Modern machine learning is great for helping scientists sort through huge, unwieldy data sets. But it’s less useful for things that require inference or reasoning – both vital to the scientific process.

One group of scientists are now trying to fix this problem with a completely new kind of machine learning. This new approach aims to find the underlying algorithmic models that interact and generate data, to help scientists uncover the dynamics of cause and effect. This could aid researchers across a huge range of scientific fields, such as cell biology and genetics, answering the kind of questions that typical machine learning is not designed for.