Human-First AI Manifesto for Monitoring & Evaluation
Human-First AI in Monitoring and Evaluation:
Principles and Practice
Human-First AI is an ethical approach ensuring artificial intelligence supports—rather than supplants—human judgment, contextual understanding, and professional accountability in monitoring and evaluation. It prioritizes human purpose, ethical safeguards, and evaluator responsibility over automated efficiency. Download it here to access the full Human-First AI in Monitoring & Evaluation manifesto.
Introduction
As artificial intelligence becomes embedded in evaluation tools—from qualitative analysis to report drafting—the sector faces a critical choice: adopt AI as a neutral instrument or embed it within a human-centered ethical framework. The Human-First AI manifesto provides that framework for Monitoring and Evaluation (M&E) professionals worldwide. Developed through consultations with evaluators, data scientists, and humanitarian organizations, it responds to the humanitarian AI paradox: widespread individual use amid low organisational readiness. By 2026, every evaluator will encounter AI-generated content; this manifesto ensures they do so with integrity, transparency, and accountability.
What is Human-First AI in Monitoring and Evaluation?
Human-First AI refers to the deliberate design and deployment of artificial intelligence systems that augment human evaluators without replacing their core functions. It recognizes that AI can process patterns at scale, but only humans can interpret political context, cultural nuance, and ethical implications. This approach aligns with principles from the OECD AI Principles and the UNESCO Recommendation on AI Ethics.
How does Human-First AI differ from conventional AI approaches?
Conventional AI often prioritizes efficiency and automation, sometimes at the expense of context. Human-First AI embeds human oversight at every stage—from defining evaluation questions to validating outputs. It treats AI as a co-pilot, not an autopilot.
Key characteristics of Human-First AI in M&E:
- Purpose defined by humans: Evaluation questions, learning priorities, and success criteria remain human-led.
- Contextual interpretation: AI identifies patterns; humans assign meaning based on political, cultural, and ethical context.
- Ethical guardrails: Safeguards for privacy, consent, and bias mitigation are non-negotiable.
- Professional accountability: Evaluators sign off on findings; AI does not assume responsibility.
Why Does Human-First AI Matter for Evaluation Professionals?
The UNICEF AI for Children project and World Bank AI initiatives highlight risks when AI systems lack human oversight: biased algorithms, misinterpreted data, and eroded community trust. For evaluators, Human-First AI preserves the profession's core values—independence, credibility, and utility—while harnessing AI's analytical power.
What happens when AI is deployed without human-first principles?
Case studies from humanitarian settings reveal that AI-only approaches can amplify existing biases, ignore local knowledge, and produce recommendations that are technically correct but contextually invalid. Human-First AI acts as a corrective, ensuring evaluations remain grounded in lived experience.
The Ten Principles of Human-First AI for M&E
AI executes tasks; evaluators set the agenda.
Pattern recognition is machine; interpretation is human.
No workflow optimization justifies ethical shortcuts.
Evaluators sign reports; AI assists but does not answer.
AI recombines; evaluators imagine.
Synthetic insight cannot replace community memory.
Outputs are suggestions; evaluators decide.
AI summarizes; humans reflect and internalize.
Lowered barriers must not erode standards.
AI reflects human choices, not conscience.
How to Implement Human-First AI in Evaluation Workflows
Implementation begins with organisational change, not technology procurement. Drawing on NetHope's AI readiness framework, M&E units should:
- Assess governance gaps: Review existing policies on data quality, algorithmic bias, and informed consent.
- Define acceptable use cases: Distinguish low-risk (summarization) from high-risk (targeting) applications.
- Build human-in-the-loop protocols: Ensure every AI-assisted finding is reviewed by a qualified evaluator.
- Document AI involvement: Make it transparent which parts of a report were AI-generated.
Practical Tool: AI Usage Disclosure Template
To operationalize Principle 4 (Accountability is Human), include this statement in evaluation reports:
“This report used AI-assisted tools for [initial transcription / literature scanning / draft summarization]. All outputs were reviewed, validated, and interpreted by the evaluation team, who take full responsibility for the findings and conclusions.”
Frequently Asked Questions
What is Human-First AI in simple terms?
It's an approach where AI supports human evaluators without replacing their judgment, ethics, or accountability.
How does Human-First AI relate to evaluation ethics?
It operationalizes ethical principles—like respect for persons, beneficence, and justice—in AI-assisted evaluations.
Can AI be used for qualitative data analysis under this framework?
Yes, but only as a first-pass tool; themes must be validated by human evaluators with contextual knowledge.
Where can I learn more about responsible AI in M&E?
Explore the EvalCommunity AI in M&E course and the OECD AI Observatory.
Authoritative Resources
- OECD Artificial Intelligence Principles – International standards for trustworthy AI.
- UNESCO Recommendation on AI Ethics – Global normative framework.
- World Bank AI in Development – Policy guidance and case studies.
- UNICEF AI for Children – Child-centered AI principles.
- NetHope AI Readiness Framework – Organisational assessment tools.
Conclusion
Human-First AI is not a rejection of technology but a commitment to using it wisely. For Monitoring and Evaluation professionals, it ensures that as AI becomes ubiquitous, the core values of the profession—independence, credibility, and utility—remain intact. By adopting these principles, evaluators can harness AI's power without surrendering their essential role as interpreters of human experience.
Human-First AI Manifesto for Monitoring & Evaluation
Preamble
Artificial Intelligence is transforming Monitoring & Evaluation faster than any tool before it. But AI is not the evaluator. You are.
This manifesto defines the principles that ensure AI strengthens — not replaces — human judgment, ethics, creativity, and responsibility in evidence-based decision-making.
Ten core principles
AI does not decide what matters. Humans define evaluation questions, learning priorities, and success. AI supports execution, speed, and scale.
AI detects patterns; humans interpret meaning, context, and significance. No algorithm can replace professional judgment.
Optimization never justifies ignoring bias, privacy, or consent. If AI cannot be used ethically, it should not be used.
Evaluators sign reports; AI assists but does not answer. Professional responsibility cannot be delegated.
AI recombines existing knowledge; evaluators imagine new questions, approaches, and solutions.
Community knowledge, historical memory, and local expertise are irreplaceable by statistical patterns.
Outputs are suggestions, not verdicts. Evaluators retain the right to question, override, or discard AI-generated content.
AI can summarize data, but reflection, adaptation, and internalization happen among people.
Lower barriers to entry must not lower ethical or methodological standards.
AI reflects human biases and choices. It has no conscience; therefore humans must govern its use.
Closing Commitment
As Monitoring & Evaluation professionals, we commit to:
- ✓ Use AI responsibly
- ✓ Lead with ethics
- ✓ Protect human judgment
- ✓ Strengthen — not surrender — our expertise
- ✓ Stay curious, critical, and creative
Evidence serves people.
Evaluating Youth Employment Programs Using Predictive Analytics
Education Evaluations in Low-Connectivity Environments
Monitoring Climate Resilience Projects with AI and Satellite Data
Making Peace with Logframes: A Guide for People Who Hate Them
