Why humanitarian work needs its own ethical framework
Most charters for "responsible" AI are written for the private sector: they talk about commercial bias, regulatory compliance, brand reputation. The humanitarian sector asks different questions. Can an AI be trusted when the data concerns people in a situation of vulnerability? Who remains accountable if the AI gets it wrong? How do you explain a decision to a donor, to an auditor, to an affected population?
TRUST (Transparency, Reliability, Usefulness, Supervision, Tightness) is an attempt to answer those questions concretely, not with abstract principles, but with criteria you can check to judge whether an AI application really meets what the sector requires.
The 5 principles
A data flow you can follow from end to end
- An AI pipeline you can follow from input to output.
- Steps that can be audited and traced.
- Methods and tools stated openly (summarising, classification, and so on).
- The ability to explain how the result was produced.
An AI paired with checks, never left to itself
- AI controlled by verification scripts.
- AI coupled with fixed rules (IDs, lookups, validation).
- Source data is mandatory.
- No result without verified evidence.
AI where binary logic reaches its limit
- Cognitive complementarity: AI steps in where binary logic no longer suffices.
- Reading nuance (qualitative material, sentiment, themes).
- AI = reasoning and interpretation; code and Excel = deterministic calculation.
- Strictly targeted use, so nothing drifts.
The human being stays accountable for the output
- The human being stays accountable for the output.
- AI does not do 100 % of the work.
- A critical read through is mandatory before anything is shared.
- AI is a tool, not a magic wand.
Data kept under control, never scattered
- Data stored on secure servers.
- Only secure, non commercial AI is used.
- Audit and transparency on how data is used.
TRUST, as applied at Opti'
This framework is not just a statement of intent. It drives concrete design choices.
Transparency
OptiBot answers from your own documents and identifiable sources (your personal library) rather than from claims nobody can check.
Reliability
The generators (XLSForm, reports) combine AI with fixed structural rules. A generated XLSForm respects the syntax, it is not "half invented".
Usefulness
AI analyses and interprets (free text, qualitative answers). Calculations and cross tabulations stay deterministic, never left to the chances of a language model.
Supervision
Every generated document (report, ToR, questionnaire) stays editable and must be read through before it is shared. Opti' never publishes anything in your name.
Tightness
Conversations and documents are encrypted, hosting is in Europe, and the service is GDPR compliant. For the most sensitive data, OptIA mode processes everything locally, with no third party commercial AI.
About the author
Thomas Lequeux is a consultant specialising in the digitalisation of MEAL for the humanitarian sector, and the creator of Opti'. His work is at lequeux-thomas.com.