IF ANYONE can build quantum computers big enough, they could have a valuable talent that鈥檚 long been sought by artificial intelligence researchers. A new algorithm developed by physicists shows they should also be able to spot patterns in apparently random noise at undreamed-of speeds.
Today鈥檚 computers process data in the form of voltages representing 1s and 0s. A quantum computer, however, exploits the bizarre phenomenon of 鈥渟uperposition鈥�, in which particles spin opposite ways simultaneously, to make so-called 鈥渜uantum bits鈥�, or qubits. These are essentially on and off at the same time, allowing them to be used in two calculations simultaneously. As the number of qubits increases, performance speeds rise exponentially, so a 200-qubit computer would be able to run more calculations simultaneously than there are atoms in the Sun.
But so far, the world鈥檚 most advanced quantum computer is IBM鈥檚 7-qubit device, which is no more powerful than a pocket calculator. Researchers hope that bigger ones will be able to factor huge numbers 鈥� the basis of many secret codes 鈥� in seconds, beating standard computers by years.
Advertisement
But aside from code-breaking, there have been surprisingly few useful applications mooted for quantum computers. Now Ralf Sch眉tzhold, a theoretical physicist at the University of British Columbia in Vancouver, has worked out a way to make quantum computers pick out a pattern from otherwise random data far faster than standard computers.
Pattern recognition is a cornerstone of AI research. While people can see at a glance that there is a chequerboard region in the dots below, today鈥檚 computers struggle to do this. 鈥淎 big product of human intelligence is recognising patterns,鈥� Sch眉tzhold says.
When a classical computer tries to find patterns, it runs what鈥檚 called a Fourier transform, which describes the image in terms of its frequency components. But this is a slow 鈥渟erial鈥� process: the program has to finish analysing one piece of data before moving to the next.
Sch眉tzhold鈥檚 new algorithm exploits a quantum computer鈥檚 ability to do many things at once to seek out parallel lines among random collections of dots. By using qubits to hold large amounts of image information in superposition, it would allow a quantum Fourier transform to examine big chunks of images in one fleeting glance, making the whole process much faster.
One application of the algorithm, suggests Seth Lloyd at Massachusetts Institute of Technology, might be in military aerial surveillance. This could allow patterns to be found in grainy images that are light on data. That would be a big advantage if, for example, you have a plane that鈥檚 flying over a battlefield and you only have a few seconds to pick out a tank.
