The OECD and EU AI Act definitions of an AI system state that, having been deployed, such a system may exhibit adaptiveness. In conceptual terms, what does adaptiveness describe within these definitions?
- AThe system can run on a range of hardware platforms, from a handheld device to a data-centre server, without its code being rewritten.
- BThe system can accept several input formats, such as text and images together, and reason across them to produce a single output.
- CThe system always produces the identical output when given identical input, so its behaviour stays consistent across every run.
- DThe system may continue to change its behaviour after deployment, for instance by learning from new data it encounters in use. Correct
Why A is wrong: Portability across hardware sounds like a flexible, adaptive quality, but it concerns deployment engineering rather than how the system behaves, and the definitions tie adaptiveness to changing behaviour, not changing platforms.
Why B is wrong: Handling multiple input formats describes multimodality, which is a separate concept, so confusing it with adaptiveness is tempting but the definitions reserve adaptiveness for post-deployment behavioural change.
Why C is wrong: Identical outputs for identical inputs describe determinism, which is closer to the opposite of adaptiveness, so this inverts the concept the definitions intend.
Why D is correct: The definitions describe adaptiveness as the capacity for a system, after deployment, to alter its behaviour through ongoing learning, which is why both phrase it as a possible rather than guaranteed property.