Google has a lot of computers. By many accounts, it has more computers than any other company in the world. Yet, even with so much horsepower at their disposal, Google’s researchers keep running into barriers when trying to solve certain complex problems, particularly those tied to artificial intelligence. Google, in effect, has been stumped.
“We have already encountered problems we would like to solve that are unfeasible with conventional computers,” John Giannandrea, a vice president for engineering at Google, said during a press conference on Tuesday. “We want to understand the future that may lie ahead of us in non-conventional computing.”
One type of machine Google has increasingly turned to for help is called a quantum computer. Such systems tap into the seemingly magical properties of quantum mechanics, the field of science that deals with how atoms and other tiny particles work. They can be used to solve problems that traditional computers simply can’t handle.
On Tuesday, Google issued its most optimistic statements to date around the technology, declaring that the still-primitive quantum machines will probably evolve into revolutionary systems for the computing industry and perhaps, for mankind. The event was held on the NASA Ames campus in Mountain View, Calif., where Google is teaming with NASA and D-Wave Systems, a maker of quantum computers, to build a computing lab. Their work has been underway for a couple of years, but only recently—thanks to a larger, upgraded D-Wave machine—have the researchers seen truly promising results from experiments.
Google revealed on Tuesday that recent test calculations show that a D-Wave computer can obliterate the work of a standard computer chip in performing some tasks. In one test, the D-Wave machine needed just a single second to process calculations that would have taken a standard machine 10,000 years to solve. Overall, Google said the quantum machines appeared to perform 100 million times faster on certain problems. Such a speedup would be a true rarity in the history of computing.
Some serious caveats surround these accomplishments, however. D-Wave’s computer is far from a general-purpose machine. It can perform only a limited set of quantum calculations, and just a few people know how to shape problems suitably for the computer. As a result, Google has been relegated to running what amount to test operations on the D-Wave system, rather than the code used in the company’s day-to-day operations. “We need to make it easier to take a practical optimization problem as it occurs on some engineer’s desk,” Hartmut Neven, a director of engineering at Google, said at the event. “We need to make the input into the machine easier. That is not there yet.”
Google is using the tough optimization calculations in some of its advanced AI technology that everyday people touch. (Its photo-search tools and voice-recognition technology are among the most obvious examples.) But those calculations are done on thousands of interlinked traditional computers. The hope is that Google could someday turn to quantum computers to complement its standard systems and come up with more breakthroughs on as-of-yet unsolvable problems. “It may be several years before this kind of work makes a difference to Google products,” said Giannandrea.
The D-Wave machine, which is also being used by NASA with hopes of improving its simulation and encryption technology, relies on what are known as quantum bits, or qubits. Unlike a typical binary digit that must be either a 1 or a zero, a qubit can be a 1, zero, or a state somewhere in between at any moment. It helps to have a degree or two in physics to fully understand how quantum computers work, but the upshot of the technology is that the machines can simultaneously consider an incredible number of possible solutions to a problem. This makes quantum computers well-suited for optimization problems, in which, for example, someone might be trying to find out the best way to route the traffic of thousands of planes going into and out of an airport. It so happens that much of today’s cutting-edge AI software relies on crunching similar sets of these tricky optimization problems.
Neven has spent the most time of any Google employee working with D-Wave machines, and he sees promise for them in areas such as improving battery technology, desalinization machines, and solar cells. The unique qualities of qubits may lend them to uncovering properties about materials, which could result in much more efficient industrial machines. “Because the operating system of nature, as far as we understand it, is quantum physics, you need a process that acts on quantum physics to describe parts of the universe,” Neven said. “Sooner or later, quantum computers will be the tool of choice to solve these problems.”
—With Jack Clark