All models are wrong, but some are harmful

A key challenge in building AI systems is creating a mathematical definition of the subject of interest. This causes a problem because no mathematical definition can capture the full scope of the natural world. This is further complicated by the fact that these definitions are created by teams of AI developers and will inevitably reflect the attitudes of these developers. This issue affects all AI systems. But that doesn't mean we shouldn't build any models for anything. A model can bring a benefit, even if it isn't perfect.