Governance, assurance and oversight
The accountability vocabulary - who answers, who checks, who can challenge. Conformity assessment is the EU AI Act gate for high-risk systems before market, and explainability (why an output) is distinct from interpretability (how the model works inside).
This is the accountability vocabulary: who answers, who checks, who can challenge. AI governance is the rules, policies, processes and accountability structures steering responsible AI across its whole life cycle, and Accountability - people and orgs answerable for a system's functioning and impacts - is its anchor.
- AI assurance → the measurable mechanisms (evaluations, audits, certifications, documentation) that demonstrate trustworthiness and compliance.
- AI audit → an independent, systematic review of a system or its governance against defined criteria.
- Oversight → the supervision arrangements, human or institutional, that monitor, review and can intervene.
- Transparency → making information available to stakeholders: a system's existence, capabilities, data, logic and limits.
- Contestability → people's ability to challenge an AI-influenced decision and obtain review or redress.
- Impact assessment → a risk tool assessing benefits, risks and limitations throughout the life cycle.
- Human in the loop (HITL) → keeping a human in the decision path to review, confirm or override.
- Human-centric AI (designed around human needs and dignity) and Trustworthy AI (lawful, ethical and robust) frame the values.
Conformity assessment is the process demonstrating a system meets specified legal or standards requirements - and specifically the EU AI Act's gate for high-risk systems before they reach market. Don't confuse it with an audit (independent review) or impact assessment (risk tool).
Explainability → describing in plain terms why a model produced a particular output. Interpretability → how the model works inside, consistently and predictably. Explainability is the why this output; interpretability is the how it works.