Machines might work better when the parts don't match
Here is something that goes against what I thought I knew.
If you have a network of things that need to work together, like generators on a power grid, oscillators in a material, or neurons firing, the standard assumption has been that they work best when they are as identical as possible. Matched parts sync up. Mismatched parts fight each other. That is how I would have explained it to anyone who asked.
A group at Northwestern led by physicist Adilson Motter published a paper in Science on September 17 saying that is often wrong. They built a mathematical framework that tells you when differences between the parts, which they call disorder or heterogeneity, actually make the whole system more stable rather than less.
And this is not just theory looking for an application. They point to real cases where it already shows up: power grids, engineered metamaterials, computation in the brain. The framework is a way to work out in advance how much variation helps, and what kind.
What I like about this is the second half of it. It also gives a reason why nature is so messy. Neural, biological and ecological networks are full of variation, and we usually treat that as noise the system has to cope with. This suggests some of it is doing actual work.
The practical version is that an engineer could deliberately design parts to differ, instead of spending money to make them match. That is a strange idea to get used to, but the math is there now.