The past week was dominated by stories about how artificial intelligence has finally, finally made significant inroads in the chemical sciences. We have been hearing a lot about this for the past five years, but so far AI has mainly responded with “about as much as you can expect from what’s been available for several decades.” It’s fantastic that people are talking up our progress, but the claims come with some holes in them.
That didn’t stop scientists, academia, investors, and the scientific press from going wild — the Telegraph put it simply, “This is the coolest thing you will read all week.”
Still, it’s hard to be upset about the progress.
There are two important reasons for that. First, advances in AI are worth celebrating: They represent the use of computers to solve problems in new and better ways. It’s the triumph of applied complexity theory. It also fits in with an even bigger momentum in the chemical sciences, as new discoveries are beginning to radically redefine the science.
But the technologies also represent a second important feature of AI, also worth celebrating: As well as exploiting complexity theory, AI seems to contribute an important form of creativity. Just as computer engineering tools can predict complex outcomes, AI can produce results similar to those predicted: Just ask The Daily Beast.
Here’s a telling example of one of these amazing results. Here’s The Daily Beast, reporting on the challenge of solving a longstanding riddle: How do you measure the complexity of a molecule based on its surface? You can get several estimates, but none can perfectly fit the puzzle, no matter how you measure it. To find out what the closest approximation was, you first need to figure out what is impossible to measure from either extreme. Then you throw all these guesses together, see which are the closest. This is a pretty typical self-referential exercise, similar to questionnaires, but much, much faster.
Because scientists at the Fourth National Coding Conference, which has been gathering for the past five years, were flying in for the last week, they were able to work out a method, just like that: