Search giant Google is a data-collection empire, and while in many countries that has raised concerns over privacy, Google’s relentless collection of information has led to all sorts of internal problems over how to identify and categorize patterns in the data glut. One such quandary is called computer vision, the process of programming machines to recognize objects in images, something conventional computers don’t do very well. In 2007, one of Google’s top scientists, Hartmut Neven, called Geordie Rose, a UBC physics PhD and co-founder of D-Wave Systems Inc., in search of a solution.
Based in Burnaby, B.C., D-Wave is the world’s first commercial quantum computing company, and Neven called Rose, the operation’s chief technology officer, because quantum computers have important strengths conventional computers lack. D-Wave’s machine, for instance, has a talent for computer vision. Collaborating with D-Wave, Google researchers developed an algorithm for teaching the quantum computer to recognize cars in images. The process is complicated, Rose says, but the product is simple: “There’s a bunch of pictures with cars and a bunch of pictures without cars, and we threw ’em in the blender and out came a piece of software that, when given a new picture, will determine whether or not there’s a car in it.”
The D-Wave system’s capacity for acquiring new pattern-recognition capabilities demonstrates successful “machine learning,” one of the holy grails of computer science. It didn’t grab headlines the way Jeopardy! champion Ken Jennings’s recent loss to IBM’s Watson did, but it’s an achievement that may very well have quietly ushered in a new era of computing.
Quantum computers use quantum bits (qubits), which behave according to the rules of quantum mechanics, to solve problems. While the bits that conventional transistor-based computers use have only two states (on or off, one or zero), qubits can register information in more complex ways, enabling the machine to perform tasks that would confound even the most powerful traditional computers. D-Wave’s approach uses a process called “quantum annealing” that the company describes as harnessing “a fundamental principle of nature in both quantum and classical regimes, the propensity for all physical systems to minimize their free energy.” Beyond this explanation, the workings of the machine—a 37-cubic-metre black box in which a thumbnail-sized superconducting chip is cooled to within 0.01 degrees of absolute zero—are shielded from the layperson’s understanding by the opacities of quantum theory.
Exactly what principles of quantum physics are exhibited in the interaction between the chip’s 128 qubits, and to what extent they are exhibited, will remain fodder for conferences and blogs for decades to come. What’s important now is that it does, indeed, work. Twelve years after its founding, D-Wave is living up to its tag line “The quantum computing company.” With its November 2010 sale of a D-Wave One system to American defence behemoth Lockheed Martin, a deal worth an estimated US$10 million that includes support staff and maintenance, the company has silenced its critics.
Like Google, Lockheed Martin has big data problems. According to spokesman Thad Madden, one of the most costly and labour-intensive parts of the process of developing large, complex systems, such as those that run modern ships and aircraft, is testing the billions of different scenarios the systems might encounter. Madden says the D-Wave One “can significantly reduce these costs and offer the potential for significant savings in cost avoidance and schedule delays.”
The size of the opportunity for D-Wave is, according to Rose, more a function of the research and development budgets of Fortune 500 companies than the market for supercomputers. From pharmaceuticals to finance to defence, the potential applications for the D-Wave One’s capabilities test even its creator’s imagination. He claims that the big problems these companies face have, until now, not been considered solvable by computers: “When you start to think about what quantum computers might be used for in the future, and particularly in these artificial-intelligence type things like learning, some of these problems become computing problems.” It’s going to take both vision and learning, it seems, just to figure out what problem to solve next.
