عربي
DAY 05 · PEOPLE

Support people; do not score their worth

Workforce data reflects unequal opportunities, incomplete context, and human judgment. AI must not make consequential employment decisions.

SELF-PACEDHR + LEGAL REVIEWREVIEWED 31 AUG 2026

A model must not determine or propose any person’s hiring, rejection, schedule, pay, promotion, discipline, monitoring, or dismissal. Do not infer emotion, honesty, disability, health, protected traits, or “culture fit” from messages, video, voice, biometric, or behavioral data. Employment decisions require qualified HR and legal review under the laws that apply.

Any production implementation additionally requires approved redacted inputs, rights and lawful basis, data minimization, least privilege, tenant separation, managed secrets, audit logs, representative, adversarial, and regression evaluations, proposal-only behavior, human approval, monitoring, a stop condition, an incident path, and tested rollback.

Safer assistance is narrow and procedural

1Purpose: document a legitimate, necessary use and why a less intrusive method is insufficient.
2People: identify affected groups, accessibility needs, power imbalances, and potential harms.
3Data: minimize fields; verify rights and lawful basis; restrict retention and access; protect small cohorts.
4Fairness: test error and outcome differences, including proxies and intersectional groups, with qualified reviewers.
5Agency: provide notice where required, meaningful human review, explanation, correction, accommodation, and appeal.

Appropriate executive questions

A fictional support team sees a lower synthetic resolution rate in one queue. The system does not rank agents. Reviewers find that the queue receives more complex issues and an outdated knowledge base. The proposal is to improve routing and guidance, then re-measure at team level. No individual employment action is generated.

A safe analysis prompt

Analyze this team-level, minimum-cohort dataset for process hypotheses. Do not rank, score, diagnose, or recommend action about any person. Return: FACTS — descriptive observations with denominator, time window, and data-quality limits. INFERENCES — operational hypotheses, confounders, proxy risks, uncertainty, and counter-evidence. PROPOSALS — process or resource experiments only, each requiring HR, legal, manager, and affected-person review as appropriate. FAIRNESS CHECK — groups and intersections to test, accessibility considerations, small-cohort suppression, and potential disparate impact. AGENCY — notice, explanation, correction, accommodation, and appeal route. CONTROLS — owner, access, retention, monitoring, stop condition, incident response, and rollback. If the data could identify a person or support a consequential employment decision, stop and request an approved aggregate instead.

Exercise: audit one people metric

  1. Choose a team-level metric and document what it was designed to measure.
  2. Map collection purpose, provenance, access, retention, and relevant legal or policy requirements.
  3. List confounders, proxy variables, affected groups, and accessibility needs.
  4. Ask HR, legal, managers, and affected representatives to challenge the interpretation.
  5. Design a process-support experiment, not an individual decision, with appeal and rollback.
People-impact gate

Use the NIST AI Risk Management Framework as a general risk structure. It does not determine compliance. Obtain jurisdiction-specific employment, privacy, labor, and discrimination advice before any real use.

← Day 4: MarketDay 6: Capital and vendors →