Citizens Juries on Artificial Intelligence

A I is being used to train computers how to do many things, for example, making medical diagnoses. Scientific studies have shown that AI can diagnose skin cancer from skin images as reliably as dermatologists. But how do we know whether we can trust a diagnosis? If we are given it by a human doctor, we can ask for an explanation. However, the most advanced AI systems are very complex and therefore they may not be able to explain to humans how they reach their decisions. We could require that these systems are always able to give an explanation that people can understand, but doing so may come at the expense of less accurate decisions. So, there is a tradeoff between accuracy and ‘explainability’ of AI systems.