Black Box Model

A machine-learning model whose internal workings are not directly inspectable or interpretable by a human, either because the model is architecturally complex (deep neural networks, large language models) or because it is proprietary and the developer does not disclose its internals. Most assistive AI products deployed today — object recognition, speech captioning, text simplification — are black boxes in both senses, which is the motivation for explainable AI. Accessibility practitioners should recognise that black-box operation compounds exclusion: disabled users cannot easily verify outputs, cannot contest errors, and often have no recourse when automated decisions about accommodation or eligibility go wrong.