Guides / Glossary
AI governance, in plain words
The terms that come up in AI law, standards and board papers, each defined in a sentence or two, with a link to where we say more.
A
- Accountable person
- The named individual who answers for an automated decision system's outcomes and has the authority to override it. A person, not a committee or a department. Naming one is the core of the AIC standard. The AIC standard
- Accreditation
- Formal recognition by a national accreditation body that a certification body is competent to certify against a given standard. In South Africa that body is SANAS. AIC is not yet accredited and says so on its disclosures page. AIC's accreditation status
- AI management system
- The policies, roles and processes an organisation uses to govern its AI, of the kind ISO/IEC 42001 specifies. It describes how AI is managed, not whether a particular decision was fair. ISO/IEC 42001 in South Africa
- AI registeralso AI inventory
- A current list of the AI systems an organisation uses, with what each decides, who is accountable for it and how it is overseen. Boards cannot oversee what they have not listed. King V and AI
- Algorithmic Rights
- The five rights AIC's standard is built on: human agency, explanation, empathy, correction and truth. Together they mean a person can reach a human, understand a decision, be treated fairly, have a wrong decision put right and know when AI is involved. The Declaration of Algorithmic Rights
- Automated decisionalso Automated decision-making, ADM
- A decision made by a system without a person meaningfully considering it before it takes effect. Under POPIA section 71 the concern is a decision with legal or substantial effect, based solely on automated processing, that profiles the person. POPIA section 71 explained
C
- Certification body
- An organisation that assesses others against a published standard and certifies those that meet it. Its independence from what it certifies is what gives a certificate its value, which is why AIC does not consult on systems it certifies. AIC's impartiality
- Conformity assessment
- Under the EU AI Act, the process of showing a high-risk AI system meets the Act's requirements before it is placed on the market or put into service. The EU AI Act and South African companies
- Contestabilityalso Right to contest
- Whether a person affected by a decision can find a route to challenge it, have a person reconsider it and see the outcome change when they are right.
D
- Data subject
- Under POPIA, the person to whom personal information relates. Section 71 protects data subjects from decisions based solely on automated processing.
- Deployer
- Under the EU AI Act, an organisation that uses an AI system under its own authority, as opposed to the provider who develops it or places it on the market. Deployers of high-risk systems carry their own duties, including human oversight.
- Disparate impact
- When a decision process produces materially worse outcomes for one group than another, whatever the intent. Often measured with a ratio of selection rates; the four-fifths rule treats a ratio below 0.8 as a warning sign.
- Drift
- A change over time in the data a model sees, or in how its outputs relate to reality, so that a system that was accurate and fair when tested stops being so. One reason point-in-time audits are not enough.
E
- Explainability
- Whether the reason for a particular decision can be given in plain language to the person affected, and whether that reason is the one that actually drove the decision.
H
- High-risk AI system
- A category in the EU AI Act covering systems used in areas such as employment, credit, education, essential services and law enforcement, which carry the Act's heaviest obligations. The EU AI Act and South African companies
- Human in the loopalso HITL, Human oversight
- A person who reviews and can change a decision before it takes effect. It only counts if the person has the information, the time and the authority to disagree; signing off every output unread is not oversight.
I
- Information Regulator
- South Africa's independent regulator for POPIA and the Promotion of Access to Information Act. It investigates complaints, issues enforcement notices and can impose administrative fines. AI governance in South Africa
K
- King V
- The King Code on Corporate Governance for South Africa, the successor to King IV, which places the governance of technology, including AI, with the governing body. King V and AI
O
- Override
- A person changing or stopping an automated decision. A working override that is actually used, with reasons recorded, is the clearest evidence that human oversight is real.
P
- POPIAalso Protection of Personal Information Act
- The Protection of Personal Information Act 4 of 2013, South Africa's data protection law. Section 71 governs automated decision-making. POPIA section 71 explained
- Profiling
- Automated processing that evaluates aspects of a person, such as their reliability, creditworthiness, health, behaviour or likely performance.
- Proxy variable
- An input that is not a protected characteristic but stands in for one, such as a postcode that tracks race. Removing the protected attribute does not remove the bias if a proxy remains.
S
- Self-assessment
- An organisation's own review of itself against a standard, without independent verification. Useful to find gaps; not a certificate. AIC Aware is AIC's free self-assessment. AIC Aware
T
- Transparency notice
- Telling a person, specifically and before a decision affects them, that AI is involved and how they can ask for a person or an explanation. A generic clause in terms and conditions usually falls short.