Things get weird at the atomic scale.
The rules of classical physics governing the objects we can see and touch break down. Particles can occupy two places at once or connect across vast distances, conditions known as superposition and entanglement (or what Albert Einstein dismissively described as “spooky action at a distance.”)
Scientists have explored for decades the theoretical possibilities of applying quantum mechanics to computing. But D-Wave Systems has been working to push the field into the practical realm, using an approach known as “adiabatic quantum computation.” The Burnaby, British Columbia, company, founded in 1999, released what it describes as the first commercial quantum computer in 2010.
Conventional computers deal with binary bits of information, 1s or 0s. But a quantum computer manipulates what are known as qubits (or quantum bits), which can be 1s and 0s at the same time, leveraging the power of superposition. Such a machine depends on entanglement as well, performing many operations on the same data simultaneously.
No one can really say for certain where those helpful distant qubits are operating. The leading bet among physicists is the many-worlds interpretation, which would suggest quantum computers offload processing to parallel universes. (No, seriously — read this!)
D-Wave has signed big-name customers including Lockheed Martin, Google and NASA, but its claims remain controversial, with conflicting reports on whether the machines really leverage superposition and entanglement. Indeed, whether it’s faster than conventional computers at all seems to depend on the problems and algorithms in question.
Earlier this month, D-Wave’s Chief Executive Vern Brownell and DFJ’s Steve Jurvetson, one of the company’s earliest and biggest investors, sat down for an hour-long interview with Re/code. The discussion spanned the critiques of the company, the science of quantum computing and the next steps for D-Wave.
Notably, the latter includes the forthcoming D-Wave processor, which Brownell says will end all doubt that they’ve leaped ahead of classical systems — and will forever leave them behind.
You can watch Brownell and Jurveston discuss the weird world of quantum computing in the video below:
Brownell added that the company’s 70 or so peer-reviewed scientific papers already confirm they’re tapping into quantum mechanics.
If the machines are achieving “quantum speedup” and growth curves keep pace with predictions, Jurvetson believes we could be on the precipice of a fundamental shift in computing — an exponential upon exponential leap that reshapes our assumptions about what machines can do.
The interview that follows has been edited for length and clarity.
Re/code: What kinds of possibilities does quantum computing open up? How can it change the sort of problems that we can apply computing to?
Brownell: We’re at the dawn of this computing age, so things will change over time, and we’ll see a broader and broader set of applications. But today we focus on three problem domains that we think are best suited to this particular type of quantum computing.
Those include machine learning, which is one of the most interesting things going on in computer science today. AI 2.0 and useful AI have really revolutionized the way a lot of folks do things today.
The second thing is the broad set of optimization problems. In logistics, for example, you’re trying to find optimal routing and things like that. They are very complex and scale very quickly with the number of variables and interrelationships you’re trying to optimize for.
And then the third class is what we call sampling. The best example for this is perhaps in financial services, where Monte Carlo simulations represent the largest workloads in most investment banks. It’s used to model things like risk in portfolios — and that’s a fit that works very well with this type of computer.
We’re particularly excited about things like working with DNA-SEQ to find better cancer cures, or doing financial modeling, or, with Lockheed in particular, helping them verify their flight control systems.
But the most important challenge for us is to scale our software capability. The tools aren’t where they need to be; we need to have compilers and higher-level [application programming interfaces] and a full [software development kit] that will allow the technology to be used by a broad set of developers around the world.
Offering this kind of capability in the cloud, so you can access it as you would any other classical resource, will dramatically open up the opportunity here and get a lot of great minds thinking about, “How can I use this new tool that’s unlike anything else in the computing space and help solve human-scale problems?”
Why did DFJ decide to invest in this space — and in this company in particular?
Jurvetson: We look for companies that are unlike anything we’ve ever seen before, with a bold vision to change the world and run by passionate entrepreneurs who get you jumping out of your seat.
This qualifies in spades.
The other part of your question is very easy to answer: When we first invested 12 years ago, it was without a doubt the only company trying to ship a commercial-grade quantum computer, as opposed to just doing research.
Clearly, if we can scale computation into new domains, to outstrip Moore’s law itself, this is very valuable. What’s so very different about quantum computing is, as you add qubits, it’s almost like an exponential on top of an exponential. Every qubit is roughly doubling the power of the computer.
Once the performance of one of these machines meets or exceeds a classical computer, there’s no looking back — and there’s nothing the classical computer industry can do to catch up. I believe that’s unprecedented in business.
There has been some controversy over what D-Wave has achieved, with varying results from various tests. In a paper published in Science in July, a research team at the Swiss Federal Institute of Technology reported: “We found no evidence of quantum speedup when the entire data set is considered, and obtained inconclusive results when comparing subsets of instances on an instance-by-instance basis.” What was your response to that report?
Brownell: That group of researchers basically looked at some benchmarking results that were published, not by us but by a third party, and took that one specific problem and built a system that could perform better than we were able to perform. Or at par with the way our computer could perform for that one specific problem.
It turns out it’s almost a mistake because there shouldn’t be a speedup over classical systems for that particular problem that was benchmarked. Another researcher named [Helmut] Katzgraber [at Texas A&M] proved that you really shouldn’t see a speedup in that kind of problem.
Shortly, we’re pretty confident that you’ll see results that definitely show us scaling better than the best known classical algorithms for those problem sets. We’re starting to see quantum accelerations, if you will, start to take off and cross over what classical systems can do.
So stay tuned for information there. We’ve had a history of knocking down our skeptics.
We’re now on the verge of definitively showing, in this scientific way, [quantum] speedups. I think that’s going to become an historic moment in computer science history.
Jurvetson: Another way to look at it is, what do the customers say?
The one I’ve watched for the longest time is Google. So just look at what they’ve said on their blogs. In 2009, they claimed that in their machine learning applications for recognizing images, the D-Wave System was already outperforming their data center.
Then they reaffirmed that with the D-Wave Two purchase, that system outperformed what classical computers could do. So you could say, “Gosh, maybe that’s the ultimate test for a business, do the dogs like the dog food?”
Every one of D-Wave’s customers has asked to purchase, and in most cases has purchased, more than once. It doesn’t answer the science, of course, but it does answer the question of whether there is value in the market.
So what do you make of Google deciding they’re going to build their own quantum hardware and hiring John Martinis (a professor of physics at the University of California, Santa Barbara) to lead it, who has praised your efforts but also suggested you’re on the wrong path.
Brownell: Uh, I don’t think so …
I can read one quote. Referring to this notion of coherence (the difficulty of keeping qubits in their quantum states long enough to perform computations), he said: “They conjecture you don’t need much coherence to get good performance. All the rest of the scientific community thinks you need to start with coherence in the qubits and then scale up.”
Brownell: What that particular quote is referring to is that there are different ways to build quantum computers and most of the rest of the community is working, at least as of today, on “gate-model” quantum computing, where coherence is very important.
In fact, they haven’t really been able to build any quantum computers because of this coherence issue.
But the adiabatic quantum computer (D-Wave’s approach) is inherently more robust against the decoherence process (see a good explainer on why here). But you’ll also note that that’s probably an older quote. I think John is on the record saying he’s going to be working with Google to do “annealing,” which is the adiabatic style of computing.
So we’re pleased that he’s joined that, I think it’s a validation of the work that we’ve done.
What’s the best guess as to when this will become a more mainstream approach to how computing gets done?
Our business is basically doubling every year in different dimensions.
In parallel universes?
Ha, yeah, right. And even the more pedestrian revenue and financial metrics like number of customers and so on. We believe we will continue on that path and we’ll accelerate that path.
Then layer on that today’s customers buy systems from us and put them in their data centers. Offering this as a cloud service could allow us to provide this as a service to anyone who needs it. My belief is that, in roughly five years, this could be a service that’s ubiquitous.
Developers developing an iOS app may decide, “I want to access quantum resources for these particular parts of my problem” and use those resources as freely as they use classical resources today in the cloud.
I know of Lockheed, Google and NASA, but you said you’re doubling every year. Are there other companies that haven’t been announced?
There are other customers. We have a relationship with the U.S. intelligence community. We can disclose that In-Q-Tel, the investment arm for the CIA, is one of our investors, so there’s interest in activity going on in those spaces.
Sorry, they’re an investor or customer?
Jurvetson: But, it’s kind of like, they often have a customer in mind when they invest.
Can you tell us about your product plans and pipeline?
Our next-generation processor will be 1,000 qubits, actually more precisely 1,152, and that’s going to be released early next year. We already have several customers waiting for that processor and we have about four of those systems in our laboratory today undergoing development and tests.
It not only increases the number of qubits, it also has significant improvements in other important dimensions of performance. So certainly this next processor is going to be very exciting.
To learn more about the weird world of quantum mechanics and computing, check out Google’s video below:
This article originally appeared on Recode.net.