Early adopters align to scale responsible AI across CGIAR R&D
AI is changing how research gets done. Across CGIAR, teams are already using AI to design studies, manage data, and communicate results. As CGIAR's Executive Managing Director noted earlier this year, these advances raise strategic questions for how the system moves forward.
CGIAR's Digital Transformation Accelerator (DTA) has launched a multi-year effort to bring AI into food, land, and water (FLW) systems research with discipline and measurable value. The International Water Management Institute (IWMI) is coordinating a core part of this work: assessing current use, aligning practices, and creating the conditions to scale responsible AI across CGIAR.
Why this matters for IT and development teams
Research is full of repetitive, high-friction tasks: data prep, literature synthesis, multilingual reporting, and model runs. AI can compress those cycles. The challenge is doing it with solid data foundations, clear guardrails, and shared components instead of isolated pilots.
An IFPRI-led survey of 418 staff showed modest but growing AI use. It also flagged real concerns: data quality, limited capacity, and a lack of clear guidance on responsible use.
The Colombo workshop: mapping, vision, execution
From 24-26 September 2025, IWMI and DTA brought AI Champions from 11 CGIAR centers to Colombo, Sri Lanka. The goal: map what exists, define where to go next, and plan how to get there across FLW systems.
As IWMI's Deputy Director General for Research for Development put it, the workshop helps position CGIAR as a leader applying AI for sustainable food, land, and water systems in a climate crisis context.
Mapping the present: tools in play
Champions shared active tools and lessons learned. Highlights included:
- SNAP (Semantic Natural Language Processing Aggregator Platform) - from the CGIAR System Office, auto-generates summary reports from CGIAR's research databases.
- Water CoPilot - an IWMI assistant that answers water management questions for the Limpopo River Basin using real data and modeling, supports multiple languages, and can render relevant graphics.
- Digital Twin for the Limpopo River Basin - IWMI's simulation environment to explore scenarios and decisions at basin scale.
- Ask the Data - a CIP tool that lets researchers query complex datasets with natural language, targeting the data prep workload that can consume up to 80% of a scientist's time.
- Topic-Specific News Agent - an IFPRI service that continuously tracks news on focused topics (e.g., export restrictions for key commodities) and delivers curated, multilingual summaries.
The session gave a clear picture of where capabilities exist, who to call for help, and which roadblocks keep surfacing.
Envisioning the future: human-centered scenarios
Weeks before Colombo, CGIAR experts met in Cali, Colombia, for a scenarios workshop hosted by the Alliance of Bioversity and CIAT, IWMI, and DTA. Instead of chasing tech trends, the group used design thinking and speculative futures to center smallholder farmers, communities, policymakers, and researchers in the Global South.
The takeaway: explore plausible futures, weigh risks and opportunities, and use scenarios to handle uncertainty while strengthening relationships and ideas. CGIAR needs clear strategic paths, and diversity across centers and nationalities is essential because digital technologies already affect society and agriculture.
From vision to operating model: OKRs and the AI Co-Scientist
In Colombo, Champions built practical visions for AI-augmented R&D over a 3-5 year horizon. Using LEGO Serious Play, teams modeled the people, workflows, values, and failure modes that will matter. They then drafted common objectives and key results (OKRs) to track progress.
Teams also explored an AI Co-Scientist concept: tools that speed routine tasks, suggest and check hypotheses, assist with experiment design, and support multilingual communication. Using a human-centered design approach, participants prototyped where this fits within CGIAR's workflows and how to keep humans accountable for decisions.
One core outcome: a shared vision for AI-accelerated R&D in FLW systems and the definition of a supporting AI platform to supply tools, data access, governance, and services in a consistent way.
What IT and development teams can do now
- Audit data readiness - schema discipline, metadata, versioning, access controls, and data quality gates. No reliable data, no reliable AI.
- Build as a platform - shared services for auth, logging, evaluation, prompt/version management, and API-first tools to avoid one-off pilots.
- Start with high-friction tasks - data prep, literature search, synthesis, multilingual summaries, geospatial preprocessing, and simulation orchestration.
- Bake in responsible AI practices - bias checks, red-teaming, model cards, eval sets, and audit trails. Use the NIST AI Risk Management Framework as a baseline.
- Design for low-connectivity and multilingual needs - edge-friendly inference, caching, and localization for Global South contexts.
- Measure outcomes - time saved, accuracy/error rates, reproducibility, and decision lead time; tie them to OKRs.
- Upskill the team - retrieval design, prompt engineering, evaluation methods, and data governance. If you need structured paths, see the AI certification for coding.
- Plan compute and cost observability - mix open-source and managed models; track token/compute budgets; enforce privacy and export controls.
What's next
Across CGIAR there is deep, multidisciplinary AI expertise. With a shared vision and platform approach, efforts can be coordinated to reach scale.
The roadmap and OKRs co-developed in Colombo will be reported to CGIAR leadership to shape an organization-wide approach to responsible AI-accelerated R&D in FLW systems. The consultation, "Accelerating Food-Land-Water Systems Research in CGIAR through Responsible AI Integration," was held 24-26 September 2025 in Colombo, Sri Lanka.
This work was supported with grants from CGIAR and Google through the CGIAR Accelerator on Digital Transformation.
Your membership also unlocks:
