AI won’t transform agricultural research on its own. Institutions must.

By Eliot Jones-Garcia, Senior Research Analyst, IFPRI, for CGIAR Digital Transformation Accelerator

Agricultural research is at a turning point. 

As climate pressures intensify across food, land, and water systems, Artificial Intelligence (AI) presents an opportunity to keep pace. 

But only if institutions are prepared to change with it. 

Early experience with generative AI shows why. Despite an estimated $30–40 billion invested, the vast majority of organizations report little to no measurable return. Too often, new tools are added to old workflows, without changing the systems needed to use them well.

A survey of about 500 CGIAR researchers revealed a familiar pattern: strong interest in AI, but limited adoption. Not because researchers are resistant, but because the conditions for meaningful use are not yet in place.

4 points to address to make the most of AI for food systems research

AI transformation is often framed as a question of adoption. It can better be understood as one of adaptation. 

For AI to deliver value, organizations must rethink how research is designed, conducted, disseminated. This means moving beyond isolated pilots toward collective workflows and governance standards that enable human and machine intelligence to work together effectively. 

Our study points to four priorities. 

First, reduce the cognitive burden on researchers.

Many scientists today are overwhelmed by administrative tasks which prevent them from fulfilling more critical and creative tasks that inform good research. AI can help by automating routine work: cleaning and annotating data, summarizing literature, drafting reports, preparing meeting notes, translating materials, organizing field documentation, etc.

But this also presents a significant risk of over-automation. On one hand, it could undermine the culture of mentorship through which early-career researchers learn. On the other, it could create a growing expectation that scientists should produce more with less. 

We recommend institutions should prioritize quality over quantity, protect time for reflection and mentorship, and resist importing a model of disruption into research environments where trust, rigor, and collaboration matter. AI transformation should be an opportunity for the human work of science to flourish: using AI to reduce unnecessary burdens while strengthening the judgment, creativity, and relationships that make research valuable.

Second, strengthen research processes, not just outputs.

Researchers are understandably cautious about integrating AI into core scientific work, given concerns about quality and rigor. Yet this caution is compounded by limited AI literacy training and weak institutional coordination, leaving researchers caught between the sheer amount there is to learn and the fear of being left behind or being displaced by technical experts. 

Rather than treating this caution as resistance to change, institutions should recognize it as a responsible response to real risks, and meet it with selective, strategic integration.

This starts by looking inward before looking outward: understanding the value of agricultural research, and how its history of impact can inform the future. 

For example, agricultural research organizations steward some of the world’s most valuable datasets for training AI systems that respond to the realities of food, land, and water systems in the Global South. That gives researchers a critical role not only in using AI, but in shaping how it is developed, and ultimately, whose interests it serves.

The partnership between CGIAR’s Consortium for Spatial Information and Ai2’s OLMo Earth platform reflects this approach: not simply providing access to AI tools, but ensuring they informed by high-quality data and designed around real-world challenges in food, land, and water systems. 

Third, create clarity at the system level. 

If AI is to be used responsibly across research organizations, researchers need more than encouragement. They need clarity. One of the clearest lessons from our investigation is that uncertainty about what is allowed, what is safe, and what is responsible can slow transformation as much as lack of interest or capacity. 

This is a system-level challenge. Without shared standards, ethical frameworks, and governance structures, researchers are left to make individual judgments about tools whose risks and implications are still evolving.  

Institutions therefore need to make responsible AI use easier to navigate. This means setting clear expectations while also creating protected spaces for experimentation, where researchers can learn, test, and improve without fear of getting it wrong. 

At CGIAR, initiatives addressing these concerns include a guide on the Ethical Use of AI in Food, Land, and Water Systems Research, which helps researchers navigate methodological and ethical challenges, and a Multidimension al Digital Inclusivity Index,  to support the responsible design and implementation of digital and AI tools. We also regularly host public webinars and write thought-provoking articles, like this one, to help researchers develop confidence and skills and create space for discussion and learning. 

Finally, build the infrastructure for collaboration. 

AI is only as powerful as the data it can access. Yet in many research systems, data remains fragmented, difficult to find, uneven in quality and constrained by concerns around ownership. 

Investing in shared, interoperable data systems, and in tools that make knowledge searchable and usable across teams, is foundational to a meaningful transformation. 

At CGIAR, Fairgrounds is being developed as a collaborative, federated data platformhttps://www.fairgrounds.ai/ bringing together agricultural data from around the world while respecting ownership, permissions, and responsible data stewardship. Rather than moving data into a single repository, Fairgrounds brings models to the data through clear licensing and permissions. In this way, it balances openness and trust: expanding researchers’ capacity to use AI and scientific models, while safeguarding sensitive information. 

This infrastructure also creates the conditions for new forms of innovation. For example, CGIAR is leveraging AI71 s engineering expertise to develop CGIAR AI Hub’s products, such as Genebanks AI, an AI-powered platform linking genetic, trait, and phenotypic data across CGIAR’s eleven Genebanks. The goal is not simply to build new tools, but to connect data, models, and expertise in ways that make collaboration easier and scientific discovery more powerful. 

A moment of choice 

Agricultural research organizations face a clear choice today.  

They can adopt AI unevenly, tool by tool and team by team, using it mainly to speed up existing tasks. That path may look productive in the short term, but it risks doing more harm than good: weakening scientific judgment, fragmenting knowledge, and creating pressure to produce more without improving the systems that make research valuable. 

Or they can embrace a deeper transformation, one that rethinks how science is done, how teams collaborate, and how knowledge flows. 

AI will not replace researchers. But it will reshape research. 

The institutions that succeed will not be those with the most advanced tools, but those that create the conditions for those tools to matter. 

***

Learn more:  

• CGIAR Accelerator on Digital Transformation. 2025. CGIAR AI landscape 2025: Needs, opportunities, and next steps. Digital Transformation Report. CGIAR System Organization. https://hdl.handle.net/10568/178766 
• Jones-Garcia, E., Malapit, H., Magalhaes, M., Go, A. and Bryan, E. 2025. Ethical use of artificial intelligence in food, land and water systems research: a guide for equity and inclusion. Nairobi, Kenya: CGIAR Gender Equality and Inclusion. https://hdl.handle.net/10568/176705  
• Blog: Beyond the hype: Centering humans in CGIAR’s genAI research 

Main image: Group photo of participants at the workshop “Accelerating Food-Land-Water Systems Research in CGIAR through Responsible AI Integration”, held 24-26 September 2025 in Colombo, Sri Lanka. Credit: IWMI. Written with Julie Puech, CGIAR Digital Transformation Accelerator's Communications Lead.