Safety & ethics
Explainability is the ability to understand and describe why an AI model produced a particular output, instead of treating it as an unexplained black box.
A neural network gives an answer, but its reasoning is spread across billions of numbers. There is no plain-language logic to read. That makes it a black box. Explainability is the effort to open the box. It tries to say which inputs mattered most, what the model was 'looking at,' and whether its reasons make sense.
Some methods highlight the parts of an image or text that drove a decision. Others build a simpler stand-in model that mimics the big one in a readable way. A newer branch, called interpretability research, looks inside the model itself. It tries to work out what individual neurons and circuits actually represent.
Explainability matters when decisions affect lives, such as medicine, lending, hiring, and criminal justice. Some laws now say people have a right to an explanation for automated decisions. One caution: a chatbot's written 'reasoning' is itself generated text. It may not reflect what really happened inside.
A loan application is declined, and the bank's explainability tool shows that a short credit history, not the applicant's address, was the main factor.