Technical documentation of an AI system is a set of materials that describes what the system does, how it was built, and how its reliability was verified. For regulated deployments, it is mandatory and serves as evidence towards authorities and customers alike. It contains the purpose and intended use, the architecture and components used, including third-party models, the origin and nature of the training and testing data, the method of evaluating accuracy and bias, known limitations and scenarios in which the system fails, measures for human oversight, and a record of versions and changes. When purchasing a ready-made solution, most of the documentation is supplied by the vendor, but responsibility for its existence lies with whoever places the system on the market under their own name. Request it before signing the contract, not only at the time of an audit.
See also: Provider and deployer, High-risk AI system, AI management system.
