Contestability
The person a decision is about can challenge it, and the challenge can change or overturn it. The route is visible to them.
View glossary entry →Public memory for contestability terms
A dated log of how terms like trustworthy AI and responsible AI drifted from commitments into labels, with stable definitions to cite instead.
Summary
Keep contestability language anchored
A log of when each governance term lost its enforcement mechanism, with a definition to cite in its place.
Takeaways
- Record when a popular term turns into a checklist with no remedy.
- Cite the glossary entries for contestability, accountability, and time transparency.
- Attach the source, so each term can be traced to where it was first written.
Related sections
Why this matters
Language drift hides accountability gaps
A term that loses its evidence requirement stops binding anyone.
Contestability terms travel through policy, procurement, and public narratives. When a term like “responsible AI” becomes a label rather than a measurable commitment, people lose the ability to demand remedies. The definitions in this appendix record what each term committed to when it was written, so a citation carries the commitment, not the label.
- Each term maps to an action someone can be held to, not to an internal checklist.
- A stable definition gives an auditor, a buyer, and a journalist the same thing to check.
- A drift log records the gap between what a term promised and what was delivered.
Timeline
Language drift and memory log
How five accountability terms shifted in policy and practice, 2019 to 2024.
-
2019
Trustworthy AI becomes the umbrella phrase
The OECD principles frame trustworthy AI around transparency, robustness, and accountability. The phrase later came to be used as a compliance label, with no route to a remedy attached.
Source: OECD AI Principles -
2020
Responsible AI consolidates around vendor playbooks
Industry guidance offers responsible AI toolkits, but often names no owner and guarantees no way to challenge a decision. The language moves from shared obligations to internal checklists.
Source: Google responsible AI practices -
2022
Rights framing enters public policy
The White House Blueprint for an AI Bill of Rights centers notice, explanation, and human alternatives. Enforcement stays discretionary. The word “rights” is used, but no review clock is set and no one is named to answer an escalation.
Source: Blueprint for an AI Bill of Rights (White House archive) -
2023
Risk management vocabulary takes hold
The NIST AI RMF sets out governance and accountability functions. Organizations often stop at scoring risk, and publish no evidence that a person can get a decision reversed. The language moves toward accepting risk without the means to remedy it.
Source: NIST AI RMF 1.0 -
2024
Regulatory language converges on transparency and oversight
The EU AI Act emphasizes transparency, human oversight, and documentation. Without a shared baseline for challenges, an implementation can still leave out the outcomes of challenges and the reversals that someone has to carry out.
Source: EU AI Act overview
Explainability for accountability
An explanation names who owns the decision, the evidence it used, and how to get it reversed.
View glossary entry →Appeal passage rate
The share of challenges decided in the person's favor. A rate near zero suggests challenges change nothing. A high rate is evidence the rule is wrong, and should send the rule back to its owner.
View glossary entry →Time transparency
The person can see where their case is, what it is waiting on, how long is left, and what would escalate it.
View glossary entry →Design authority
A named owner can set safeguards and approve reversals.
View glossary entry →Turn memory into enforceable requirements
Cite this page as an appendix in briefs, audits, and public statements.
- Cite the glossary entry, not the bare term, in briefs, audits, and press summaries.
- Put a number next to each term: the appeal passage rate, the review clock, or the remedy log.
- Record each time a term is used as a label with nothing to enforce it.