The human brain can still outperform our best algorithms on various tasks. Some tasks, such as object identification, are not really surprising – our brain itself has been optimized by evolution to be quite good at this. But there are other types of issues that are a bit of a surprise, such as some forms of optimization.
You would expect a computer to be reasonably good at finding optimal solutions. But when it came to figuring out the optimal structure of a protein, people playing the FoldIt game managed to beat some of our best software. Now you can add a second task where our brains get ahead: figuring out the best way to perform quantum manipulations. All it took was to turn quantum mechanics into a game.
Algorithms often fall short in optimization problems because of the way they are structured. It is easiest to think of this idea as a landscape with peaks and valleys. The algorithm simply starts by choosing a large number of random locations within this landscape and then tries to move uphill from each of these locations. Once it finds a collection of peaks, it can compare them to find the highest peak it found, which may be the optimal solution.
Unfortunately, it’s also possible that all starting points lead to what are called local maxima: nearby high points that aren’t actually the highest in the landscape.
This is where people can excel. Instead of being stuck with a local maximum, they can intuitively sense that a better solution might be available and use some trial and error to find ways to get there. This is exactly what happened with FoldIt, where people managed to figure out that shaking specific protein structures could return them to a more energetically favorable state.
But protein structures are inherently visual and somewhat intuitive. Quantum mechanics is generally anything but. So how did the new project turn out?
The authors were working on a specific type of problem that could aid in the development of quantum computers. Their system involves a series of atoms held in an optical lattice. It is possible to trap hundreds of atoms in these systems, allowing these systems to scale well. But to perform operations, you have to physically move atoms so that they interact with each other. (Researchers use a form of light called optical tweezers to do this.)
Move the atoms too slowly and the system loses the quantum state you need for the interaction to take place. But move it too fast and you also disrupt the system. “There is a shortest process duration with perfect fidelity, referred to as the quantum rate limit (QSL),,” the authors write, “which imposes a fundamental limit on process duration.” So figuring out the best way to approach the speed limit is a big challenge – and an optimization challenge.
It was also a challenge to turn it into a game. But the authors, from Aarhus University in Denmark, visualized it as a liquid collecting at a low point of a flexible line. Your job as the player was to bend the line in such a way that as much liquid as possible would collect in a target location. Move it too slowly and your score would suffer. Move it too fast and it would slosh everywhere, which also lowers your score. They called the game “Bring Home Water” and turned it into an application called Quantum Moves, which is available for download (it’s also on the App Store and Google Play).
https://www.youtube.com/watch?v=3U-hzGgUsHY
A little Bring Home Water in action.
When you grab and bend the surface, you are actually doing the same thing as moving the optical tweezers. In fact, the equation is so direct that the results could tell researchers where to place the tweezer relative to the atom to perform the equivalent manipulations. The liquid is just the quantum state of the atom, and moving it too fast introduces excitations into its wavefunction.
Once again, some players beat our best algorithms, as they were willing to sacrifice a bit of stability to get a good time. Overall, they found similar solutions to those of the algorithms in far fewer tries.
So the authors came up with a hybrid solution. Instead of providing their optimization software with arbitrary starting points, they posted it with some of the best solutions described by the readers. The resulting analysis returned a value for the quantum rate limit that was less than 70 percent of the value produced by the algorithms alone. In other words, it shaved more than 30 percent of the time required.
The authors were so impressed with the results that they already have another game available on their website, and they plan to expand the approach to a number of other issues.
Nature2015. DOI: 10.1038/nature17620 (About DOIs).