Explainable AI (XAI)
Also known as: XAI
What is Explainable AI (XAI)?
A set of methods that make an AI model's output understandable to people: why it reached a decision and which factors weighed most. It turns a black box into something you can question and defend. This matters most where decisions carry real consequences, like credit, hiring or healthcare, and where the law expects a reason.
Why it matters
Two things make explainability harder than it sounds. First, an explanation can be convincing and still be wrong. A plausible story about why a model decided something is not proof that this is what actually drove it, and a confident but misleading explanation is worse than none. Second, different people need different explanations. A regulator wants to know a decision was lawful, a rejected customer wants to know why, an engineer wants to find the fault, and one output rarely serves all three. There is a cost as well. The models that explain themselves most readily are often not the most accurate, so buying explainability can mean giving up some performance, and that trade should be a deliberate choice rather than an accident.
In practice
A hospital weighing a model to flag patients for earlier review leans towards a simpler, more explainable one over a marginally more accurate black box, because a clinician has to understand why a patient was flagged before acting on it, and be able to justify that afterwards. The small loss in raw accuracy buys something worth more here: decisions people can question, defend, and correct when they turn out wrong.
