Friendly artificial intelligence (FAI) AI Safety; Human-centered AI
noun phras
Definition: A hypothetical form of advanced AI designed to have a beneficial, or at least non-harmful, impact on humanity by remaining aligned with human interests and values; historically, the term also denotes the research program concerned with designing such systems. The term is strongly associated with Yudkowsky’s early work on safe AGI, but in current scholarship it is often treated as a historically important precursor to the broader field of AI alignment rather than the dominant contemporary label [Yudkowsky 2001].
Example in context: “One obvious suggestion has been that any AI which achieves or approaches human-level intelligence should be programmed as ‘Friendly AI’ …” [Li 2021].
Synonyms: friendly AI; FAI
Related terms: AI safety; AI alignment; beneficial AI; value alignment