Why it matters how we visualise how machines detect cancer

And now for something maximally different from the first recommendation. This is as nerdy as it gets. As you probably know, machines have become very good at spotting cancer, in many cases better than doctors (longtime subscribers might remember this excellent piece). One crucial bit, though: With every diagnosis, there will be uncertainty. And when a doctor reviews the machine’s assessment, she needs to see this, too. So what’s the best way to visualise uncertainty here? It’s complicated.