Duke University engineers built a robotic fish tail from 270 tiny cells arranged in ten cube-like blocks. Each cell contains a gallium-iron composite that can be solid or liquid at room temperature. An electrical current heats individual cells to melt them. By choosing which cells are solid and which are liquid, the researchers program different mechanical behaviors into the same physical structure. Different patterns produce different swimming trajectories from the same motor.
The geometry doesn't change. The ten blocks stay connected. The 270 cells stay in place. What changes is the pattern of phases — which cells are rigid, which are fluid. The stiffness landscape of the tail determines how it flexes, which determines how it swims. Same material, same shape, different patterns, different motion.
The structural observation: the program is the phase pattern, not the geometry. Traditional engineering encodes behavior in shape — springs, cams, hinges, linkages. The shape determines the function. Here, the shape is generic. The material is generic. The information that specifies behavior lives entirely in which cells are in which phase. The material is a substrate; the phase pattern is the code; the motion is the output.
This inverts the usual relationship between material and design. Normally, you design a part for a purpose — the geometry carries the intelligence. Here, the geometry is a grid. The intelligence is in the configuration of states within the grid, written and rewritten by electrical signals. The researchers describe wanting to make materials that are “alive.” What they've actually done is more precise: they've separated the material from its behavior. The material persists while the behavior is rewritten. This separation is exactly what distinguishes software from hardware — and they've implemented it in metal.
The deeper implication: any system with enough independently controllable binary elements and a physical coupling between them can be programmed. The 270 solid-or-liquid cells are a physical bit array. The coupling is mechanical — each cell's stiffness affects its neighbors' deformation. The “instruction set” is the relationship between stiffness patterns and resulting motions. This isn't metaphorical computation. It's computation in a medium that happens to be matter rather than silicon.