A team spread across Washington University in St Louis, the Indian Institute of Science, Heidelberg University, Johns Hopkins, and UC Santa Cruz built a computer to solve the kind of problem that stumps today's chatbots, and they describe it in a new paper out of the Indian Institute of Science. As they put it:
Today, AI models may have the capability to write novels and even steer a spacecraft. But give them a logistics network, a microchip to route, or a cryptographic lock, and they stall. These are combinatorial problems.
A combinatorial problem is one where the number of possible answers explodes as the inputs grow — think of every possible order to visit a list of cities, or every way a protein chain can fold up. The team's machine is a brain-inspired computer that leans on quantum tunneling physics to grope through that vast field of competing possibilities and pick out a good one.
The design "does not simply compute a solution," the researchers write — instead, it hunts for the answer, drifting toward stability as a physical system settles into its lowest-energy state. They ran it on protein folding, where a floppy unfolded chain has to find its single most stable shape, and watched it pass through the loose intermediate forms that real proteins move through.
Led by Shantanu Chakrabartty of Washington University, the group argues that brute force has run its course:
The hardest computational problems are not waiting for faster chips — they are waiting for machines that compute in a fundamentally different way.
The work appears in Nature Communications, and the machine runs on ordinary CMOS hardware — the same stuff in everyday electronics — with a mathematical guarantee that it eventually lands on the optimal answer.
Previously:
