noun
A black box in artificial intelligence refers to models, especially large language models, where the internal decision-making process is opaque to users. Unlike traditional software, these models don’t follow explicit rules but generate outputs based on patterns learned from vast amounts of data.
The term comes from engineering, where a ‘black box’ describes a device whose internal mechanisms are irrelevant to its function. In AI, this means users interact with the model’s results without understanding how it arrived at them, which can be both a strength (efficiency) and a weakness (lack of transparency).
For professionals, this changes how they trust and debug AI tools. For example, if a language model generates a misleading answer, users can’t easily trace the error back to a specific training flaw. In 2023, a study found that 68% of developers using AI models struggled to explain incorrect outputs, highlighting the challenge of debugging black-box systems.
When an AI chatbot gives a wrong historical fact, users can’t see why it made that mistake, making it harder to correct.
In other languages
- FR boîte noire
- ES caja negra
- DE Blackbox
See also: Model interpretability