Key questions on the use of big data in farming: An activity theory approach

Abstract

Big data represent a pioneering development in the field of agriculture. By producing intuition, intelligence, and insights, these data have the potential to recast conventional process-driven agriculture, plotting the course for a smarter, data-driven farming. However, many open issues about the use of big data in agriculture remain unanswered. In this work, conceptualizing smart agricultural systems as cyber-physical-social systems, and building upon activity theory, we aim at highlighting some key questions that need to be addressed. To our view, big data constitute a tool reciprocally produced by all the actors involved in the agrifood supply chains. The constant flux of this tool and the intricate nature of the interactions among the actors who share it complicate the translation of big data into value. Moreover, farmers’ limited capacity to deal with data complexity, along with their dual role as producers and users of big data, impedes the institutionalization of this tool at the farm level. Although the approach used left us with more questions than answers, we suggest that unraveling the institutional arrangements that govern value co-creation, capturing the motivations of farmers and other actors, and detailing the direct and indirect effects that big data (and the technologies used to generate them) have in farms are important preconditions for setting forth rules that facilitate the extraction and equal exchange of value from big data.

Keywords:

• Big data
• Smart farming
• Value
• Farmers
• Cyber-physical-social systems
• Activity theory

1 Introduction

In the late 1990s, participants in a symposium on precision agriculture characterized smart farming as a high knowledge-intensive form of practicing agriculture, which uses insights from both basic and applied science to capture the effects of problems and actions in space and time, thus having the ability to enhance the productive potential of farm enterprises while, in parallel, reducing the environmental externalities of agriculture (Lake et al., 1997). Despite its vagueness, this definition indicates that smart farming is much more than the fine-tuning of technology and other inputs, which is the case in precision agriculture. More than two decades later, the interest in smart farming has grown apace, fueled by the huge opportunities new technological developments offer today. Using farm robotics, Internet of Things, and cloud computing capacities, and exploiting the interoperability of databases and the use of real-time data, smart farming can improve farm management, while ensuring the maintenance of natural environment (Raja et al., 2018; Jayaraman et al., 2016; Wolfert et al., 2014).

Smart farming can be seen as the end of a continuum ranging from traditional, labor-intensive forms of agriculture, where past experience and intuition are used to inform farmers’ decisions, to technology-intensive, data-supported forms of precision agriculture, where field-specific data are used to help farmers make appropriate decisions on the production process (Kritikos, 2017), to a data-intensive form of farming, where smart devices and intelligent decision support systems can exploit the potential offered by big data, leading to data-driven decision making. Hence, smart farming can be conceived of as a data-driven form of agriculture, in which decision making processes are based on explicit information derived from big data collected through heterogeneous sources (Wolfert et al., 2017).

The use of big data for guiding farmers’ decision making is a real revolution which has the potential to transform current “process-driven agriculture” – where the emphasis is in the process of using scarce resources to produce crop and livestock products – to a new, data-driven agriculture. To better illustrate the difference between process- and data-driven (smart) farming let us take a look at an example. Summer 2017 in some Mediterranean regions was characterized by very high temperatures and high-precipitation events. Such a combination offers a favorable environment for grapevine diseases like downy mildew, which can significantly reduce both the quantity and the quality of the produced wine. Indeed, in Italy, Spain and France wine production dropped to a historic low (Organisation Internationale de la Vigne et du Vin, 2017). Being traditionally based on process-driven approaches, most Mediterranean viticulturists experienced wide departures from normal yields, since they followed the standard plant protection measures and agronomic strategies. On the contrary, having access to weather data, and constantly monitoring temperature, soil moisture, leave hydration and relative humidity data, or even having the potential to receive downy mildew alerts and advices – facilities offered by some smart systems (European Commission, 2017; Pérez-Expósito et al., 2017) – data-driven viticulturists might have the opportunity to adjust their phytosanitary treatments so as to increase the effectiveness of pesticide interventions, and to reduce both their spraying costs and the environmental impacts of fungicides. In other words, although in process-driven agriculture viticulturists failed to achieve high efficiency while maintaining product and ecosystem value, data-driven decisions helped farmers reducing the production losses, lessening the pesticide costs, and minimizing the environmental impact of plant protection.

Of course, the notion that data are important for informing farmers’ decisions is not new. Price- (Kruse, 1958) or production-related data (Aakhus, 1947) were always considered valuable inputs in the decision making process. As Brown noted in the mid-1950s, availability, timeliness, and accuracy of high volume datasets can secure the viability of agrifood supply chains (Brown, 1955). Nevertheless, although “traditional” data – i.e. information that can be transferred between data originators, providers, users or third parties (Copa-Cogeca et al., 2018) – have the potential to show past or present trends, big data, by taking advance of interconnectivity (Zikopoulos and Eaton, 2011), datafication (Walker, 2014) and digitization (Loebbecke and Picot, 2015), can combine complex and large datasets (Bajaj and Ramteke, 2014) thus offering entirely new sources of current and detailed information derived not only from structured but also semi-structured or even unstructured data (Franks, 2012). This amalgamation of real-time multi-sourced information – derived not only from on- but also from off-farm data (Liang et al., 2018) – offers farmers better opportunities to diagnose potential risks, to identify possible alternatives, and, more importantly, to predict future occurrences than traditional analytics which are based on past data. Consequently, big data can supply farmers with quick-witted intelligence, this way transforming traditional (process-driven) agricultural systems to smarter, data-driven systems.

Although the original definitions of big data emphasized their great volume, today the term encompasses not only the amount of data available but also the technologies and expertise used to generate, collect, store, analyze, present and use these data (Yang et al., 2017). Big data are characterized by their great volume (size of the datasets), their huge variety (heterogeneity of data types and sources) and their substantial velocity (speed of data generation) (Gandomi and Haider, 2015). These three characteristics (commonly referred to as “the three Vs” in the literature) permit the emergence of hidden value from big data (Hashem et al., 2015). Having the potential to glean important and applicable information from high-volume datasets (Sagiroglu and Sinanc, 2013), big data analytics can produce new forms of value, by helping actors make smarter, faster and impactful decisions (Ahmed et al., 2015) through informed intuition, intelligence, and insight (Abdrabo et al., 2016). According to Padgavankar and Gupta (2014):

-	informed intuition refers to the ability to predict future occurrences and identify appropriate actions,

-	intelligence relates to the offering of solutions to real-time problems,

-	insights emerge from the process of reviewing occurrences and actions.

Moreover, big data can bring together different actors, through value-generating information. Farmers, processors, manufacturers, agrochemical companies, and technology suppliers (AgTech companies) are all involved in and affected by the use of big data in agriculture (Bronson and Knezevic, 2016), whereas researchers, practitioners and policymakers can draw important conclusions and design appropriate intervention schemes when needed (Chen et al., 2012). In this vein, big data have the potential to enhance collaboration among different nodes of the agrifood supply chain, facilitating the expression of what some scholars term connective action (Cieslik et al., 2018; Leeuwis et al., 2018).

However, a crucial question is whether these data can really lead agriculture to a sustainable future. Despite the considerable amount of literature dealing with the issue, today our comprehension of how to use big data to ensure the sustainability of agrifood production is still at an embryonic stage. In the early discussions on smart farming, the ability of smart technologies to improve agricultural efficiency occupied a central position (Porceddu and Rabbinge, 1997). Indeed, later works confirmed that smart (data-driven) farming has the potential not only to increase farm production and efficiency (Borchers and Bewley, 2015), but also to help farmers minimizing microbiological (Oliver et al., 2017) or disaster-related (Řezník et al., 2017) risks, to reduce fertilization costs (Bendre et al., 2016), to facilitate crop disease management (Garg and Aggarwal, 2016), to minimize depletion of natural resources (Gupta et al., 2016), to mitigate climate change (Scherr et al., 2012) and to enhance food security (Ribarics, 2016). However, already from the very early stages of smart farming expansion, doubts about its potential impacts on ecological sustainability were raised (Athanasiou, 1998). Today, many scholars express several concerns on a broad range of social and ethical issues associated with big data and smart farming (Bronson, 2018; Eastwood et al., 2019; Carbonell, 2016). Kritikos (2017) provides a wide list of socio-ethical imperatives associated with the use of data in agriculture, including both data-related perils – such as dependency risks, potential lock-in effects, data concentration, and transformation of farmers into information tools – and sustainability challenges.

Notably, as a research topic, the issue of big data is characterized by diverse and, oftentimes contradictory arguments. Hype and confusion abound in the scientific debate (Mithas et al., 2013). Depending on the point of view, different features of big data are emphasized. Hence, although some claim that big data can provide answers to the most pressing problems generated by process-driven agriculture, others speak about a dark side of big data. Nevertheless, to better grasp the effects that big data can have on global farm architecture, it is important to clearly conceptualize not only how these tools are used, but also by whom they are used and what determines their ability to transform farming.

In this work, our purpose is to decipher the ways big data connect farmers, farms and agrifood systems, and to provide a list of questions that can open up new research agendas in the field of smart farming research. Building upon activity theory (Engeström, 1999), and drawing from different fields of study, we attempt to sketch a new framework for understanding how farmers interact with big data with the aim to reach specific outcomes related to their farms. We also discuss the mediating role of the wider network of actors who share the work of production and use of big data in these interactions, and the ways formal and institutional rules catalyze the application of big data in agricultural systems. In doing so, this study moves beyond prior work in this area, by shifting the focus from the ways big data can be applied in farm settings to the interactions between the components of activity systems that big data mediate, and by setting the scene for identifying caveats and new research lines in the field of smart farming.

2 Smart farming as a cyber-physical-social system

In their highly cited work, Wolfert et al. (2017) conceptualize smart farming as a “cyber-physical system” in which smart devices (the “cyber” dimension of the system) are the locus of control for the farm (the “physical” dimension). Of course, in such a system humans still undertake different activities – like the analysis of data and the planning – but different appliances perform most of the tasks required to keep the system operating. However, in our view, this system also entails a social dimension, emerging from the fact that cyber-social information (social information exchanged within the cyberspace) is inextricably linked with big data (Barnaghi et al., 2015). Actually, there is little – if any – doubt that the interactions between individuals and/or social groups constantly change the cyberspace through the production of new data (Zeng et al., 2019). Moreover, human action is a necessary condition for transforming big data and the farm, while, at the other end of the spectrum, humans continue to be in charge for the decision making processes – since some indications confirm that even sophisticated and highly intelligent decision support systems are differently approached and used by farmers (Lundström and Lindblom, 2018; Alvarez and Nuthall, 2006) – which, in their turn, further alter the physical and the cyber dimensions of the system. Consequently, we argue that it is more precise to define smart agricultural systems as “cyber-physical-social” systems (Sheth et al., 2013).

The social dimension of this system refers to the sum of entities which provide and use data, that is all the immediately involved actors in the supply chain of agrifood products – farmers, manufacturers, wholesalers, distributors, and consumers – but also intermediately connected actors who might be involved in the various upstream and downstream flows of goods, services, resources and capital, such as intermediate suppliers and customers, collaborative or competitive organizations, institutes or communities, as well as a cloud of “external” actors like governmental organizations, research and policy institutes, banks, etc. (Lioutas et al., 2018; Mentzer et al., 2001). Big data link all these actors through shared cognition – the ability to build institutional connections that help them interpret cues in a similar way (Wertsch, 1991; Cannon‐Bowers and Salas, 2001) – thus augmenting their ability to produce meaning necessary for sustaining systems performance. Hence, big data are used as glue, connecting the physical and the social dimension of the system.

3 A brief insight into activity theory

Activity theory (or cultural-historical activity theory) draws from thoughts developed by a group of thinkers belonging to what today is called the School of Russian Social-Psychology. The two main exponents of this school – Leontyev (1978) and Vygotsky (1978) – aimed at elucidating and explaining the relationships between “subjects,” “objects,” and “tools” used to transform these objects. Subjects, objects and tools constitute what can be termed “activity system.” In plain words, every activity – which consists of a set of intentionally performed goal-directed actions (Roth, 2007) – can be captured as an interaction between a subject and an object with the aim of transforming the object through the use of various tools (Russell, 1998). This conceptualization is also evident in interactional psychology, where an individual’s actions are considered as the outcome of her/his interaction with the specific situations she or he encounters. In this vein, individuals have to be conceived of as active agents, who intentionally engage in an interaction process (Endler and Magnusson, 1976).

Engeström (1987) expanded upon Soviet scholars’ work, by designing a more complete theoretical framework of human activity. Instead of some related theoretical foundations, Engeström’s (2005) view is based on a systems approach to conceptualize human (intentional) activities, without considering actors as “black boxes” lacking internal properties. This philosophy emphasizes the continuous process of transformation across time; a process that extends beyond individual levels, thus affecting and being affected by a wide array of actors (Engeström, 1999). Following this line of thinking, current conceptualizations of activity theory have moved beyond the issue of mediation through specific tools, to incorporate diversity and interactivity between activity systems (Bakhurst, 2018).

According to activity theory’s basic premises, individual activities are “practical,” take place in social contexts, and can be mediated by various artifacts, i.e., technological developments, laws, methods, practices, etc. (Engeström and Middleton, 1998; Kuutti, 1996). To produce any result or to transform an object to an outcome, a subject – after developing a level of motivation – uses the available tools ¹ to effectively interact with the object. However, the activity is also influenced by a community of actors who share a common interest in the process. A set of rules intermediates the subject-community relationship, whereas a certain division of labor regulates the interrelation between subject, object and the community (Engeström, 2001; Nardi, 1996) (Fig. 1).

Fig. 1 The structure of an activity in the framework of activity theory – adapted from Engeström (2001).

Activity theory has drawn much attention from theorists and researchers interested in explaining the ways technology can be exploited within activity systems in the field of education (Isssroff and Scanlon, 2002), ergonomics (Engeström, 2000), transportation (Ettema, 2018), and so on. Indeed, as Kaptelinin and Nardi (2018) explain, activity theory can frame the human-technology interaction with a meaningful context, offering opportunities to better capture the ways technology affects – and is affected by – individuals and groups, as well as illustrating the real meaning of technology for people. The application of activity theory in the study of human-technology interaction represents a shift in scientific focus from the technological inventions to the ways human actors interact with technology and with each other within a framework determined by specific requirements and constraints (Bannon, 1995). Therefore, activity theory, by depicting “the doing of the activity in a rich social matrix of people and artifacts” (Kaptelinin and Nardi, 2006, p. 9) can shed new light on the ways big data influence the interactions between farmers, farming, and agrifood systems.

4 Applying an activity theory approach to reframe smart farming

To frame our analysis we used the model developed by Mwanza and Engeström (2005), widely known as the eight-step model. According to this model, to operationalize the activity and sub-activity triangles (Fig. 1), one needs to define the activity under investigation (1 st step: Activity), to understand the “why” behind the activity (2nd step: Objective), to identify the actors who perform the activity (3rd step: Subjects) and the tools that mediate this activity (4th step: Tools), to determine the rules that constrain and regulate the activity (5th step: Rules and regulations), to understand how labor is divided among the actors who participate in the activity system (6th step: Division of labor), to define the community of actors involved in the activity (7th step: Community), and to identify the expected outcome of the activity (8th step: Outcome). Some brief descriptions of the terms used in the eight-step model for the case of smart farming are provided in Table 1.

Table 1 The eight-step model of activities (adapted from Mwanza and Engeström, 2005, p. 459) in the case of smart farming.

Step	Focal point(s)	Description
1	Activity	The optimization of farmers’ decision making process, so as to achieve better product quality and high product quantity with lesser cost and with a more targeted use of resources
2	Objective	The motives that drive farmers to use the available tools so as to transform the object of activity (the farm) into valuable outputs
3	Subjects	Farmers
4	Tools	Big data (encompassing both data and technologies used to generate them) and previous knowledge
5	Rules	Legislative rules that regulate the use of big data and institutions that govern the relations among the actors involved in the activity system
6	Division of labor	The ways the responsibilities of value creation are distributed within the community
7	Community	The wide array of actors involved in the supply chain of farm products and data (including consumers)
8	Outcomes	The production of high-quality products in high volumes, ideally with respect to the environmental, social and economic sustainability of the agrifood systems

As Mwanza (2001) notes, defining the elements of an activity system helps us answering a series of questions that focus on the interrelations between systems’ components. These questions (presented in Table 2) drove our analytic plan. The first of the two main sections that follow is dedicated to analyzing the subject-tools-object triangle (Fig. 1). Our analysis starts with a brief description of an activity system in the case of process-driven farming (Section 4.1.1), so as to help us better illuminate what differentiates big data from conventional technologies. Then, we present the special characteristics that account for the very nature of big data (Sections 4.1.2 and 4.1.3). In the next sections we shift our focus on the relation between tools and subjects, examining the ways big data affect farmers’ decision making process (Section 4.1.4) and analyzing the issue of data ownership (Section 4.1.5). The Section 4.1.6 aims at scrutinizing the issue of farmers’ motivation. Finally, we turn our attention to the relation between big data and the farm (Section 4.1.7). The second part of the analysis is devoted to identifying what community means in the case of smart farming (Section 4.2.1), to examining the ways the division of labor affects production of and access to value emerging from big data (Section 4.2.2) and to discussing the issue of rules that regulate (or should regulate) the activity (Section 4.2.3).

Table 2 Questions on the interrelations within the activity system.

Components of the activity system	Questions to ask
Subjects and Tools	What tools do the subjects use to achieve their objective and how?
Subjects, Community, Division of labor, and Tools	How does the division of labor influence the way the subjects satisfy their objective? How does the division of labor affect the way the community achieves the objective?
Rules and Subjects	What rules affect the way subjects achieve the objective and how?
Community and Rules	What rules affect the way the community satisfies their objective and how?
Tools and Community	How do the tools in use affect the way the community achieves the objective?

Adapted from Mwanza, 2001.

4.1 Subjects, objects and tools

4.1.1 Activity systems in process-driven farming

In the case of process-driven farming, farmer – the subject of the activity – reciprocally interrelates with the object (the farm) to produce certain outcomes. It is this attempt to yield specific results that initiates every action. Traditional farm technologies (tractors, combines, fertilizers, pesticides, irrigation systems, etc.) are tools that mediate this process, facilitating outcome attainment. Nevertheless, apart from technological instruments, the term “tools” in activity theory also encompasses the subject’s knowledge, skills and competencies (Engeström, 2000). These internal tools correlate with technologies affecting the mediation process, determining in this way the length of the interaction between subject and object.

4.1.2 Big data as tools: Inseparability of production and consumption

In the case of smart farming, the interrelations between farmers, farms and big data are quite more complex than in process-driven agriculture. Farmers occupy a double role in the activity system. First, they are users of data, since they utilize the results of data analysis to deal with real-farm problems (Evans et al., 2017). Second, although farm-related data refer to issues pertaining to crop and/or livestock responses to various physical conditions, growth factors and climatic variables, farmer’s management practices heavily affect the information content of these data. In this vein, farmers are also co-producers of data. Hence, they simultaneously construct physically (through their decisions and actions) and socially (through the different meanings they ascribe to these data during use) big data. This peculiarity generates a different form of “duality.” As Orlikowski (1992) explains in her seminal article, tangible technologies are created by some actors and then used by others, which reify and institutionalize them, through embedding them into specific organizational structures. In the case of big data, however, farmers’ dual role complicates the process of institutionalization, since every action made by a farmer influences both the quantity and the quality of big data. This raises an important question:

Q1. How does the duality of farmers (as data users and co-producers) influence the institutionalization of big data?

Remarkably, the inseparability of production and consumption is a key feature differentiating big data from any other technological development ever appeared in the field of agriculture. For instance, the way a farmer uses her/his tractor affects the tractor only in the long-term, so the quality of the tractor remains relatively stable when the analysis is restricted to a limited time frame. On the contrary, differently from other – tactile – tools, big data are in continuous evolution (Chen et al., 2014), in the sense that every action generates new data, thus altering the nature of the tool. So, farmers have to continuously adapt to the constantly evolving nature of big data and to develop new knowledge and skills (internal tools) so as to effectively transform big data (external tools) into usable decisions. Nevertheless, it is questionable whether the farmer, as an individual agent, has the ability to deal with the ever-increasing complexity of the big data (Mittelstadt and Floridi, 2016). Absorbing the meaning of the complex insights offered by big data analytics is not an easy task, and lack of understanding these insights significantly restricts an actor’s ability to act upon them (LaValle et al., 2011). Moreover, even when information is rich and insightful, heuristics are often used to guide decision making processes (Klein and Thompson, 1984). Therefore, a germane question is:

Q2. Do farmers have the ability to deal with the ever-increasing complexity of big data?

4.1.3 Big data, bigger data and more cooked data

The dramatic increase of data generated by sensors and computers, not only in agriculture but in every arena of economic and social life (Khan et al., 2014), raises some new concerns about the quality of the information provided through data analytics. First of all, the size of data is not a synonym of data quality. As van den Hoven et al. (2012, p. 157) put it, “bigger data are not always better data.” Second, potential biases are more possible as the number of data increases. Although a variety of data cleaning applications emerged already from the early stages of big data explosion (Rahm, 2000), cleaning is in itself a subjective process which might lead to serious biases (Boyd and Crawford, 2012). Apart from cleaning, all the data handling operations carried out lead to a different form of data. As Bowker (2013) notes, data are never “raw” – i.e., data “generated and collected without editing or any processing” (Copa-Cogeca et al., 2018) – but rather “cooked,” in the sense that a degree of processing always exists. The term “processing” encompasses every operation performed on data or datasets, including, for example, data organization, structuring, adaptation, etc. (Copa-Cogeca et al., 2018).

When viewed within an activity theory context, this transformation of tools further increases the complexity of the activity system (Petersen et al., 2002; Bertelsen, 2000). Indeed, this transformation from a raw to a cooked form – owing to the conditions of collection, storage, and transmission – might heavily affect the ability to precisely interpret data and the capacity to draw usable conclusions, whereas it increases the complexity of the activity system and puts in doubt the accuracy of big data (Fan and Bifet, 2013). On the other hand, some scholars raise serious concerns about the power of various actors to hijack data during their processing so as to gain competitive advantages (Moura and Serrão, 2015). Consequently, the quality of big data is an outcome not only of processes used during data acquisition, preprocessing and analysis, but also depends on the intentions, orientations, and motivations of different actors. Here, a new question comes to the fore:

Q3. How does transformation (quantitative and qualitative) of big data affect the quality of information provided to farmers?

4.1.4 Big data and farmers’ decision making

In the growing body of the relevant literature, big data are often viewed as a tool that “guides” decision making (Phillips-Wren and Hoskisson, 2015; Gopalkrishnan et al., 2012). If that is true, then big data analytics provide specific opportunities of action, determining to a great extent farmers’ decisions. Indeed, studies converge to show that such tools can increase farmers’ decision making performance by offering specific solutions (Capalbo et al., 2017; Zhang et al., 2012). In this vein, big data open-up specific pathways of action, which farmers can follow to solve problems of their farms. Nevertheless, as Morozov (2013) states, this reliance on fixed norms of action might reduce actors’ creativity and experimentation, thus restricting their ability to discover and exploit – even by mistake – new paths. At the other side of the coin, as structurational models of technology suggest (Orlikowski, 2008), the solutions provided by big data are differently experienced and used not only by different actors but also by the same actor at different times. In this view, the ways big data are interpreted and used depends on a wide array of factors, including personal variables (beliefs, attitudes), characteristics of the task at hand, and sets of priorities (DeSanctis and Poole, 1994). So, a critical question is the following:

Q4. Do big data represent a tool that “guides” farmers’ decision making processes or do they offer abstract information that is differently elaborated and used by farmers?

To answer this question, one needs to shift focus from how big data impact a farm to how big data analytics are used within complex systems that entail social, technological and physical dimensions (Rose et al., 2018). Although, traditionally, technology is treated by researchers as an independent variable that affects, changes and transforms an object (Orlikowski and Iacono, 2001), it is perhaps more important to understand how subjects simultaneously interact with big data and objects (their farms). As Fisher et al. (2012) argue, to capture the potential of big data, a more concise focus on the experience of user interaction with these data is needed.

4.1.5 A shortlist of “hot” issues: Data ownership and data privacy

Notably, another characteristic that distinguishes big data from technologies used in process-driven agriculture is the lack of ownership. Although a farmer can claim that a tractor is “her” or “his” tractor, no one can support that big data are her/his “property.” So, every farmer has to deal with a tool that is collaboratively developed and used, but its ownership is unclear. This peculiarity of big data generates two major issues. First, despite their contribution to the development of the tool, farmers lack control over the data exchange process in terms of data customization. Bronson and Knezevic (2016) also stress this issue by observing that farmers do not have a voice in shaping the context of data collection, analysis, and use.

Second, as we explain in a following section, by eliminating individuals’ opportunities to set-up clear zones where other actors do not have access to their “own” data (Mai, 2016), the lack of ownership rights is associated with the issue of privacy (Shepherd et al., 2019), thus complicating the relationships between farmers and other members of the community (Kempenaar et al., 2016; Sykuta, 2016). The practice of signing agreements with technology providers creates a sense of uncertainty and perceived risk about the potential uses of farmers’ data (Desouza and Smith, 2014). Hence, despite the existence of regulatory frameworks and industry standards – at least in developed countries – on the use of big data (Kshetri, 2014), farmers are sometimes unwilling or even reluctant to share their data (Jakku et al., 2019). Although the issue of willingness traditionally receives less attention than the quality of the connection between different actors through data sharing practices, it is inextricably linked with the performance of the whole agrifood supply network (Fawcett et al., 2007). Consequently, a crucial question is:

Q5: How can we increase farmers’ active engagement in big data production and use?

4.1.6 Subjects and motives

To better capture the ways an activity is performed it is fundamental to understand what gives birth to subjects’ actions, by focusing on the motives that spur individuals to act upon an object with the aim of producing certain outcomes (Foot, 2014). In other words, a challenging task is to delineate how farmers’ motivations, goals, and intentions determine the limits within which big data and smart technologies can be fully exploited at the farm level. Several indications confirm that farmers’ actions have different economic, managerial, social, and also psychological precursors (Fang et al., 2018; Lioutas and Charatsari, 2018; Brown and Roper, 2017), while their motivational orientations guide their engagement in various activities (Charatsari et al., 2018; Triste et al., 2018; Charatsari et al., 2017). A more concise focus on the motivation behind adoption and use of big data is expected to shed further light on the process of farm transformation, since it can help us put action into contexts that integrate environmental, personal, and procedural factors (Szalma, 2009). As Engeström (2000, p. 964) puts it, motives give to actions “their ultimate continuity, coherence, and meaning.”

From an activity theory perspective, different motives – sometimes conflicting (Sannino, 2011) – can drive subjects’ behaviors. This type of “poly-motivation” (Kaptelinin, 2005, p. 6) raises some new conundra concerning the ways multiple motives emerge, the processes by which motives are generated within and beyond the need to transform the object, and the ways motives are associated with certain activities instead of other sets of activities (Kaptelinin, 2005). Moreover, social contexts catalyze subjects’ motivation (Allen et al., 2011), whereas oftentimes both personal and collective motives may operate in tandem (Hyysalo, 2005). Therefore, two questions emerge:

Q6. To what extent is the use of big data a self-determined and autonomous decision?

Q7. How is the motivation process reshaped and embedded within a community of actors?

4.1.7 Back to the basics: rethinking the farm

As the object of the activity, the farm drives while in parallel determines the length of any possible human action. Following Engeström’s objects-as-concerns view (2009, p. 304), a farm can be viewed simultaneously as a generator and a focus of “attention, motivation, effort, and meaning.” In this vein, the farm serves as the basis for the development of motivations and the setting of goals to be attained by farmers, whereas it is in continuous transformation and reorganization. Big data (the tools) are used to gradually transform the farm, thus facilitating the achievement of the desired outcomes (higher production, better product quality, more efficient farm management, reduction of environmental impacts, etc.). Nevertheless, as in any activity, potential negative outcomes are always possible.

However, the interrelation between big data and the farm represents the less discussed aspect of the activity system in the literature. Quite surprisingly, there is broader speculation about the legitimacy of using technology in process-driven crop and especially animal production than in data-driven farming. The limited emphasis on the issue probably owes to the intangible nature of big data. Nevertheless, big data have both direct (referring to the new allocation of resources due to investments in smart technologies needed) and indirect effects on the farm (e.g., the use of wearable sensors for data collection purposes in productive animals). Hence, most of the concerns over the negative effects of technology on farms are multiplied in the case of smart farming. Among the few studies addressing this dimension of smart farming, issues like animals’ freedom of choice (Driessen and Heutinck, 2015) or animals’ autonomy (Schewe and Stuart, 2015) hold a predominant position. The emphasis of big data analytics on the outcome (i.e. higher performance, lower effort) leads to an over-commodification of farms (especially of farm animals) which creates new ethics. This adds a new question to the discussion on big data in farming:

Q8. Do (and, if so, how) the novel digital ethics affect the quality of natural resources used in farm production?

4.2 Community, division of labor, and rules

4.2.1 Defining community

Perhaps the most challenging task when applying activity theory in the field of smart farming is to define what community means. A simple approach would be to describe the community in terms of access to the object. In this sense, the community is the group of persons who “hold” the farm (and, thus, they share the labor). However, within an activity system, access to the tool is also a characteristic of the community. Consequently, the term community can be thought of as a group of individuals who engage in processes of collective action (Ostrom, 2010, 2007) driven by the common interests they share on the object and the tools (Engeström, 2014; Bellamy, 1996).

In process-driven farming, the term “community” is liable to a wide array of interpretations, ranging from the farm family to farm enterprise (including the farm family and the personnel), to the actors that supply farmers with products or services aimed at helping them to effectively use the different tools so as to transform the object of the activity (the farm). These suppliers can be manufacturers of technologies, providers of repair services, advisors, extension providers or other actors involved in the field of agricultural production. Nevertheless, when the discussion turns to data-driven farming, the term is also used to denote the sum of players involved in the processes of producing, handling, and using big data. In other words, in the case of smart farming, it is not only the involvement in the production or translation of technology which defines the community. All the entities who provide, analyze and use data shape an informal community which is characterized by a remarkable heterogeneity, in that involved actors have significantly different aims and worldviews (Eastwood et al., 2019). Consequently, the term community encompasses the whole supply chain (i.e. the actors involved in the supply of inputs, products, and services, but also the entities which collect, store and process data) – also including consumers (Jay, 2007). Although the latter conceptualization of the community seems exaggerated, it is well known that nodes in a supply chain are firmly connected so that every change at a farm level generates fluctuations that increase in volume moving downstream the supply chain (Lioutas et al., 2018). In any case, the interactions between the community and the three nodes that constitute the upper triangle of activity theory’s model (Engeström, 2001) shape the effectiveness of the transformation process, thus deserving intensive inquiry. To put in context these interactions, we should first find answers to the following questions:

Q9. Which actors affect (and are affected by) the use of big data in farming?

Q10. How do different actors interrelate with farmers and with each other?

4.2.2 Distribution of power, labor and value

Traditionally, in the relevant literature, the issue of power is tied to that of control over data management processes (Carbonell, 2016; Andrejevic, 2014). Haire (2014) supports that today the distribution of value from big data is unequally allocated across agrifood systems, with farmers enjoying a limited share of it. In this perspective, the actors who collect and analyze data receive the lion’s share of value from big data. “Value” is the key-word here. To our view, what creates the power imbalance within a community is the uneven access not to big data but to the value emerging from them. Stubbs (2016) also argues that the central question in the discussion on data access is not who owns the data, but who owns the value of them. Besides, as Mittelstadt and Floridi (2016) note, although the notion of disempowered data subjects prevails, it is still unclear how data subjects – without having the skills and competencies needed – can exercise their access rights so as to independently exploit the potential of big data. On the other hand, albeit often portrayed by popular media as dark centers of power, big players in the field of smart farming are the only actors who seem to have the competence (both technological and operational) to handle and produce meaning from these data, thus having better access to their value.

Today, there is little doubt that the distribution of power within this community of actors follows the investments made by different actors in labor, technology, human resources and expertise (Andrejevic, 2014). In other words, those actors who have the capacity to develop a high level of competence in dealing with big data are the most privileged (Boyd and Crawford, 2012). Nevertheless, in this view, the value of big data is based on a neoclassical economics framework: farmers aim at maximizing the utility of big data, whereas AgTech companies, which process and analyze data, have as a central purpose the maximization of direct and indirect profits. Hence, these players “produce” value (and, consequently, enjoy the major share of it), whereas farmers “consume” this value or, as Vargo et al. (2008) put it, “destroy the value.” In other words, the value is considered as an asset that is to be delivered along with big data to their potential users.

However, within complex systems consisting of different actors, value is realized only when “consumers” (in our case the farmers) make use of a product or service (in this case, the big data) (Lusch and Vargo, 2006). In terms of current marketing thinking, all the actors participating in the community are linked with relationships aimed at translating big data into value. To achieve this purpose, actors integrate resources and work under the aim of providing value, not only for their own benefit but also for the other members of the community, thus engaging in “actor for actor relationships” (Polese et al., 2017). Such a view acknowledges the importance of all the involved actors in the process of value creation. To use an example, in the BIGt&u project – a Dutch consortium consisting of horticultural companies, growers associations, governmental agencies, information technology providers, and knowledge institutes – big data referring to all aspects of horticultural products marketing (consumer data, production data, and supply data) represent the glue that connects actors with different roles, aims and interests (Verwaart, 2015). Each one of these actors contributes data and/or work to the platform. In this sense, the labor is divided among all the involved members, although each actor undertakes different tasks. Nevertheless, this division of labor does not necessarily mean that actors cooperate with each other. It is well known that agrifood systems operate under conditions of coopetition (Walley and Custance, 2010), which means that actors have to cooperate while at the same time they compete with one another. Although current work in the field emphasizes the competitive nature of the relationships among community members and farmers, to capture the multidimensional tenor of value it is essential to focus on the ways actors reciprocally co-create value.

Importantly, the collectively produced and used big data can pave the way for the development of an alternative, customer-dominant logic (Heinonen et al., 2010), where actors who collect and handle data are viewed as “value facilitators” (to use the term of Grönroos and Gummerus, 2014) and farmers create their own value, by embedding big data into their social and/or professional contexts, activities, and practices (Heinonen et al., 2010). In addition, consumers of farm products can gain a more central position in the value ecosystem, since data on their preferences can be used to inform farmers’ crop choices. Going back to the BIGt&u example, insights from consumer data can help farmers adjust the allocation of their resources so as to correspond to the real market needs. However, the degree to which the value production process is consumer-driven, farmer-driven or controlled by other centers of power depends on the institutional arrangements that govern the activity system. Hence, irrespective of the actors involved in the activity system, there is a need to shift attention from the roles of individual players to the institutional factors that coordinate the community.

Therefore, researchers have the challenging task to define how value is co-created and what value means for each actor by moving beyond conceptualizations of value as the return of labor and invested capital. To borrow the terminology from systems ecology literature (Gren et al., 1994), the value of big data can be divided into primary and secondary. Primary value refers to the development and maintenance of the activity system; that is the ability of big data to connect the elements of the system, to ensure system’s productive capacity and to increase its resilience. For instance, helping farmers make sound and timely management decisions, big data can help them and other community members to avoid disturbances (weather hazards, market changes, etc.) by simultaneously reducing the economic waste associated with agricultural production. On the other hand, secondary value pertains to the outcomes that extend beyond the activity system. For example, by supporting the efficient use of natural resources (Lamrhari et al., 2016), big data can lead to more ecologically sustainable agriculture. The issue of value is quite multilayered and intricate, that makes it impossible to fully elaborate on it here. However, we offer a series of indicative questions:

Q11. What does value mean for farmers and other actors?

Q12. How is value (co-)created?

Q13. How do power dynamics and institutional factors shape the value (co-)creation frameworks?

4.2.3 Rules

The issue of rules is the most addressed in the literature on big data. As posited in the activity theory, the interrelations between the subject and the community are mediated through shared and agreed rules (Dissanayeke et al., 2014). Consequently, a fundamental element to be clarified refers to the norms, rules or other regulations which affect the activity under study (Mwanza and Engeström, 2005). Hence, an – old but always relevant – question is:

Q14. Which are the implicit and explicit rules affecting the way the activity takes place?

The creation of shared rules of thumb is as fundamental as complex in big data management. A huge amount of studies underlines several socio-ethical dilemmas associated with access to big data – but with a limited emphasis on big data value, as we discussed in the previous section. There is a consensus among researchers that if, on the one side, we are witnesses of a new era of agricultural production (where big data drive the evolution of farming), on the other side, privacy concerns occur (Whitacre, 2014). The four main concerns that need to be considered in rule setting are the issues of ownership, privacy, security and openness (Janssen et al., 2017; Stubbs, 2016; Sykuta, 2016). These questions are particularly urgent in supply networks dominated by big players like multinational corporations located upwards in the supply chain.

Nevertheless, another important task is to elucidate what rules are about. With the term “rule” we mean not only laws but also social, cultural and other patterns of interaction, as well as practices that influence the activity. Engeström (1987) assimilates rules to procedures, norms, and conventions that frame behaviors of members within an organized social scheme. This definition of rules draws on the institutional tradition, both the original institutional economics (Hodgson, 2007) and the new institutional economics (North, 1990), according to which institutions include both formal and informal conventions. Therefore, the definition of the rules needs to be contextualized on the basis of the three key “w-questions”: who, where, and when (Kébir, 2017):

Q15. Who defines the rules, where these rules are applied, and when (new) rules are needed?

The first strand of these questions refers to the key-actors involved in rule setting along the supply and data chain; that is the “stakeholder network” including big data users, companies specialized in data management, regulatory and policy actors (Wolfert et al., 2017, p. 71). What makes the setting of rules difficult is the growing number of players involved, both in data management (producers, data collectors, independent agricultural data banks, data cooperatives) and data uses (farmers and ranchers, retailers, industry groups, environmental interest groups) (Stubbs, 2016).

The second strand is related not only to the territorial dimension but also to the business context and to the institutional context. In the rich corpus of literature on the issue, the discussion on the very nature of agrifood supply chains holds an ascendant position. Supply chains in the agrifood sector are often dominated by the presence of large AgTech providers (actors that collect, aggregate, and process data), playing a monopolizing role with respect to the weak nodes of the chain, like farmers (Sykuta, 2016). The difficulty of rule design is drawn on the need for catching farmers’ trust in achieving agreements on rule identification and specification. This is particularly true in the family farm business, where the family sphere overlaps the farming one. Consequently, widening the horizons of knowledge access opportunities is a necessary step, through the involvement of all members of the activity system in processes of discussion, debate, and reflection (Ahonen et al., 2000; Engeström, 2001).

Finally, the third strand refers to the ways current or past negative/positive experience affects the patterns of rule setting. Paying attention to the historical evolution of the contexts within which economic activities take place can help us better capture the future trajectories of any activity (Welter, 2011). This dimension is linked with institutional arrangements, given that institutions – by determining the contexts of transactions and interactions (Vargo and Lusch, 2016) – shape the ways different actors experience their role in the activity system, thus offering the basis for the (re)designing of rules. When dysfunctional, these institutions impede the co-creation and sharing of value, whereas when instrumental facilitate the systemic attainment of common goals (Nelson and Nelson, 2002).

5 Discussion and conclusions

The discussion on the interrelation between big data and farming is shrouded by significant counterarguments. From the fear that an Orwellian future is near to the overenthusiasm for – and overconfidence to – the potential of big data in regenerating agriculture there is a wide variety of opinions and points of view. Our work, resting upon the premise that smart farming is a cyber-physical-social system, and applying an activity theory perspective, aimed at shedding new light on the ways the dimensions of data-driven agricultural systems intercorrelate. In doing so, the article provides a new series of questions, constituting not a new research agenda, but a map of areas in which new research agendas focusing on creating and answering different sets of sub-questions should be developed.

The fifteen questions emerged from our analysis (Table 3) pertain to different relations within the activity system. Big data – the tools in our system – not only facilitate the transformation of the farm through their enormous potential, but also mediate between farmers and communities, and between communities and farms, whereas they prompt – and are subjects to – the development of new rules, and lead to a reorientation of the division of labor in the agrifood system. Nevertheless, big data radically differ from conventional technologies like the tractor repeatedly used as an example in the previous sections, because they are in continuous evolution and are characterized by ever-increasing size and complexity. As Karanasios (2018) notes, these kinds of hybrid tools are produced by the concurrent action of subjects and communities, thus transgressing traditional notions of technology. The observation that farmers, although contributing to the development of the tool, don’t have full access to it seems paradox when we frame our discussion in terms of conventional technologies, but in the digital world it is a common ethos.

Table 3 Summary of the key questions on the use of big data in smart farming.

Interrelated components of the activity system	Key question(s)
Subjects and Tools	Q1. How does the duality of farmers (as data users and co-producers) influence the institutionalization of big data?
Q2. Do farmers have the ability to deal with the ever-increasing complexity of big data?
Q3. How does transformation (quantitative and qualitative) of big data affect the quality of information provided to farmers?
Q4. Do big data represent a tool that “guides” farmers’ decision making processes or do they offer abstract information that is differently elaborated and used by farmers?
Q5: How can we increase farmers’ active engagement in big data production and use?
Q6. To what extent is the use of big data a self-determined and autonomous decision?
Subjects and Community	Q7. How is the motivation process reshaped and embedded within a community of actors?
Tools, Rules, and Object	Q8. Do (and, if so, how) the novel digital ethics affect the quality of natural resources used in farm production?
Community and Tools	Q9. Which actors affect (and are affected by) the use of big data in farming?
Community and Subjects	Q10. How do different actors interrelate with farmers and with each other?
Subjects, Community, Division of labor and Tools	Q11. What does value mean for farmers and other actors?
Q12. How is value (co-)created?
Q13. How do power dynamics and institutional factors shape the value (co-)creation frameworks?
Rules and Tools	Q14. Which are the implicit and explicit rules affecting the way the activity takes place?
Community and Rules	Q15. Who defines the rules, where these rules are applied, and when (new) rules are needed?

However, access to big data – albeit associated with the creation of digital divides in the relevant literature (Crawford, 2013; Nunan and Di Domenico, 2013) – is of less importance when compared with the question of access to the value derived from the use of big data. Surprisingly, the issue of value has attracted only a limited share of attention by scholars. Not rarely, in lieu of a more accurate way to defining it, value is conceived of as a synonym to farmers’ decision making performance. To our view, value can be distinguished into primary and secondary, where the former refers to the value enclosed within the activity system and the latter to the value extensions beyond the system. Such a differentiation could help researchers assess the value of big data not only within but also outside the zone of the activity system. Nonetheless, in networked systems, value takes different forms, as it is uniquely and phenomenologically experienced by the beneficiaries (Vargo and Lusch, 2017). So, a critical task for scholars is to clarify what value means for farmers and other community members, and then to examine how power dynamics affect access to this value. A first step to this end should be the identification of actors who collaboratively produce both the big data and the value extracted by them. A more concise focus on the community formed by these actors – and the institutions that it nests – can deepen our understanding of the ways interrelations and everyday practices lead to the co-creation of value.

Perhaps it goes without saying that these actors have different purposes, foci, and intentions. However, what is common within the activity system is the aim to produce value from big data. Nevertheless, the institutionalization of data is a provocative conundrum, especially for farmers who often lack the competence to deal with the labyrinthine nature of big data. To unravel the mechanisms with which farmers transform big data into value we need to illuminate the process of combining conventional wisdom with the insights offered by data analytics. The answer to this riddle cannot be simple, since each farmer adopts different patterns of interaction with big data, depending on her/his motivational orientations. The term motivation here refers not only to the goals set by individuals in relation to the object of the activity (Locke and Latham, 1990) – which have received the most interest so far – but also to the intrinsic appeals that guide behavior (Ryan and Deci, 2000).

At the other end of the spectrum, the farm – as the object of activity – is in constant evolution. It consists of a sum of entities (including the land, the cultivated crops and livestock, farm machinery, buildings, production infrastructures, etc.) that, as Miettinen (2005) argues, are embedded in different socio-economic activities. Hence, the value proposition within this system can be seen as the “meeting point of several interrelated activities” (Spinuzzi, 2017, p. 268). Under this prism, the farm determines both the contexts of farmers’ actions and the tools used, this way also defining the community of actors that mediates the subject-object relation. The transition from process- to data-driven farming inevitably leads to the involvement of more actors in the system and requires the creation of new value propositions. Nevertheless, since such a shift can redefine the meaning of farming, we need to understand why farmers are doing farming under different situations.

Another interesting avenue for future research is to explore how big data affect not only the performance but also the structure of the farm and the quality of farm life. Both inanimate and – especially – animate elements (productive animals which are often transformed into data collection appliances) of a smart farm are variously affected by the technologies used for data collection purposes, as well as by the application of big data analytics in farm practice. Therefore, the reconceptualization of the interplay between farmers, animals, and big data should be brought to the attention of researchers and policy-makers (Holloway et al., 2014). Other sets of rules, relating to the interrelation between subjects, objects, communities of involved actors, and tools are also necessary.

To sum up, our analysis instead of offering answers to the pressing dilemmas associated with data-driven agriculture provides fifteen questions and sets a foundation for future research. Activity theory affords researchers and policy makers an opportunity to taxonomize many challenging questions relevant to the potential benefits and the important risks associated with the use of big data in farming. This work represents a first attempt towards this direction, by depicting some intriguing interrelations among farmers, farms, big data, communities, and rules. Although other frameworks – like, for example, Latour’s (1996) Actor Network Theory – that have been previously used to deal with the ways agricultural systems process inputs transforming them into outputs (Noe and Alrøe, 2012, 2006) can also depict the ways big data produce value for the agrifood sector, activity theory differs in that it distinguishes between different entities within the system (Miettinen, 2001), thus putting emphasis on the co-evolution of these entities in the framework of the activity (Miettinen, 1999). Interestingly, this co-evolution generates conflicts, imbalances, and breakdowns, or – in activity theory vocabulary – contradictions.

Contradictions refer to misfits within elements of the activity systems (primary contradictions), between them (secondary contradictions), between different activities (tertiary contradictions), and between different developmental phases of a single activity (quaternary contradictions) (Foot, 2014). As Foot (2014) explains, these contradictions reflect the richness of a system and its ability to move forward, so, their careful consideration can open-up new outlooks for understanding the developmental potential of the system. Despite the fact that our aim was not to perform a contradiction analysis, our approach can throw some light on the secondary contradictions, which concern the conflicts between different components of the system, such as subjects and tools (farmers’ views on the ways big data reconstruct the farm), subjects and the community (since the community reconciles the subject-tools relationship but absorbs an important portion of the produced value), tools and the rules that frame the use of big data, etc. Our list of questions (Table 3) might help to identify some of these contradictions.

Nevertheless, farming is by definition characterized by primary contradictions – emerged from the conflict between use value and exchange value (Foot, 2001) – since the “doing” of farming has both an inherent and a commoditized value. On the other hand, smart faming is in itself a tertiary contradiction, since new tools – referring to both new data and their new uses – already used in other, more advanced activities (Engeström, 1987) are continuously introduced in the activity system. Furthermore, given that both the practice of smart farming and the research in the field are in an early stage of development, the identification of quaternary contradictions – inter-activity conflicts (Foot and Groleau, 2011) – is still largely undermined. Although it is important to understand how these contradictions affect the activity system, such an analysis is left for other researchers.

Moreover, it is worth mentioning that our analysis was concentrated to some universally shared features of data-driven farming systems, but it must be acknowledged that activity systems might be quite different depending on the type of farming they are referred to or on the available infrastructures. In supporting this argument, relevant work indicates that farms with higher levels of integration into the value chain can better exploit the opportunities offered by big data (Fleming et al., 2018), whereas the big data software capacity is concentrated in countries with advanced economies (Hilbert, 2016). Moreover, one can expect significant differences between developed and developing countries in the levels of big data availability or data literacy.

Obviously, much future work remains to be done on the modes activity systems operate, the ways in which value is produced, and the forces that govern the production and distribution of value. Hence, the present work must be viewed as what was designed to be: a first step towards a systemic analysis of activity systems in the case of smart, data-driven farming. To achieve a sound knowledge base on the topic, both theoretical and empirical research on specific activity systems are needed. Future researchers can expand our framework by adding and studying new concepts. Two central issues in activity theory – yet, not touched upon in this study – are historicity and expansive transformation. Historicity refers to the need to view the problems and potentials of activity systems against their own history (Engeström, 2001; Foot, 2001). Although it was beyond the scope of this article, the difficult task to put smart farming systems into a historical context – so as to understand how past activities, ideas, tools, and motives shape current trends of activities – can offer a wide range of new insights. Expansive transformation, on the other hand, refers to the ways the whole activity system is transformed as an outcome of the gradual resolution of contradictions (Engeström, 2001; Avis, 2009). We hope the questions posed here can provide a suitable jumping off point for those who will undertake the challenging task to study how data-driven farming is expansively transformed.

Acknowledgment

We would like to thank the two reviewers and the editor for their valuable comments and suggestions, which significantly improved the quality of this article.

Notes

1 Although Engeström also uses the words “mediating artifacts” (2001) or “instruments” (2000) in this article we adopted the term “tools” used in his later writings.