The recent bridge collapse in Minnesota happened in large part, people think, because the bridge was of a non-redundant design. Which is to say that if certain elements of the bridge failed, there was no backup, and the whole bridge would fall. Which it did.
We are told by the authorities that modern bridges are not built this way anymore, that they always have built-in redundancy. But redundancy is expensive: You kind of have to build everything twice. What if there were a better way?
Stephen Wolfram, with whom I co-founded the company that makes the Mathematica software, decided to take a look at the problem using the technique of experimental structure generation. You can read about it in his blog post.
The idea is to use simple programs to generate vast numbers of possible bridge designs—say, different possible layouts of struts—and then run a simulation on each one to see how well it performs. Some will be obviously stupid, some will be the same as current designs, but if you're lucky, maybe some of them will be better. They might find a way to spread the effects of a failure in one place out over the whole structure, for example.
This type of exploratory search for designs is the ultimate form of thinking outside the box. The structure-generating program intentionally doesn't know anything about good bridge design: It's going to come up with things that make no sense whatsoever. But experience in other fields shows that sometimes the completely ridiculous idea is the one that turns out to work. —Theodore Gray