Quite recently, I had the chance to discuss with Philippe de Reffye, a French researcher who spent his life studying how plants grow. He built a generic mathematical model that can simulate the growth of any plant, assuming we can mesure some environmental parameters to calibrate the model. He wrote a very complete book about it (in French).

The interesting point from my perspective is that his work managed to lower the computational complexity of simulating plant growth from an exponential to a linear function. What would take days to calculate can now be done in seconds, and it grows linear w.r.t. the plant complexity.

Philippe and I talked about AI. His opinion is that AI can help calibrating the models. However, without the maths, you cannot really understands what happens when the plants grow and you have no chance finding an optimization of the kind he found.

Philippe warned me about the danger of replacing maths with ML when we don’t have a model. Indeed, it is a potential workaround, but it is expensive in computation time and it blurs our understanding of the world.