IBM's Dynamic Duo: Jerry Chow and Jay Gambetta Discuss All Things IBM Q

Jerry Chow and Jay Gambetta Chat with Whurley about QISkit,  Artificial Intelligence, and the Myths and Misconceptions about Quantum Computing

When I visited Germany a few months ago for the BlueYard event, I had the pleasure of meeting Jerry Chow. Jerry is the Manager of Experimental Quantum Computing for IBM. He’s also super friendly and easy to talk to. Jerry introduced me to his cohort Jay Gambetta, another awesome dude who just happens to be the Manager of Quantum Theory, Software, and Applications at IBM. A couple of weeks ago I reached out to see if they would be willing to do an interview. To my surprise, they both said yes! So let me get out of my own way. Here’s superposition.com’s first double interview with two titans in the field.

How does quantum computing compare to classical computing? Will quantum computers replace classical computers?

Chow: You can think of quantum computing as computing with the full force of nature. Processing information as is governed by the natural laws of physics, specifically quantum mechanics. In some sense, classical computing is just a subset of quantum computation. You can think of a classical computer as a handicapped quantum computer, because you are limited to the computational space of classical information as bits.

In the near term, quantum computing will not replace classical computers completely, since the classical computers we build are already very good and useful. Rather, they will likely work in concert with one another. We will likely use quantum computers for specific classes of problems which are too unwieldy to be performed on classical computers. These problems themselves might be contained within larger problems, which would be broken into parts that are better solved on quantum, and others better solved on classical.

How would you explain to a layman the difference between a quantum computer that uses the gate model to solve problems and one that’s using quantum annealing?

Chow: The circuit model for universal quantum computation makes use of quantum controls to manipulate the states of qubits to explore an exponentially large computational space. We often hear about the ideas of quantum superposition and entanglement with regards to quantum computing. It is through this circuit model with controls (or gates) applied to the qubits that it is possible to fully make use of this space.

By designing the appropriate algorithm, we can find routes for quantum speed-up. Within the circuit model for universal quantum computing, one can run the known canonical quantum algorithms with speed-up over classical computers—Shor’s algorithm for factoring large numbers, Grover’s search algorithm, and others.

However, to actually run these universal algorithms, the circuit model quantum computer needs to also be fault-tolerant and have built-in quantum error correction. Otherwise, the physical implementations of qubits which comprise a quantum processor suffer from decoherence and errors, which result in the loss of quantum information (and thus the inability to make use of entanglement and an exponentially large computational space).

Implementing quantum error correction for fault tolerance results in the need for a much larger overhead of qubits, pushing into potentially the hundreds of thousands, if not millions of qubits. Even with the holy grail of a fault-tolerant quantum computer far off in our sights, there is real value in finding applications on near-term quantum processors of just a few hundred qubits. We might find real quantum advantage in applications areas such as quantum chemistry, optimization, and machine learning.

The contrast of universal quantum computing to quantum annealing is that quantum annealing addresses a single class of graph problem that is embedded on the physical device hardware. It is not a general-purpose quantum computer on a potential path towards a fault-tolerant universal quantum computer.

What are the biggest engineering challenges in quantum computers?

Chow: All physical implementations of quantum computers require physical qubits. Qubits come in many different shapes and forms, ranging from trapped-ion or atomic systems, to photons, to semiconductor spins, to integrated superconducting circuits. Our approach at IBM is to work with superconducting circuits based on small Josephson-junctions.

The biggest engineering challenges for many of these approaches is to balance coherence (how long the quantum information lives in the qubits), controllability (how easy it is to address and perform operations on the qubits), and connectivity (how to scale up and interconnect a large number of qubits).

For quantum systems, these three challenges often compete with one another. With regards to our superconducting qubits approach, advancement along all three of these dimensions requires improving the core materials and processing, the integration methods for microwave hygiene, and low-noise room temperature electronic control systems.

On top of that, a significant challenge to making use of quantum computers is the enablement of a software stack. We started the open-source Quantum Information Software Kit (QISKit) precisely for that reason. Pairing it with the back ends we make available through the IBM Q experience, we enable anyone anywhere in the world to program and run a quantum computer.

Can quantum computing help solve the Moore’s law problem?

Gambetta: People often like to discuss quantum computing in the context of Moore’s law. We actually like to think beyond that. Calling it a solution to the Moore’s law problem makes it seem like quantum computing is simply another node along the developmental path. Rather, quantum information is an entirely different paradigm from classical information, so quantum computing really is different altogether. It’s not a matter of being faster than existing computers. It’s more a matter of being different, and way more efficient for a particular set of extremely difficult problems.

IBM recently announced a 50-qubit quantum computer. It seems like 50-qubits is where we start hearing talk of quantum supremacy. Why is that such a significant number? What is a machine like this capable of?

Gambetta: The other talking point that is making the rounds when it comes to quantum computing is this concept of quantum supremacy. The popular discussion around the phrase emphasizes a single point at which quantum computers are suddenly capable of outperforming classical computers, and that it’ll happen around 50 qubits. But to us, it feels a bit naive to try and pinpoint this one moment. For one thing, it’s not simply about the number of qubits, but in fact the depth of the quantum circuit you can run. Both contribute to forming a measure of a quantum computer’s performance, which we’ve started calling quantum volume.

On the other hand, we expect that over the near term we will be seeing more and more demonstrations of quantum advantage. There will be cases where quantum computers outperform classical ones, and we are most excited in getting to cases with real-world applications. In some sense right now, the whole world should be excited that it is in a quantum-ready phase. We are all developing the tools and infrastructure to, in the not-too-distant future, reach this realm of commercial quantum advantage.

What trends do you see in high performance computing that will interact with quantum computing?

Chow: HPC is of course at the heart of many of the most challenging problems, such as optimization and chemistry. In a lot of these areas where the solution spaces become too large as the problem size grows, HPC will run out of steam. It’s in those cases that there will be a need to fill that space with a quantum computing solution. The interaction between HPCs and quantum computers, therefore, is an important one to define for near-term application. This is why we will need to properly define the software stack and interfaces, and the reason for a particular partnership that our team is engaged in with Oak Ridge National Laboratories and the Quantum Computing Testbed Pathfinder program.

How will quantum computing change, influence, or incorporate technologies like A.I., machine learning, big data, and the cloud?

Gambetta: AI and quantum computing are two technological buzzwords thrown around so much that you think, “There has to be a connection, right?” Well we certainly are intrigued by the possibilities, especially with regards to AI for quantum computing, and then also conversely, quantum computing for AI. On the first front, we are actively exploring how there might be ways to speed things up like characterization or calibration of quantum computers through leveraging machine-learning techniques. On the latter, figuring out the learning and large feature spaces in the quantum realm and whether there might be novel paths to take.

We are really excited about these problems, and, through our recent partnership to form the MIT-IBM Watson AI Lab, are undertaking a collaborative research effort on this front with top talented students and faculty. Together, IBM and MIT scientists are investigating the “physics of AI,” which involves new research into AI hardware materials, devices, and architectures, including quantum.

How will the way we currently think about solving problems with a computer have to change?

Chow: This is a great question that will be answered in this new age of learning how to program on a quantum computer. Depending on the interface between classical systems and quantum computers, the software stack needs to be arranged in such a way that parts of problems can be isolated and ferried out to either the classical part or the quantum part.

How can developers get involved with quantum computing today?

Gambetta: We like to define two types of developers: q-developers and your traditional developers. Our open source software kit, QISKit, already targets the q-developers. These are primarily comprised of research scientists that want to develop and developers that want to learn quantum. They are making tools that use quantum circuits, including transpilers, mappers, schedulers. Starting now, we are also extending QISKit to make verticals that traditional developers who do not know much about quantum will be able to integrate into other projects. This is going to be an exciting time.

What’s the response to QISkit been like so far? Can you give a few examples of projects that use QISkit?

Chow: We have all been super excited by QISKit and being able to work on this open source platform. On www.qiskit.org you can see the number of external researchers that are all contributing to it. It’s opened up a different way of enabling research and providing open access tools to truly accelerate a technology. Prof. James Wootton from U. Basel has used QISKit both for his research to study a repetition code, as well as to enable the generation of “quantum” emojis for fun.

This is also just the beginning for us. We have planned a road map to keep peeling back layers of the onion and enable more functionality and access to things like pulse-shaping. Definitely stay tuned, but also get involved. Feedback and input from the community really help drive this.

What are some of the most prominent myths or misconceptions about quantum computing?

Chow: One, that quantum computers are just another point on the Moore’s law. Two, that quantum computers are generically faster than classical computers. Three, that quantum computers that will crack all your passwords are just around the corner.

Gambetta: Four, that quantum computers compute “all solutions in parallel.”

What problems do you run into with software simulations of quantum computers that are running on traditional computers?

Chow: Understanding noise and cross talk, things that can lead to error. We often simulate a quantum computer, but include real physical noise and error models so that we can accurately compare with real quantum systems.

What applications of quantum computing are you most excited for?

Chow: In the near term, certainly getting to solve problems in quantum chemistry. This hearkens back, of course, to Feynman’s statement about simulating quantum mechanics. But there’s a fundamental “harnessing nature” aspect of this problem that just makes it so cool. Use and program a quantum system to understand a natural quantum system.

Was there anyone whose work inspired you to pursue a life in the sciences? Is there someone working in quantum today that inspires you?

Chow: My father was a physics professor, which started me down this path. Within the field, I’m fortunate to have developed most of my fundamental experimental know-how through working on top research teams led by Charlie Marcus (back when I was an undergrad), and then of course Rob Schoelkopf as a grad student.

Gambetta: I have a lot of respect for many people working in quantum. Too many to name. The two that have had the biggest impact on me are My Ph.D. supervisor, Howard Wiseman, and my post-doc adviser, Steven Girvin.

Correction: December 5, 2017

Jay Gambetta’s role with IBM is Manager of Quantum Theory, Software, and Applications; not Manager of the Theory of Quantum Computing and Information.