Quantum Supremacy with Dario Gil (IBM) | Disrupt SF (Day 3)
4:54PM Sep 7, 2018
We're going to start with quantum computing. So I know like 1% of what there is to know about quantum computing. So hopefully this will be highly educational for all of us. Please welcome to the stage from IBM Dario Gil, to talk about quantum advantage and your moderator Frederic Lardinois.
Any quantum physicists in the house who got up early for this? Yes, my team. Awesome. Thank you. IBM has been working on on quantum computing since the what the 70s?
Even today. There's very little that's intuitive about quantum computing. If Jordan knows 1%, I think she's exaggerating there, but 1.01% is what I know. So if I gave you this stage for two minutes, and you had to explain to us what quantum computing is all about, what do we need to know, to understand quantum computing?
Yeah, I would start with the foundational principles of what makes a quantum computer, you know, potentially powerful. And this three key elements, right? I mean, a quantum computer fundamentally exploits the principles of quantum physics to operate very different in the computational realm compared to a classical computer. And the first principle is the principle of superposition.
And now one way to, you know, understand that it is imagined that you have a coin and in the classical world, we have our zeros and ones and we flip the coin, it's in one of the two states. But if we were spinning the coin, we would see that, you know, in some senses in a superposition of heads and tails, right, so that's Principle number one.
Let me stop you for one second there. I think we're supposed to see something on the screens and we're not so if anybody can bring that up in the back or not. I'll let you talk. But there should be something up at some point, you will see something up there.
Yeah. So. So that's the first principle. Then the second one is something called entanglement, which is a, you know, a beautiful understanding of, of the nature of our physical world.
And now, to just give you an analogy, imagine you have two coins, right? So each one can be in the superposition of heads and tails. But now we have an interesting situation that they're entangled which we get to do when we have quantum bits instead of coins, then then we can have these correlations these perfectly correlations at a time of measurement.
So if we did a particular measurement, if one was measure heads, the other one will be measure heads too. And if one was tails, the other one will be or tails as well. Now that is something that you do not see in the classical world. Right. So the type of measurement that can happen when these or qubit entangle is fundamentally different than what you can see classical.
And the third principle is the principle of interference. So once we have our qubits or quantum bits and the superposition on entangled state, we get to do an operation that we cannot do classically, which is to in a computation sense, which is to interfere.
And interference is the idea that you can sort of cumulatively add amplitudes so that the peaks go up or you can cancel. If you have a positive and negative amplitude of the same amount, you can cancel that and that's why, you know, when we see it on waves, or noise cancelling headphones, we have that idea.
So those three principles are the kernel of how you go about building a quantum computer and what gives it its power and I'll share in a little bit perhaps how an algorithm works and what's the power.
Did everybody get that? Did you mostly learn how to cheat in Vegas with coins?
That's right. If nothing else quantum coins
Quantum coins will let you win in Vegas. For a second, maybe talk to us how that's different from a classical computer?
Right. So. So if you look at in our classical systems, the contrast is you have these bits of zeros and ones. And the logical operations is what we all learning in high school logic of the ands and the ors, and the nots and so on, right. So a classical computer is nothing more than a device that is able to process the logical operations expressed in bits, right in zeros and ones.
In our quantum world, now, we have these qubits. So the way an algorithm works is when you see on screen is a good representation. Here's kind of what you have to do. The first thing you do is you put the computer in a superposition of states.
So let's say we had just two cubits so we have two to the power of n, where n is the number of qubits. So we have two qubits, two to the power of two is for so our computer is in a 00011011 state. Step one.
Second state is we got to inject data into the computer. So each one of that sphere that you see where the dots is represented, that state is represented by an amplitude on a phase. And putting data into the computer just means changing the face of my four states in this case.
And the third step is you can take those states that are written on face and interfere them and the algorithm is the sequence of expressions that manipulates those states such that the right answer is maximized. And that is the power right because I get to have many many states representing complex information and I get to do this process of interference to find the right answer and that gives enormous power.
What then today there's a quantum computer actually look like?
They are actually beautiful machines and the good news is we have one that you can see
This was not practice at all
yeah and so so here's what they look like. You can see a you know a real the inside of a quantum computer. There are different ways to implement quantum computers. The technology that we use at IBM is based on superconducting qubits.
So what you're seeing here is the inside of a cryostat. So at the very bottom of this device, that
Explain maybe for a second, what's a cryostat.
It's it's basically a refrigerator right? And and what happens is because these qubits are very delicate, right? Disentangle, entanglement, and superposition is very delicate. Anything from the outside world can couple to it and destroy its state.
So well, one way to isolate it and protect it is to cool it. So at the very bottom of it, where you see a metal canister it is it operates at about 15 millikelvin, so that's about 100 times colder than outer space. And inside that that canister we have a quantum processor that we we fabricate and that is the one that contains inside the qubits.
The wire that you see are superconducting coaxial cables because the input into a quantum computer, it's a sequence of microwave pulses. It goes all the way to the quantum processor. It performs a logical operations. And then we extract the readout, the signal goes to the coaxial cables. And then we interpret the signal.
Now people have been working on this for 30, 40, 50 years at this point. There was a long time when nothing. Well, a lot of things happen in theory, but suddenly we actually have machines.
Like this one. What happened? What changed?
Yeah, great question. So. So in the from technology side of the house is first the qubits got a lot better. So there's a key measurement on the qubits, and then called coherence time. Basically, it alludes to the fact of how long do you have a qubit before it becomes a bit. I mean, how long do you have this superposition, entangled state to do computation.
For superconducting devices, in the late 90s, to give you a flavor for how long that time was, was about one nanosecond. That's how much it was
Were you able to do anything in one nano
No, because you couldn't do with one nano second, you could do some early demonstration of the principle. But you couldn't run an algorithm because an algorithm is a sequence of gate operations, right? That you have to do to manipulate the qubits. So at present in the devices that we have available for anybody to use all over the world, we have ranges of between 50 and 100 microseconds is how long it lasts.
So the key one key answer to your question is the devices have gotten a lot better. Cryogenic technology, the ability to cool and maintain it stable, has gotten a lot better. And the control electronics and the algorithms associated with the control also have improved. So for the first time ever, we're in a dream that has been going on for many, many decades. Finally, we've been able to build working quantum computers.
And now we're going to do something today on stage that has never been done, we're gonna do a live demonstration or running a classification algorithm with a with a live quantum computer.
Absolutely. And what can you know, 50 microseconds still doesn't sound like a lot long time?
It's not a long time.
No, not at all. What can you do in those in that short time frame?
Yeah. So you can actually start implementing 10s and dozens or even hundreds of operations
So you can run short algorithms to do that we've done demonstrations of simulating molecules we published on you know, in Nature last year, the largest molecule ever simulator with a quantum computer, we've done classification tasks and actually they they community has have now hundreds of scientific publications using the devices that we have put online to to test the principal quantum mechanics and to run algorithms.
So now you can actually run things.
Why don't we do that?
Yeah, let's try let's try
Let's see what that looks like.
Yeah, so first one, we're gonna see if we can connect live so you can see in one over laboratory in Yorktown Heights, so it's about 45 minutes north of New York City. And this is one of our, of our quantum computation centers. And you're seeing now you're hearing let's listen for a second, what is the sound.
So, that chirping sound is associated with the pumping the gases that have to go inside the cryostat which you're seeing it in close with a white cylinders to be able to extract energy from the system. And cool it progressively. So the sound of a quantum computer of this type is that chirping sound, right? That goes in there.
So, okay, so what we're gonna do is we're going to run one of the algorithms that you can run is our quantum super Vector Machine. So we're going to do a very simple classification task. And further, I'm going to ask you whether we should classify a cat or a dog.
You made me, I like dogs, but it's quantum stuff so it feels like cat is just the right thing to say.
Okay, sounds good.
Let's do that.
Alright, so we are going to send an experiment to do a classification of either a cat or a dog and an emoji and we just send it out. And let me explain you what's happening right now.
So the way it works is, and you'll be able to, you know, run these kinds of experiments from from your browser. This is kind of how the quantum system works. you implement the, the algorithm and you have a both a graphical user interface and you can implement it in Python.
And you send your what's happening right now, is we're sending those microwave pulses. It's traveling inside the cryostat it is performing this superposition and entanglement on the quantum bits and performing the operation. That is what is happening right now.
And then you're extracting the signal signal out and because it's a probabilistic machine, right, and you have to repeat the number of shots that you do many times.
Once they're back there, probabilistic machine, but so what does it mean?
So what it means is that when you perform the measurement in the end, you're sort of collapse in this idea of the weight function of superposition. But you have to accumulate statistics to have high confidence in the result, which you can get. But one measurement is not enough. Right?
So, okay, so and then. So this is what happened physically. And then let me just give you a conceptually, what is the idea of a quantum machine learning algorithm? Right? So we've all gotten used to the idea that if you pick a, you know, a classical support vector machine algorithm, let's say you have that line and you have two types of, of of data, right, the light dots and the dark dots when they are expressing one dimension. If I asked you Can you define a line that separates the dark dots and the light dots?
The answer is you cannot right not with one line, but if you just take that one dimensional layout and curve it, it is clear then now we can find a line that separates the two.
So a lot of what we do in machine learning is this idea of mapping and inputs space to a high dimensional space and find a hyper plane that separates the classes. Okay. So based on this idea, we've implemented something called a quantum super Vector Machine. And what it is, is that in this case we are using the state axis is that I have, which is the entanglement, which is a classic property, I don't have classically to embed the features to do the separation.
So we've codified that the way the algorithm looks. So you just have a flavor for what the code looks like is, you know, it gets expressed in you're seeing the series of like, you see an H and Z and something that's called like, score.
And basically, in this case, you have two qubits and you read it from left to right, like if you're reading music, and the the blocks you see are the gate operations and then these you don't have to worry about it. You can write it in code, right? So the way gets expressed in QISKit in in the open source environment that allows you to program the systems is it gets expressing code and then use you send it to run.
So let's see what we we've got.
Let's Let's, before we do that, just a question about this code. Do I need to be a quantum physicist? So I can write this code?
Because most developers are not.
That's right. Yeah. No, no longer the case. So it used to be like, two years ago, before we put the first quantum computers on on the cloud that Yeah, you did have to be a physicist and tune stuff right now, there is a much more sophisticated and higher abstraction stack so that you can make call functions to algorithms.
So if you want to, you know, on QISKit, we release something also called Aqua that if you want to operate at the level of implementing the algorithm to do like machine learning or search or so on you can run it at that level right so you still have to know a few things about it. But you don't have to be a quantum physicists that has changed dramatically and you all have access to it.
So it gave the right answer, right? So we send that and you know, and and that's kind of remarkable, right? That, you know, finally, we've gotten to this stage where quantum computers are real, they work as systems. And there's a wonderful community that has been created around this.
And they can distinguish between cats and dogs now
Which is the third step of getting to the internet age. Right? And, and, you know, so, so people are running all sorts of experiments all over the world on this and just to give you, you know, you can scan people have created quantum games, right, exploiting the principles of quantum mechanics in our quantum system.
No Doom yet. No, but these are great game of quantum battleship, okay, where are the principles of how you sink things is, is is very different. So in our community, you can see, you know, there's over 120 technical publications outside of IBM of people using our systems and this enormous creativity you can scroll and see some of the things there and the community has been amazing I mean, if you want to see a short video of what has happened over two years. So we put the quantum online on the cloud.
And that was in May 2016. And this is what happened in the last two years. So you're seeing how what you're seeing on the world map is the executions and our real quantum hardware. And, and we went through two distinct phases, right? The first year or so people had to drag and drop the logical operations.
You say that like it's a bad thing?
No, it was easy. It is perfect still. And people use it a lot to learn. So what you do is you can learn what a logical operation is in quantum and you can drag and drop on the qubits and you can see so it is a perfect tool and is using universities all over the world to learn.
And this gets executed on a real machine?
Yes, you can execute we have three freely available quantum machines two five qubits and a 16 qubit and then for our or client engagements and collaboration. We are both at 20 and plans for the 50 as well. So this is for education and learning. And you can all try. And you can run on a simulator. And you can run on the actual hardware, right? it executes on the quantum machine. What we then did is because some feedback from the community and to increase the productivity is we released and we created QISkit.
And now that's is the most widely adopted open source environment in the world for quantum. And what it allows you now to do is to implement it with higher productivity, right? And there's a lot more routines in there and you can do things like, you know, algorithm experiments, noise experiments, simulation, and so on.
And it's well documented and there's a good community and if we can show the video then what happened is we went through that first and you saw the growth and then since then the growth has really accelerated. So I present we're close to having 100,000 users right that have run I think it's five and a half million experiments quiz kit itself has you know, over 100,000 downloads, so there's a lot of people trying.
And you describe them as experiments? And that seems fair, I think at this point.
Right. How long we're running a little short on time here. But how long is it going to take before this is more than experiment?
Yeah. So. So there's different ways we look at it in two ways. First, we look at like, the world has to get quantum ready, right? And when we mean by that is that you cannot just think classically in the traditional algorithms and say, the quantum computer is just going to map them to what I'm doing. So there's a massive job to do and you can all be part of this is some learning the new principles are very fundamental new form of computation that will be part of the permanent landscape of computing.
But the second part is then what you do with it and to solve some problems. So that second phase is this idea of quantum advantage. And we believe we've set ourselves the goal that within three years right to do three years is to have a first demonstration of a practical problem that you can do with a quantum computer that has some commercial a practical advantage compared to do any classically kind of the horizon.
What if you had to guess what will that look like?
We believe that the first application to have an impact will likely be in the chemistry world and the materials world. So Richard Feynman famously sort of proposed the idea that if we wanted to simulate and understand nature, we should build a quantum computer because quantum operates according to the laws of nature, a quantum computer.
So we think that being able to model chemistry and that will have important implications to develop new materials. And over time, Life Sciences and drug is one. The second one is we're very excited about what's happening, we show the example of the quantum support vector machine.
So empirically, you know, one contrast I would just quickly before we put these kinds of systems for everybody to tinker and play with everybody was playing a much more theoretical game is what can you do if you had a perfect machine?
I think what is very exciting and this community should be very relevant to express that is that for the first time, you can just play and empirically try stuff out. So I think that we may see opportunities in being very clever about doing quantum AI.
Is there a chance, though, that this is not going to pan out, and you'll not be able to scale this up?
There's always a chance on that, right. It remains in the level of technological sophistication, as you probably can allude, you know, on a scale of, you know, one to 10 of technical difficulty. This is a 10, right? It's very, very sophisticated system and theory. So, one of the big big battles first how we're going to deal with error Stein hurting quantum machines.
So the holy grail of quantum computing eventually is to build a fault tolerant machine, the universal machine that you can compete in definitely will we get there is a big question. Okay, right. And, and then the second one is, before we get there, what are enough algorithmic advances that can exploit the noisy systems that we have today.
And I really believe passionately that we're going to overcome both. And we're making more progress, nine quantum computing that we've ever made. And you can see the rate of progress. But one big part is to have more and more talented people engaging learning how to program these things. I think there's going to be a whole new feel about quantum developers. And it is the beginning of something very fundamental.
So, you know, all the extraordinary folks are here and there are listening, engage with it, right, there's a way for you to get access to computers and try them. And that's really the path to solve problems.
Is there 10 second answer since we're running out of time, but it's the room here for startups?
Absolutely. You know, absolutely. So we have we launched something called the IBM q network to collaborate and engage and get people access to the hardware stack. And we have over 12 startups right now that are a part of the Q network they operate at different levels of of the stacks and we're working on applications are working and algorithms or noise. So, absolutely, you know, we, you know, we really value that. And it's a growing community as part of the ecosystem we created.
What one final point is that we've shown all the demonstrations and content on a Jupiter notebook. We've made this all available. So you can all get it at qiskit.org and you can get the notebook and all the material that was shown and you can try try stuff out on your own.
All right. Well, thank you for doing the first quantum computing lecture.
I know a first today. Thank you.