AI Ethics for AgDev Toolkit overview

The Toolkit is a decision-support resource for teams designing, developing and deploying GenAI systems that use agricultural or agriculture-adjacent data.

Its activities and supporting resources help your team move from high-level ethical principles to structured discussions about your own project. Use the public activity summaries to identify relevant resources, then access the associated materials to work with your team.

Data that is reusable is not necessarily suitable for every AI use. The way data is selected, combined, interpreted and governed can influence an AI system’s outputs and their suitability for the people using them.

The Toolkit supports team discussions about:

• Data provenance, quality, representativeness and intended reuse.
• Ethical, governance and operational risks.
• Stakeholder participation, local knowledge and feedback.
• Documentation, traceability, accountability and human oversight.
• Relevant legal and regulatory questions requiring review.

Use the Toolkit if you develop, research, oversee or implement agricultural GenAI systems, especially those that influence decisions.

It is relevant to technical leaders, data scientists, machine learning engineers, product managers, research organisations and think tanks. Teams can also involve ethics and governance colleagues, implementation partners and relevant stakeholders.

The User guide explains how the resources are organised and which roles can contribute.

Agricultural data and advice depend on local conditions, including soil, climate, farming practices and regional differences. An apparently plausible AI output may still be unsuitable for the people or setting it is intended to support.

The Toolkit’s examples include farmer and extension advice, agricultural policy and evidence synthesis, and crop education and outreach. These contexts help teams recognise why local knowledge, stakeholder participation and appropriate oversight matter.

The Toolkit contains three complementary checklists:

• Risk identification and mitigation: resources for recognising potential harms and considering safeguards.
• Participatory co-creation, feedback and model refinement: resources for considering stakeholder involvement and learning from feedback.
• Legal and regulatory compliance: resources for identifying relevant issues for further review.

Activities are organised under Design, Development and Deployment, with eight lifecycle subphases. Each public preview explains the task, relevant roles, agricultural context and intended benefit so you can choose an appropriate resource.

Full activity materials provide the questions, considerations, instructions and examples needed for team working.

The FAIR Process Framework supports Findable, Accessible, Interoperable and Reusable data. These practices provide an important foundation, but do not by themselves establish that data is appropriate for a particular AI use case.

The Toolkit addresses additional considerations when data is reused to train or fine-tune AI models, including context, representativeness, provenance, governance and lawful licensing.

The Toolkit is a baseline decision-support resource. It is not a certification scheme, a formal skills qualification or a guarantee of safety or legal compliance.

Use it alongside your project’s technical, legal and governance processes. Its value depends on the quality of your team’s discussions, evidence and follow-through.

Read the User guide to understand the resource structure and who should take part. Explore Activities to find resources relevant to your project’s task and lifecycle phase.