Il prossimoProssima guida
Sfumature di genere e audit sui pregiudizi del riconoscimento facciale
Società
GUIDA della Società
An AI audit is a structured evaluation of an AI system, or the organization running it, against defined criteria such as accuracy, fairness, safety, legal requirements or governance standards, carried out internally, by an independent third party, or under a regulator's mandate.
Ascolta questa guidaNarrato dall'IA · ~3 min
Third-party assurance matters because claims about AI systems are hard for outsiders to verify, and audits are becoming a main way regulators, buyers and the public check them.
AI audits differ in who does them and what they examine. Internal audits are run by the organization itself, ideally by a team separate from the builders; Raji and colleagues' 2020 framework proposed an end-to-end internal audit process across a system's development. External audits come from outsiders. Some are cooperative, with access to data and code; others are adversarial, like journalists or researchers testing a system through its public interface. Regulatory audits are required or performed under law. Methods fall into three broad groups. Governance or process audits check policies, documentation, roles and risk management. Technical audits test the model: performance across subgroups, robustness, security, and behaviour on edge cases, sometimes including red teaming. Outcome audits look at real-world effects, for example comparing decisions across groups or using sock-puppet accounts to probe a recommendation system. Conformity assessment is a related legal concept. Under the EU AI Act, providers of high-risk systems must show compliance before placing them on the market. For most standalone high-risk uses this is an internal control procedure, while certain biometric systems and AI in products already covered by EU safety laws may involve a notified body. New York City's Local Law 144 requires independent bias audits for automated employment decision tools, and the EU Digital Services Act requires independent audits of very large online platforms. Independence is the persistent weakness. Auditors are usually paid by the company being audited, may receive limited access, and often work without agreed standards for what counts as passing. Critics warn of audit-washing, where a narrow audit is presented as a clean bill of health. A common misconception is that an audit certifies a system as safe or fair in general; it only speaks to the criteria, scope and time period examined.
An employer in New York City using an automated resume screening tool commissions an independent bias audit that reports selection rate impact ratios by sex and race or ethnicity categories, as the city's Local Law 144 requires.
Researchers query commercial face analysis services with a balanced set of faces and publish error rates by skin type and gender, an external audit carried out without the vendors' cooperation, like the 2018 Gender Shades study.
A software company seeks certification of its AI management system against ISO/IEC 42001 from an accredited certification body so enterprise customers can see its governance controls have been checked.
A very large online platform in the EU undergoes an annual independent audit of its risk management, including its recommender systems, under the Digital Services Act.
An AI assurance industry is forming, including accounting firms, specialist startups, testing labs and certification bodies. Standards work, such as ISO/IEC 42001 for management systems and related standards for bodies that certify them, aims to make audits more comparable. Governments including the UK have published plans to grow AI assurance as a market. Key unresolved issues are auditor accreditation, access to model internals for external researchers, and who pays without creating conflicts of interest. Audits of general-purpose models are especially immature, since their uses are open-ended. Expect gradual professionalisation rather than a single agreed method in the near term.
Newsletter gratuita
Tre storie verificate sull'IA ogni mattina nei giorni feriali, scritte in un inglese semplice. Gratuito per sempre, senza pubblicità.
Un'email ogni giorno feriale. Disiscriviti con un clic. Non vendiamo né condividiamo mai il tuo indirizzo.
Mettiti alla prova
Feedback immediato su ogni risposta e un certificato condivisibile con un ID verificabile una volta superato un corso.
Sostieni l'istruzione gratuita sull'IA. AI Understanding è un'organizzazione non profit 501(c)(3) — nessun annuncio, nessun paywall, mai. Fai una donazione
An AI audit is a structured evaluation of an AI system, or the organization running it, against defined criteria such as accuracy, fairness, safety, legal requirements or governance standards, carried out internally, by an independent third party, or under a regulator's mandate. Third-party assurance matters because claims about AI systems are hard for outsiders to verify, and audits are becoming a main way regulators, buyers and the public check them.
Gli audit esterni contraddittori vengono eseguiti da soggetti esterni, come giornalisti o ricercatori, che sondano un sistema senza cooperazione o accesso privilegiato.
Gli audit dei processi verificano il modo in cui l’organizzazione gestisce l’intelligenza artificiale, anziché testare direttamente il modello.
La legge locale 144 richiede controlli indipendenti sui pregiudizi, compresi i rapporti di impatto per categorie demografiche.
Il rapporto di impatto confronta il tasso di selezione di ciascuna categoria con quello della categoria più selezionata.
La maggior parte degli usi autonomi ad alto rischio seguono il controllo interno, mentre alcuni sistemi biometrici e prodotti regolamentati possono coinvolgere un organismo notificato.
Continua a imparare
Altre guide selezionate per questo argomento
Il prossimoProssima guida
Sfumature di genere e audit sui pregiudizi del riconoscimento facciale
Società