Insights · AI Governance
Responsible AI: What It Actually Means for Your Small Project
Your team has started using ChatGPT or a similar tool to design a program or social project — but how do you avoid it becoming a source of bias or fabricated information? data.org published a practical guide for exactly this, aimed at development and social-impact organizations. Here's a direct summary of its four steps.
Last updated: October 2026 · based on data.org's guide, published October 2024
Step 1: Understand where AI actually helps
The guide recommends using generative AI as a "brainstorming partner," not a final decision source:
- Use it for initial ideas and general contextual information — the target community's landscape, the regional political climate, trends in indicators like education, income, and health
- Use it only on topics where your team already has enough knowledge to judge whether the answers are accurate and relevant
Step 2: Understand the real risks
Before any use, the guide suggests asking yourself three questions: How will I use this tool? What flaws might exist in its information? What harm could inaccurate answers cause?
| Risk source | Detail |
|---|---|
| Lack of data | Marginalized and low-income communities often have sparse data, so AI may not represent them accurately |
| Misleading data | Training data can reflect existing societal inequalities, causing AI to reproduce stereotypes |
| Four bias types | Stereotype, racial, cultural, and gender bias |
Step 3: Practices to mitigate bias
- Evaluate content: Check responses against your own expertise, and consult subject matter experts
- Cross-validate: Compare outputs from multiple tools or techniques to spot inconsistencies
- Refine your prompts: Iterate on prompts and use precise, descriptive wording
- Fact-check with ground truth: Verify conclusions and references against real data — AI can fabricate entire citations
Step 4: "So what" — next steps
The guide recommends treating AI as a tool that takes time to learn, not a complete ready-made solution. Use it to support decisions only alongside your prior knowledge and additional analysis — so it augments your team's creativity and expertise rather than replacing it.
