Datafication of Agriculture

Collecting, analysing, and managing data have been important tools for organising and governing culture and society since ancient times (Koenen et al. 2021). Nevertheless, datafication (Mayer-Schönberger and Cukier 2013), ‘the process of rendering information into machine-readable quantifiable data for the purpose of aggregation, analysis, and anticipation of human behavior and social interaction’ (Koenen et al. 2021:137), emerged late in the twentieth century, with the development of digital platforms that increasingly enable the accumulation, processing, and nearly endless recombination of large-scale data collected and scraped from across daily life. And while the datafication of daily life has been vigorously promoted as offering opportunities for innovation, rationalisation, and optimisation (Morozov 2013), there is evidence that it ‘can also create new and reinforce old divides’ that exacerbate colonial and capitalist inequalities (Koenen et al. 2021: 150; c.f., Couldry and Mejias 2019).

So too, the measurement and quantification of agriculture has long been underway, but the ‘digital turn’—termed Agriculture 4.0 (De Clercq et al. 2018)—has been viewed from within the field as ‘a transformative force in agricultural production systems, value chains and food systems’ (Klerkx et al. 2019: 1). Operating under labels such as smart-farming, precision agriculture, and sustainable intensification, the digitisation of agriculture has focused on integrating data collected both in the field (i.e., using sensors and robots) and at a distance (i.e., using drones, satellites, and weather stations) to support farmer decision-making and practice. Responding to the challenge of meeting human agricultural needs in the context of climate change, population growth and urbanisation, degradation of natural resources, and food waste, the promise is that digital agriculture will allow farms to become ‘more profitable, efficient, safe, and environmentally friendly’ (De Clercq et al. 2018: 4).

At the same time, there are growing critiques that the digitalisation of agriculture ‘may entrench productivist and neo-productivist visions of “feeding the world”’ (Montenegro de Wit and Canfield 2023: 1). Where the twentieth-century ideal of agricultural productivism focused on maximising yields through industrial-scale production and the use of synthetic fertilisers and hybrid seeds (Stone 2022), the neo-productivist turn has focused on ‘technological and productivity-oriented innovations to improve resource efficiency while reducing adverse environmental and health impacts of current food systems’ (Montenegro de Wit and Canfield 2023: 4; Wilson and Burton 2015). Key to this critique is the concern that, despite its utopian promise, the underlying biases encoded into the infrastructures of digital agriculture are likely to exacerbate—or at least do little to undermine—the inequalities in the current food system (Bronson 2019; Carolan 2020; Hackfort 2021; Montenegro de Wit and Canfield 2023). In particular, scholars are interrogating the ways that digital agriculture is likely to ‘lock in’ industrial-scale monocropping and ‘create path dependencies that entrench dominant food regimes’ (Montenegro de Wit and Canfield 2023:5)—in no small part because much of the emergent digital agriculture technology is being developed and distributed by global, private corporations that are deeply invested in the status quo. Put simply, without deliberate intervention, the norms and assumptions that underpin productivist ideals will continue to be materially encoded in the tools and technologies of digital agriculture (Bronson and Knezevic 2016; Montenegro de Wit and Canfield 2023: 5). The consequences, according to Michael Carolan, are ‘not just world-shaping, but world making’ (2022: 208; c.f., Bowker 2000).

Uncertainty and Complexity in Sustainable Agriculture Research

For millennia, humans have been developing local and traditional ecological knowledge that has helped them to effectively cultivate food and fibre in diverse environments (White and Denham 2020). However, the entire trajectory of the twentieth-century productivist model of industrial agriculture—exemplified by the Green Revolution—has been to minimise uncertainty and complexity via one-size-fits-all strategies that depend heavily on external inputs of fertilisers and pesticides. Within the US industrial model of agriculture, the Land Grant research and extension apparatus has played a central role in identifying and disseminating ‘evidence-based’ approaches aimed at maximising productivity by simplifying practice. In contrast, the sustainable agriculture model emphasises the importance of farmers’ localised, place-based, and contingent knowledge developed through the lived-experience working their land (Carolan 2006). In an examination of the co-production of expert knowledge in agriculture, one of Michael Carolan's interlocutors explains, ‘We're [agricultural scientists] trained to reduce things to nice and neat formulas—have this crop rotation, apply this many chemicals at such and such times, and you'll have a bumper crop. But this completely glosses over the more site-specific knowledge that sustainable farmers need’ (2006: 425).

With the expansion of federally funded sustainable agriculture research and outreach at Land Grant Institutions, this dichotomy appears to be breaking down. Specifically, more so-called conventional producers are beginning to experiment with conservation agricultural practices (e.g., no-till/conservation tillage and cover cropping)—even, and maybe especially, in the context of commodity production (Wade et al. 2015). This is happening for a wide range of reasons: from the practical need to mitigate soil erosion and the effects of extreme weather (like drought and flooding), to increased financial incentives, and outreach and education focused on the benefits of these practices. At the same time there is a major rhetorical shift underway aimed at reframing agriculture (and thus farmers) away from being perceived as a contributor to climate change, and toward being a key part of the solution. In all, there are increasing efforts to support ‘conventional’ growers in the adoption (and effective implementation) of conservation practices. Yet, this presents a bit of a paradigm mismatch: many ‘conventional’ producers have survived in the context of industrialised commodity agriculture—where the profits margins are very tight—by minimising the complexity of the system. As they begin to experience the drawbacks of this approach (i.e., soil erosion, fields that are too wet/too dry, and increasing costs of inputs, including fertilisers and pesticides), they learn that cover crops could help with all these things. But adopting cover crops adds complexity. Farmers now need to consider what cover crop varieties to plant given their cash crop rotation, soil, and weather/climate; when to plant and at what seeding rate; when and how to kill the cover crop; and so on. If they want to receive financial incentives, it gets even more complicated: How are the incentives managed? What paperwork is required? How does the incentive constrain what, when, and how the cover crop is planted and killed? The bottom line is that for many farmers operating in a ‘conventional’ paradigm, the increased management complexity of conservation practices like cover cropping increases uncertainty. In short, supporting ‘conventional’ row croppers in adopting these conservation practices means meeting farmers where they are³. For many, this means providing the tailored and site-specific information and support they need to implement these practices locally and use them successfully in the short and long term.

Ethnographic Context: Precision Sustainable Agriculture (PSA)

Practically speaking, PSA is made up of three main interdisciplinary research efforts: (1) A series of controlled ‘common experiments’ on agricultural research stations across the Northeast, South, and Midwest that investigate how cover crop species and management impact nutrient, pest, and disease dynamics, as well as cash crop quality and yield. (2) OFR conducted on over one hundred privately-owned corn, soy, and cotton farms using cover crops in more than 20 states. Collaborating farmers make all their own management decisions while an OFR team collects data to determine how cover crops impact water, soil, nutrient, pests, disease, and cash crops in the ‘real world’⁴. And (3) socio-economic research aimed at understanding farmers’ experiences, values, priorities, and social networks. Data from across these research arms are being integrated within interdisciplinary teams—including systems modellers who aim to integrate the on-farm data with remotely-sensed and public data to develop predictive models (and ultimately web-based decision-support tools) that can reduce the uncertainties and maximise the benefits of cover cropping for farmers. In the words of one of the Extension Specialists on our team, the hope is that ‘this is going to take the “it depends” out of cover cropping.’ Thus, PSA is squarely embedded within a neo-productivist paradigm that posits that technology will enable producers to more effectively use resources in ways that reduce the environmental costs of commodity agriculture.

As the only anthropologist and lead of the social science team on this project, I wear several hats: I direct research aimed at understanding farmers’ perspectives and experiences with conservation practices; I serve as a member of the projects’ executive team; and I practice participant observation. I seek to engage in pragmatic, collaborative research to (imperfectly) tackle so-called wicked problems (Rittel and Webber 1973), while maintaining a critical engagement to question unexplored assumptions—especially about the unfettered potential of science and technology—and push for greater transparency and equity in research and practice. Practically speaking, this means I attend a lot of Zoom meetings.

In the first three years of the project, I consistently attended the biweekly meetings for the OFR team. My objective was to understand the development of this ‘participatory’ arm of the project: how were decisions were being made, protocols developed, challenges navigated, and when/where/how were farmers involved in the process. Early on, I noted the confidence with which the researchers and technicians on this team talked about what farmers do, how they think, and what they care about. Without a doubt, my colleagues on this team are extremely knowledgeable and experienced, and they have strong relationships with collaborating farmers. Yet surely, farmers—even farmers growing a small number of commodity crops—are not monolithic. They have different perspectives, experiences, and priorities. Who is the farmer that researchers are holding in their minds’ eye? Who is missing? At what cost to science and society?

This observation influenced how my team has approached our social science research. We've developed a mixed-methods interview study to characterise the range of mindsets (shorthand for attitudes, values, and priorities) held by farmers growing the same crops we were studying in our OFR; and we are now working to understand the ways farmers navigate the process of adopting and implementing cover crops over time, and what resources they draw on to do so. This initial observation also motivated me to critically examine what we know about the phenomenon of OFR.

The Black Box of On-Farm Research

On-farm experimentation is as old as agriculture. The most common form of OFR is farmers’ experimentation on their own farms, and in the context of farmer field schools (FAO n.d.), to develop localised knowledge about crop and whole-farm management. While individual experimentation is critically important for farmers and farming communities, the results are rarely integrated into larger research initiatives and are often dismissed within the scientific community as ‘anecdotal’ due to lack of replication and control plots (Norman et al. 1998).

On-farm collaborations between farmers and researchers are highly variable both in purpose and practice. They can range from basic research aimed at answering a fundamental scientific question, to validation or ‘translational’ research focused on evaluating agricultural practices developed on experimental farms ‘in the real world’ (Woolf 2008), to outreach aimed at demonstrating practice and disseminating knowledge to a broader population of farmers (Anderson 1992; Lockeretz 1987; Thompson et al. 2019). In the Global South, OFR is frequently part of agricultural development operations targeting under-resourced farmers (De Roo et al. 2019; Shaner et al. [1982] 2018); but in the US context, OFR often aims to extend the validity of research conducted on USDA Agricultural Research Stations and Land Grant experimental stations, increasingly in the context of advancing ‘sustainable’ agricultural practices (Carolan 2006; Norman et al. 1998).

Further, although collaborative OFR is frequently framed as a participatory approach to research (Lacoste et al. 2022), the degree to which farmers are actively engaged in developing the research questions, design, or even management of on-farm trials varies widely (Johnson et al. 2004). Farmer participation in OFR can range from having virtually no input at all (essentially, researchers using farmers’ fields to produce site-specific data), to implementing a researcher-designed protocol, to managing land as they normally would (perhaps within some parameters) while researchers collect data about ‘real-world’ farming conditions, and finally to co-designing research studies and protocols in full collaboration with researchers (Lockeretz 1987; Thompson et al. 2019).

Anthropology has long played a role in advocating for greater farmer participation in agricultural research. In 1982, anthropologist Robert E. Rhoades and technologist Robert H. Booth proposed the ‘farmer-back-to-farmer’ model of agricultural development, the foundation of which the belief that ‘successful agricultural research and development must begin and end with the farmer’ (Rhoades and Booth 1982: 132, emphasis original).

Research from this period also demonstrated that the degree of farmer engagement in OFR impacts research design and implementation. Comparing models of farmer participation in research design, Jacqueline A. Ashby (1986) determined that a ‘consultive’ model (in which farmers’ priorities were identified via an agronomic survey) resulted in researchers making changes to the research design only when there was a ‘major discrepancy’ between the experimental design and farmer practices. This model ‘tended to screen out practices followed by a minority of farmers, in favour of a focus on representative practices followed by the majority’ (1986: 14). In contrast, a more participatory ‘farmer-design’ model (in which a small number of ‘innovative’ farmers taught their experimental practices to researchers) uncovered farmers’ priorities and created space for translating farmers’ experimental practices into the research design in ways that broadened the research relevance for both farmers and scientists (1986: 17).

In related work, Ashby (1987) found that actively involving farmers in OFR decisions also impacted the outcomes and evaluations of trials. By actively considering competing demands for their time and labour rather than simply maximising yields, farmers developed different management strategies. This resulted in divergent trial outcomes and evaluations of the effectiveness of the research trials by farmers. Ashby's work demonstrates that farmer decision-making is far more complex than a ‘rational-choice’ model, instead drawing upon values, social relations, and complex orientations toward risk (Barlett 2013; Boholm et al. 2013; Ortiz [2004] 2020). Further, it provides insight into the ‘efficacy-effectiveness gap’ (referring to the variability in outcomes found in the context of a controlled trial versus real-world practice) (Eichler et al. 2011), which has been repeated many times in the history of both agricultural outreach and public health.

Yet even with the expansion of farmer-collaboration in OFR, we still know very little about who participates in on-farm research. Agronomic research using OFR rarely reports anything beyond the protocol and parameters of the agronomic study itself. The characteristics of the farmers who are collaborating in the research are generally treated as a black box. As anthropologists, however, we understand that the cultural and social worlds in which people are embedded underpins the decisions they make every day (Boholm et al. 2013). Furthermore, there is the question of who is invited to participate—that is, whose voices, values, social and cultural experiences, and priorities are welcomed to the table?

To date, very little research has examined the characteristics of farmers who participate in OFR in the US context. The most comprehensive research on farmer characteristics and attitudes related to OFR comes from survey data collected with farmers in 1993 (Goodwin et al. 1997; Norman et al. 1998), and more recently by Edward C. Luschei and colleagues (2009). Analysing survey data collected with three groups of Kansas famers, B. K. Goodwin et al. (1997) found that those most likely to participate in OFR were younger, had a higher net income, and had greater crop diversity. Analysing the same data, D. W. Norman et al. (1998) found differences among the groups in terms of which collaborating organisations farmers felt best met their needs, and with which organisations they preferred to work. Specifically, farmers from the group most associated with sustainable agriculture (who were also younger, had higher education, had farmed for fewer years, farmed fewer acres, and had more off-farm employment)⁵, preferred to work with a sustainable agriculture organisation. These farmers were less confident about (and less reliant on) university-based research and extension, and even less confident about research with commercial firms. Across groups, farmers supported OFR and expressed a willingness to participate, but the farmers associated with the sustainable agriculture organisation had a stronger preference for greater collaboration in research design and implementation than the other two groups.

To my knowledge, only Luschei et al. (2009) have specifically examined the potential for participation bias in OFR in the US context. As a result of a failed effort to recruit a ‘random’ sample of Wisconsin farmers for a weed management project, Luschei et al. (2009) compared survey responses from their on-farm collaborators to a larger, random survey of Wisconsin farmers. Overall, they found that the differences between groups can be explained by the recruitment criteria for the on-farm trial.

Notably, when it comes to which farmers are engaged in OFR, several papers emphasise the role of prior relationships between farmers and the researchers or agricultural professionals (including Extension) (Goodwin et al. 1997; Kladivko et al. 2019; Thompson et al. 2019). In most cases, this relationship is framed as an advantage on which researchers should capitalise when enlisting farmers into OFR. In one of the very few papers to detail their OFR recruitment process, Eileen J. Kladivko and colleagues list the production and management criteria for potential ‘farmer-cooperators’ in a soil health research, demonstration, and training project. They further detail that their initial recruitment list ‘was generated from personal knowledge of the core team plus input from two partnering organizations’ (Kladivko et al. 2019: 12A).

This is also true within PSA. When I asked members of our OFR team to describe how they identify and recruit farmers to participate in our OFR, they overwhelmingly described working within their established networks. One member of our team, for example, explained that they were ‘looking for growers that are already involved with … farm trials, as well as targeting growers that are involved with [a local conservation agriculture organisation]. We have several county agents conducting on-farm trials, which allows us a relatively easy-in with growers.’ Another said that they identified potential on-farm collaborators by ‘talking with farmers I've met over the last 14 years. … Growers that you're having a dialogue with, who I've met at events, or when they call me, or when I talk to them on twitter.’

As anthropologists, I'd argue that we are deeply empathetic to this perspective; we understand that good ethnography is built upon strong relationships with key informants, developed and nurtured over time. These ‘thick’ and enduring relationships offer unique insight, and we rely heavily on the networks of our interlocutors to gain access communities and cultures. While I do not aim to undermine the value of these rich networks and relationships, if we have scientific goals related to the generalizability of our models and predictions, and societal goals focused on serving a broad population of farmers, then we must be extraordinary reflexive about the benefits and limitations of this approach to farmer engagement.

Epistemological Concerns and Pragmatic Consequences

At this point, we must consider the epistemological consequences of the dearth of knowledge about farmer participation in OFR. What we face is essentially a problem of ‘map-territory relation’—that is, the ever-imperfect relationship between a representation of an object (the map or model), and the object itself (Korzybski 1933). Of course, the whole point of a map or model is to create a simplified representation of reality. This is precisely why it is useful. But the process requires identifying and including the factors that matter (to whom? and for what?) and excluding those that do not. The factors deemed worth collecting, recording, and analysing become ‘data’. Yet, data—what constitutes data, how we capture it, and what we do with it—is deeply embedded in its sociality (Douglas-Jones et al. 2021).

In the context of OFR, the focus has been squarely on collecting agronomic data. Nevertheless, embedded in each data point are webs of social relations and power—including farmers’ histories with the landscape, their relationships to other farmers, and their relationships (or the lack thereof) with agents distributing state and federal agricultural resources. As noted above, the limited data we have about farmers collaborating in OFR suggests that they tend to be younger, better educated, and wealthier than the general farming population (Goodwin et al. 1997); and they tend to have prior relationships with researchers (Kladivko et al. 2019). While there is no data to my knowledge on the gender or race of OFR collaborators in the United States, the representation gaps identified in medicine index the potential consequences of this situation (Clark et al. 2019; Flores et al. 2021; National Academies of Sciences and Medicine 2022). To wit, a 2021 review of participation in ten years of vaccine trials, for example, found that Black, Indigenous, and Hispanic/Latino individuals, and older adults, were significantly underrepresented compared to the US population (Flores et al. 2021). And while representation gaps present epistemological concerns, they also have pragmatic consequences. For example, a recent National Academies report (2022) argues that underrepresentation in clinical trials has serious consequences for health—by compromising the generalizability of the research itself, limiting the potential for emergent knowledge, and restricting access to novel and experimental therapies⁶. These direct consequences also exacerbate health disparities, their related financial and social costs, and distrust in science and research (National Academies of Sciences and Medicine 2022).

In the case of OFR, a lack of data about collaborating farmers both prevents researchers from confronting the limitations of our knowledge and limits the validity of the knowledge itself: what we think we know about farmers’ needs, priorities, and practices might not be true, for example, for farmers who do not have prior relationships with researchers. However, this blind spot also limits the utility and value of our practical recommendations: if our recommendations align only with the needs, priorities, and practices of the wealthiest, most educated, and most connected farmers, we might lose our best opportunity to truly transform US agriculture at scale. Further, by materially encoding these blind spots into emerging technological models and tools, and we are likely to reinforce dominant regimes of power and privilege in agriculture (Montenegro de Wit and Canfield 2023).

Beginning to collect and report data about collaborating farmers is a critical precursor to broadening participation in OFR⁷. We don't know whose perspectives and experiences are missing until we first look at whose are represented. This will not be easy, but it is the right thing to do—for scientific reasons but also to expand the benefits of participation beyond ‘the usual suspects.’ Interviewing 40 Nebraska farmers (all men) who participated in OFR over the last 25 years, Laura J. Thompson et al. (2019) found that OFR offers a series of scientific and practical benefits, which include providing experiential learning for farmers, increasing the credibility of the research for other farmers, building farmer peer-networks, and strengthening the link between farmers and researchers which creates further opportunities for farmers to guide research priorities. These are benefits to which all farmers—not just the most well-connected—deserve access.

Anthropological Engagement with Data Science

Since the 1990s, research funding agencies have increasingly required large-scale research projects to include ‘human dimensions research’ to ensure their societal relevance and transferability into practice (Gibbons et al. 1994; Nowotny et al. 2001). More recently, upstream efforts to integrate societal values and priorities into scientific research and technology development have increasingly become known in the United States and European Union as ‘responsible research and innovation’, or R(R)I (Owen et al. 2012). Fisher et al. (2015) argue that socio-technical integration has the potential to bridge disciplinary divides, catalyse mutual learning, and transform engagement around societal issues. Nevertheless, social scientists have had a somewhat fraught experience within this space—often finding that projects are driven by scientific development, with the social science pushed to the margins (c.f., Lyle 2017). Further, social scientists may be reluctant to engage in projects where their presence is dually framed as either ‘adversarial armchair critique’ or ‘co-opted uncritical service’ (Smolka 2020: 1). Nevertheless, echoing Marilyn Strathern (2006), I argue that there is an urgent need for anthropologists ‘roll up our shirtsleeves’ and engage in interdisciplinary work with data science.

Considering the response of anthropologists to the emergence of data science, Rachel Douglas-Jones and colleagues observe that ‘anthropology has largely responded to the data moment by figuring ethnographic fieldwork as a necessary or more sensitive qualitative complement to large-scale data collection and analysis’ (2021: 11). Drawing on Clifford Geertz, Tricia Wang first argued that ‘big data needs thick data’ (2013). Elizabeth F. S. Roberts elaborated this point, making the argument that ethnography is ‘a kind of “big data”’ that can enhance the overall quality of research (2021: 356); and Andreas Bjerre-Nielsen and Kristoffer Lind Glavind suggest that ethnographic data can serve to ‘ground truth’ big data, enhance it through ‘thick description’, and provide insight into its ‘hidden dimensions’ (2022: 2).

In the context of digital agriculture, Maywa Montenegro de Wit and Michael Canfield similarly argue that ‘small-n narratives are thus critical for giving meaning to data’ (2023: 9). Within PSA, my team is doing this. Since 2017, we've collected over one hundred in-depth interviews with farmers and agricultural advisers and are developing a nuanced understanding of farmers’ values, experiences, networks, and mindsets toward conservation practices, particularly cover cropping. As we analyse and interpret these data, our next steps are to (re-)engage with farmers and other researchers to co-develop strategies to support their adoption of agroecological practices, while broadening research participation and access.

But we must not stop there. In the context of interdisciplinary engagement with data science, we anthropologists must use our seat at the table to practice what Helen Verran (2001) called ‘generative critique’ in support of responsible research and innovation (Hillersdal et al. 2020; Niewöhner 2016; Smolka 2020; Smolka et al. 2021). Our training in engaged yet critical participant observation, and our experience building long-term relationships of trust within communities, makes anthropologists particularly well-suited for this role. Within interdisciplinary data science projects, we can pay attention to embodied moments of ‘disconcertment,’ which signal ‘different ways of making an object, different ways of knowing, and different ways of being in the world’ (Smolka 2020: 13). And then we must find ways to productively communicate these moments of disconcertment to our interdisciplinary colleagues.

As I allude to above, early in my work within PSA, I experienced disconcertment as I became aware of the black box of OFR. Feeling the tension of disconcertment rise in my body during on-farm project meetings, I began to ask questions aimed at understanding my colleagues’ unspoken assumptions about our collaborating farmers’ experiences and priorities, how well they represent row crop farmers more generally, how this might affect our models, technological tools, and outreach, and how this might affect the broader goals of our work. Over the last year, I've begun to share my deep dive into OFR with my colleagues on this project, and I have partnered with another social scientist on the team to compare the survey data she has collected with our farmer-collaborators and with a random sample of farmers. When we shared our preliminary results at our annual project meeting last year, it spurred interest, as well as disconcertment, among colleagues. In an email following this meeting, one colleague wrote:

I wanted to offer my perspective regarding the engagement of ‘random farmers’ versus the usual suspects… Most, if not all, of the state extension specialists I know carry the weight of responsibility for farmers who are not engaged with their universities. We know the benefits of engagement would be mutual. […] Extension specialists and agents invest years in developing relationships with farmers. But, frankly, there are many times we are turned away—some folks don't want to work with the university, no matter who comes knocking. … sometimes we are provided with a reason (no time, no money, no labour), and I imagine there are additional reasons they aren't willing to engage. … The farmers who avoid workshops, field days, etc., aren't random. We know who they are. Some are warming up to relationships, but others are a firm no. We keep trying. Everyone should have access to information and support. […] I love working with farmers, but it will likely take more time and funding if we want to engage the previously unengaged.

In contrast to either adversarial critique or uncritical co-opted service, generative critique offers an opportunity to nurture mutual learning and greater responsivity among collaborators (Smolka et al. 2021)—to make the work more relevant, more inclusive of diverse experiences, opinions, and priorities. As of this writing, we have not yet solved the challenge of broadening on-farm participation. As my colleague suggests, there's no shortcut to building trust (whether among interdisciplinary colleagues or with farmer-collaborators), but we have submitted new proposals that would allow us to broaden our on-farm networks and deepen participatory engagement within them. However imperfect, engaging in the practice of generative critique has deepened interdisciplinary collaboration within this data science project, and driven us to (re-)consider how we might collaboratively produce knowledge, models, and tools that are more just, more inclusive, and that support more farmers to effectively adopt more sustainable agricultural practices.

Conclusion

I argue that the lack of scholarly attention to the characteristics, experiences, and values of farmers participating in OFR raises important epistemological questions, as well as pragmatic and equity concerns that need to be addressed as we develop technologies and tools intended for broader audiences. Whether working in the field of agriculture or beyond, anthropologists engaging with data scientists can draw on our training in participant observation to further responsible research and innovation. Beyond providing ‘thick’ ethnographic data to complement largely decontextualised ‘big data’ (Wang 2013), anthropologists are uniquely suited to draw on ‘moments of disconcertment’ within interdisciplinary encounters to offer ‘generative critique’ (Smolka 2020; Verran 2001). This is work we should embrace.