AI Ethics and Responsible Use
Definition
PMBOK 8 (Appendix X3.3) pairs AI’s benefits (shifting daily work from repetitive tasks toward higher-level, creative work) with risks that humans must weigh. The ethical factors:
- Bias — systems trained on biased data (or with biased algorithms) reproduce bias. Mitigations: diversify training data sets, run periodic bias-focused tests, involve different teams in developing the system.
- Privacy — AI uses large, often sensitive and regulated data sets; data must be secured and collected under a privacy policy.
- Accountability — AI may drive certain decisions, but ultimately a human must be accountable for each decision, and that accountability must be clearly defined.
- Reliability — AI output may be biased, incorrect, or irrelevant; it should be checked and validated before use.
- Safety — the AI system should be properly designed, tested, and monitored.
- Transparency — how data is used and handled, how algorithms work, and how decisions are made should be shared with end users and impacted parties.
- Copyright — ownership of AI-generated information can be unclear; know the applicable regulations and the level of human elaboration involved.
- Sustainability — each AI request consumes electricity, water, and other resources; consider this when deciding to use AI for a task.
Organizational ethical guidelines and AI policies are the basis for a shared understanding of how AI is used; project professionals should foster a culture of awareness and ethical AI use in the team.
Related concepts
Exam angle
- Data leakage scenarios: confidential data pasted into a public AI tool = an active privacy/compliance issue — first assess exposure and respond per the organization’s privacy/AI policies, not “add it to the risk register” or “ban all AI”
- Accountability can’t be delegated to the machine: when asked “who is accountable for an AI-made decision”, the answer is always a clearly defined human role, never the system or the vendor
- Validate before acting: AI forecasts and analyses are inputs to be checked (reliability), not triggers for immediate rebaselining or escalation
- Bias needs systemic fixes: right answers pair human review with diversified data and periodic bias testing — not arbitrary mechanical fairness rules and not quietly abandoning the tool