that we'd like to look at application wise with quantum computers, actually is in
chemical simulation. So what's actually interesting is that it
dates back to Feynman around 1980s when he actually talked about, wouldn't it be great
to actually simulate nature using something that follows the same quantum mechanical principles
of nature. And there's been a lot of theoretical work
going into how would you actually map say problems in quantum chemistry, for example
electronic and molecular structure, onto physical quantum bits. And it's a really neat idea in the sense that
you can you can actually try and get an analog for a physical, a real physical system such
as the energy levels of say a hydrogen molecule, but actually run it on a on a on a chip right? Run that simulation on qubits that's inside
of one of our dilution refrigerators. And so we, our team has done various both
theoretical explorations and recently experimental demonstrations of how to do some rather simple
molecular calculations. So looking at the energy the ground state
energy of a simple molecule just like hydrogen so two H's and then lithium hydride, beryllium
hydride, but very small at this at this at this stage. But it shows the type of trajectory, if you
will, of our application in the near term because at some point with these different
molecular structures you get to a point where there's too many electrons in it that it's
impossible to again, simulate in on any classic computer. And it can be rather modest molecular sizes
that that already maxed out all those supercomputing resource in the world. And there's a lot of potential there for quantum
computing to really be a game changer in that in that field. Seth, I was wondering if you could fuse, merge
for me the two great computing tasks of our time; machine learning and quantum computing? Is there a relationship between the two? Yeah. For there is the only way to get information
right now is to, you know, sort of the only way get a grant right now is to apply to do
something with big data machine learning. And then in physics the only way to get a
grant is to do something with grapheme -the material is the future along with gallium
arsenide. So the real reason is to have something so
you can get a grant that's you know graphene based quantum random access memories for the
analysis of big data. It's a winner. I guarantee it. You heard it here. So it's interesting now that we are we actually
are about to have a simple quantum computers that have, you know, tens of qubits and fifty
qubits coming up. And I think that there's a reason reasonable
path to think of having up to a thousand physical qubits over the next five to ten years. I don't think that's unreasonable to expect. What will you do with these devices? Now because they are quantum mechanical and
they're very hard to simulate classically, As Jerry was saying, quantum mechanics you
know it's famously weird and funky and quantum systems exhibit funky and strange effects
like entanglement and Einstein Podolsky Rosen correlations, and Schrodinger's cat, and statistical
patterns in data that are very hard to capture classically. They're counterintuitive, it's hard for classical
computers to capture them. So if they can exhibit these, if quantum systems
can generate these funky patterns that you can't generate classically maybe they can
also recognize patterns that you can't recognize classically. Now machine learning is about taking patterns
of data and trying to tease them out and show that they're there, it's recognizing patterns
in data. Machine learning of course very trendy right
now, justifiably so not not really because actually I think you know it's about to supplant
human beings or anything like that but because actually it's gotten good. You know there's this there's this thing called
deep learning, which when I learned about it a three or four years ago I said wow! this
is fantastic, you know computers will tell us about love and truth and you know happiness
of all this deep stuff, but no such luck. It turns out that these are just neural electronic
analogs of neural circuitry that have many many many many levels in it so they're deep
in that sense. But they actually do do problems, they solve
problems that are hard to do. Now do you get inside of a machine learning
algorithm like say the Netflix algorithm where you know you say OK what should I watch today? And Netflix says, ‘’well I think that
you would like to see Dirty Harry’ but you, for some reason my students don't watch Clint
Eastwood any longer, I don't know what it is. You know what Netflix is doing is they're
actually looking at the preferences of everybody out there who's looking at Netflix, comparing
your preferences to theirs and then you know doing what's called a matrix completion algorithm
to recommend something to you. Now if you were to program that in a co-op
into a quantum computer it turns out that their algorithm which they only run, they
run it twice a day because it's so incredibly computationally intensive, that if you do
that on a quantum computer you could have a quantum computer that had say a hundred
quantum bits and you could do ten thousand operations and it would do the same set of
operations in a quantum mechanical fashion. So we decided hey this is great we'll call
this quantum Netflix algorithm, but then I googled quantum Netflix algorithm and it turns
out that Netflix calls their own algorithms "the quantum algorithm" even though it has
nothing whatsoever to do with quantum mechanics. So using you know quantum computers, quantum
systems in general exhibit strange and counterintuitive patterns. This gives you reason to hope that they can
recognize strange patterns and it turns out that the actual stuff that they're doing already
for things like factoring numbers is great for actually finding patterns and data. And actually this is a nice application that
people have been using to demonstrate, you know, simple versions of these algorithms
on small quantum computers Obviously the world as we know it would not
be the same without computers. They're just everywhere, they're ubiquitous. Will fifty years from now people say the same
thing about quantum computers? Will they be as transformative as classical
computers have been? That's an excellent question. I view, I think a quantum computer will remain,
it might be ubiquitous but it will remain a special purpose device for various things. I don't think it will replace the computers
as we know them in its entirety. So I view the future perhaps as you having
your laptop and then a little dongle with one of Jerry's or Kathy-Anne's contraptions. And then whenever you know whenever you need
to break into somebody else's credit card or whatever it is that you want to do you
make you know you make calls to that, you know, to that special device. I think that, that's more likely picture of
the future with a quantum computer in it then, yeah. So it's more like a GPU inside of an Xbox
type of thing. Yeah. I guess. That won't be literally a quantum iPhone. Although Apple may trademark that before Netflix
or Verizon I’m more optimistic, I think, you know,
they build it they will come. Right? So once we have quantum computers to play
around which we already do have thanks to IBM, I'm all over that and people will play
around, will come up with you know more quantum apps, quapps, quapps for all. Can you trademark that? I trademarked the cloud with a q. Questions. So the four card monte example I mean it's
it looks to me like something that could basically like backwards solve any kind of cryptographic
hash like as a black box or whatever the hash is. And yet I hear about these, these quantum-proof
cryptographic methods and because it seems like it doesn't matter what the actual hash
operations are, how does that, what is the general underlying principle for these quantum
proof hash hash hash functions? So a hash function in cryptography is a function
that just like scrambles everything up in a way where you can check to see if it's been
scrambled up in a proper fashion. And inverting these, so undoing this hashing
is supposed to be hard and that's the basis for a lot of cryptographic protocols. It's still hard on a quantum computer that
is this quantum searching will allow you to get a speed up to that will allow you to you
know solve some problems that you would be able to solve classically. But this kind of hashing problem is still
hard on a quantum computer. So one of the things that's going on right
now because exactly because quantum computers are getting more powerful, though let's face
it we're still you know we can basically compute our way out of a paper bag now where previously
we couldn't compute our way out of a paper bag. So but you know even the NSA has issued an
advisory saying you know if you're going to come up with an application that's good and
will still be secure twenty years down the line it's time for you to think of something
in addition because quantum computers might be there. So people are coming to trying to come up
with what's called post-quantum cryptography and I think that you're alluding to some of
these problems there. Now you make me wonder what post quantum cryptography
could possibly be? Can you just give a simple example? So based on a, so public key cryptography
is a way where you know I send, suppose so I buy green coffee beans over the Internet
and then roast them at home. Which you actually do. Yes I do actually. So yes it's so much so much fresher that way. It really is. I highly recommend it. So what I like to say try to send buy something
from Sweet Maria's in Berkeley, you know this 10 pounds of Costa Rican, then Sweet Maria’s
sends me a big number which is the product of two smaller numbers which are prime numbers. And this is called the public key, this big
number. I could use that number to encrypt my information
in a way such that only sweet Maria, who knows the two smaller numbers, can decrypt it and
that's the basis for public key cryptography. There's a public key, which is what they sent
out there. Anybody could encrypt but to decrypt you need
you know the private key, these two numbers. This is what quantum computers can do. If they can find the private key given the
public key which would be very disruptive thing because I frankly I like buying freshly
roasted green coffee and I would be pretty pissed off if I couldn't get it. So the idea is to you see this as a rather
specific protocol, so what people are trying to come up with are other protocols where
quantum computers can't break those protocols where there is a public key you can encrypt
using the public key but then it can only decrypt using the private key. But a quantum computer can't find the private
key. And so far there's been mixed success I would
say doing this. There's not nothing's ready for primetime. Yeah I should say. This is also important because even though
quantum computer is not there you might want to encode your information in a way that nobody
can decode it in the next one hundred years and one hundred years is a very long time,
right. And then we might assume the quantum computer
could be there, I mean a big one whatever. And the post refers to the fact that even
though you encrypted today maybe you don't want it to be decrypted you know in ninety
five years by one of the successors of Jerry's computers. And there are, there are methods nowadays
in fact that cryptographers have started to develop but they're extremely impractical
at this stage, that the public keys you would have to transmit are so long that it would
take you, you know hours basically to do that. But it's it it would be wrong to say that
there is no alternative but it's not a practical alternative. More questions. I saw a bunch. How does the observer effect claim to like
retrieving information? Can you elaborate? Well it seems like if you try to like observe
the information it would collapse, right, like it would just collapse back to two bits. So like how do you maintain like the four
bits or whatever? Observing it is essential to how we actually
make use of a quantum computer, right, because it has to be something tangible that we can
put in, has to be something tangible that we can take out. So the input will be classical bits the output
will be classical bits. In-between is where we make use of superposition,
entanglement and this two to the n exponential space, state space. And the key thing is is how do you tailor
your algorithm to make use of that so that when you perform that measurement you've learned
something that you otherwise wouldn't have wouldn't have been able to calculate. So it's all about how you define those interferences
of the waves through the operations you perform in between. You only observe at the end. Before that it's considered to be rude to
look at somebody's quantum computer while it's in operation. So, that's part of the difficulty in controlling
these very complex systems is because if you make a measurement or if the environment makes
a measurement without your knowledge the same thing happens. And so you have to control the system very
well so that you only look for, the environment only measures the system at the end of the
computation. So it sounds like a lot of the stuff that
you guys are working on is kind of like a straight analogy from a classical computer
to a quantum computer where like a bit is a qubit and you're working not gates, I was
wondering if you could talk about how quantum annealers like what D-Wave works on how well
you guys work on factors into that and if there are any sort of limitations using the
annealing paradigm. Cause I know that's usually better off for
like combinatorial optimization problems. But is there any other sort of limitation
on what an annealer can do versus what a quantum computer can do? Well I think that the first thing there an
annealer is a very it's a more restricted type of problem, right, so you get your hardware,
the way that you lay down these that these circuits in an annealer you defined all the
couplings between these these these these devices and you've defined a particular energy
landscape that you want to say optimize or find the ground state for. In the case with the systems that we were
building where you have full quantum control over any of the qubits, you really can drive
the system to any kind of quantum quantum problem that you want. And so it's it's reprogrammable in that sense
and you can define your optimization landscape in more generality. But maybe Seth you can also comment on the
D-wave. Yeah, I mean, also let me say that you keep
on letting all of the people who possess prior knowledge into this room will make our lives
harder. So what's that about? So quantum annealer as as Jerry was saying
it's a it's actually a very old idea and classically, there's a notion called simulated annealing
classically where you want to solve a hard problem. So you want to find the the minimum value
of some function, many problems are like this, like the traveling salesman problem, I want
to find the shortest path that will get me through all the cities of the United States
and back to where I started. That's a hard problem. And so what you do is you map this problem
into finding the lowest energy state of a physical system and then you try to find this
lowest energy state by cooling, annealing, that's why it's called, annealing to get down
to this lowest energy state. Now quantum annealing is a sneaky trick that
does this quantum mechanically. You construct a quantum system. I mean for instance D-Wave quantum annealer
is a tunable device with up to a couple thousand quantum bits, and they can tune all the couplings
between them And then you set it up so that the lowest energy state encodes the answer
to your problem.and then you try to find this lowest energy state by kind of oozing in a
funky quantum mechanical state of fashion from some known state to this unknown state. And either it works or it doesn't. Now it is an interesting situation because
actually nobody knows if this works is supposed to work even in theory. And so it's one of these things where if you
build it and then you see what happens. They find that some fraction of the time they
actually get the right answer. There's a lot of argument about whether this
is happening in an intrinsically quantum mechanical way or not. But I mean these are very interesting systems
so I mean D-Wave deserves great credit for building a large scale quantum system. It's got lots of entanglement in it. It's got, you know, it's it's has thousands
of quantum bits and it's actually a beautiful system just for doing experiments on. I'm going to a conference in Japan next week
or two weeks from now where basically people are going to report on all the experiments
that they're doing all these different D-Wave devices to try to figure out what the heck
is going on. Great. I'm afraid I'm going to have to cut off the
questions there Thanks to all of you for coming.