Q&A: Courtney Heldreth and Diana Akrong on supporting farmers with machine learning

People + AI Research @ Google
People + AI Research
8 min readOct 1, 2020


Illustration by Shannon May for Google

This is the third in a series of Q&As with design and UX practitioners sharing perspectives on participatory machine learning. Check out our previous installments, with Sarah Gold of IF and Mat Budelman of Spotify.

Dr. Courtney Heldreth and Diana Akrong are user experience researchers in Google Research. Their work explores how AI can help improve the lives of farmers in the Global South. They look at incorporating farmers’ needs, practices, value systems, social worlds, and daily agricultural ecosystem realities into the products Google builds.

Diana is the founding member of Google Accra’s UX team. Courtney is a social psychologist. She co-leads a team conducting culturally embedded research to understand the risks and opportunities of AI for people in the developing world. They were interviewed by David Weinberger in May 2020.

David: What are some of the challenges local, small farms in the developing world are facing?

Courtney: By 2050, we’re expecting to add another 2.5 billion people to the planet, while there’s less and less arable land, and climate change is continuing to have devastating impacts on food production.

Diana: And the impact of these problems on the developing world will of course be greater than on the richer nations.

David: So, no shortage of huge, planet-threatening, unfairly distributed problems. What are some of the issues with trying to bring machine learning to bear on problems like these?

Diana: Here’s a data issue specific to the developing world.

Satellite imagery is useful in spatial analysis to provide personalized and actionable information to farmers, from weather to soil moisture levels. In much of the developing world, the resolution is not high enough to capture the details of small farms which are typical in these regions. This makes it difficult to provide accurate and personalized recommendations and predictions for those farms.

Courtney: And that really matters. The topography and parameters of a field are super variable so when you’re trying to detect soil moisture levels in a small farm, it can be very different in a farm 100 feet away. Also a lot of the farms in the developing world don’t distinguish very precisely where they start and end.

Then, there are serious issues establishing ground truth data…

David: …That is, data that a machine learning system learns from, which it’ll take as accurate, so it better actually be accurate ….

Courtney: Yes, ground truth data is really important for creating accurate machine learning models. But in the more rural parts of the world, it’s often expensive and laborious to collect data. For example, oftentimes you need to collect multiple soil samples from different parts of a field over time. That requires bringing people out to hundreds of farms, collecting a large number of samples, and sending those samples to a lab. This can be prohibitively expensive.

COVID is making it very hard for people to go out and verify the recommendations made based on satellite imagery. And it can be hard to pinpoint which data applies to which geographic spot when the concept of fixed, sharp boundaries is culturally insignificant.

Diana: That’s one important part of the sociotechnical context. There are so many intersecting factors that affect whether and how technology can be employed.

David: Such as?

Diana: One example is the network infrastructure in rural communities. The majority of network connections in rural areas in Sub-Saharan Africa rely on 2G technology with limited coverage. Expanding the network in these areas is challenging due to high operation costs and low revenue opportunities. This presents limitations for how cloud-based solutions can work in these regions and to a certain degree, how data can be collected and curated for use.

Courtney: That’s a great point. Most government data gathering serves the primary interest of policy-making and is not sufficiently local for the needs of individual farmers. While some entities don’t see enough return in investing in tech and data collection in these small farms and villages, we see it as an opportunity. How can we meet these smallholder farmers where they’re at? How do we do a better job of understanding the needs and situations of smallholder farmers to uncover the ways AI can create more economic empowerment?

David: So, you’re not looking to replace traditional practices with shiny new everything.

Diana: Not at all. We believe that there is an opportunity to use technology to augment human capabilities and processes in the agricultural value chain.

David: How do you avoid imposing solutions that make sense to the technologists but might not actually work for the farmers they’re intended to help?

Diana: The local farmers know their challenges best. So we strongly believe in participatory design: creating solutions that are useful for, and usable by, a community. And you do that by making the community part of the process.

Not only does this result in improvements that work and that respect the needs and values of the local communities, but it helps ease the adoption of those improvements. The way tech adoption works in rural communities is that the local farmers try it, and if they like it, they tell other farmers about it. Word of mouth is the best recommendation system.

Courtney: If you design for farmers rather than with them, the technology either isn’t adopted, or it doesn’t spread.

Diana: We’ve also seen cases in which good information about preventing particular crop diseases has been expressed in ways that don’t align with how farmers themselves think and talk about it. That’s another reason we’re advocating for farmers’ voices to be heard throughout the machine learning development process.

David: So how do you put together a technology that helps in such a complex environment?

Courtney: The million dollar question! Our approach has been to ask: What are the bite sized pieces that are a nice intersection between AI’s strengths and the needs of local farmers? The agricultural ecosystem is extremely complex — there are suppliers, farmers, merchants, delivery systems, customers… We can’t try to tackle it all because we know we can’t be experts across the entire value chain.

But you can’t avoid this complexity. In order for a meaningful technology solution to work for this population, you need to consider the entire value chain — farming has so many links to food security, and public health. Insects can start to breed on crops, deforestation can lead to the spread of malaria and to malnutrition. We don’t have all the answers. We’re designing for complex, delicate ecosystems.

For example, we support farmers to improve productivity, but that’s not the end. High supply can mean lower prices, which drives down farmers’ incomes, so we also need to facilitate access to information about what other farmers in a community are growing, facilitate access to markets, and do what we can to make sure that good information moves through the entire value chain.

David: If the value chain — which is really a type of ecosystem — is so complex and important, but also so large that you as a technologist can’t hope to understand it all, how do you proceed?

Courtney: Our team conducted research to learn about the challenges farmers face across this complex value chain. Through this, we identified 2 key leverage points, which are defined as places in a complex system where a small shift in one thing can produce big changes in everything. Our first leverage point is enabling sustainable agriculture practices, which is a system of practices that increases biodiversity, enriches soil, improves watersheds, and enhances ecosystem services. Our second leverage point is identifying ways to provide more financial security to farmers, thus improving their resilience.

As we begin to explore these areas, we maintain primarily focus on understanding farmers’ needs and challenges. That’s why community partnerships have been so pivotal. Having partners who have developed trustful relationships with farmers — and the government, too — is really critical to developing any technology that will actually help.

David: Building trust must be a crucial part of this.

Diana: Absolutely. Communities are built on trust, so working with them is vital.

David: What’s the current status of the project?

Courtney: We’ve been taking a very foundational farmer-centered approach about how we want to move forward to identify AI tools that will make an actual impact on people’s lives. That means getting on the ground and talking with farmers, what they need and want, how that intersects with financial security, family stability, even mental health. We are working on talking with five hundred farmers all across India, to learn about their needs and how AI can help them.

And then COVID-19 came. All face to face interactions for research have been postponed. We’ll pick up that work as soon as it’s safe to do so. In the meantime we’re focusing on expert interviews.

David: Which experts?

Diana: Researchers, agronomists, farm input distributors, agribusiness managers, policymakers, academics and farm extension workers who work in countries in Sub-Saharan Africa, India and Indonesia. These experts work closely with farmers and some of them are farmers themselves.

David: Are your results going to be made public?

Courtney: Yes. We plan on making as much of it public as we can. We’ll be publishing a research paper on farmer-centered AI, proposing a framework for understanding farmers needs and their social systems. We also plan on publishing the results from the farmer survey and hope to share that with the public.

David: When?

Diana: Hopefully by the end of the year. But COVID-19 is dictating our schedule right now. So much of the work requires travel that we can’t undertake now. We want to prioritize in-person interactions for conversations with farmers to allow us to build relationships and establish trust as we learn about their needs, goals and aspirations.

David: What constitutes success?

Courtney: This is an ongoing learning process. But we’ll count the project as successful if it results in more adoption of sustainable agricultural practices by smallholder farmers, resulting in farms that are more productive and more resilience among farmers. That’s what we hope is attainable.

Diana: Being able to maximize productivity with limited resources is important if you’re a smallholder farmer. Resilience is important because the ecosystem is so complex and unpredictable. The more we can predict and help reduce the uncertainty to control outcomes, the better for these farmers.

David: And the farmers’ voices in deciding what constitutes success?

Courtney: That’s the most important aspect of our work. We’re talking with them through our research and our partners. We will talk with them consistently throughout the development process. Because we’re UX researchers — and fellow human beings — we care deeply about making sure the voices of farmers are heard and are never lost.



People + AI Research @ Google
People + AI Research

People + AI Research (PAIR) is a multidisciplinary team at Google that explores the human side of AI.