1. Introduction

Technological innovations have driven agricultural change for centuries. With intersecting labour and ecological crises hitting the global food system, there is an urgent need to develop methods to enhance sustainable agricultural production. According to key policy actors, the so-called ‘fourth agricultural revolution’ (‘Agriculture 4.0’), driven by new digital technologies such as AI and robotics, is key to the transformation of the food system to meet the growing global demand for food within planetary limits [1,2].

Public perceptions of new technologies play a critical role in their success or failure [3–6]. Despite an emerging literature exploring the ethical impacts of AI and robotics in agriculture [7–12], there has been little attention paid to public perceptions of these technologies and their implications for the ethical acceptability of AI and robots in agriculture.

This paper draws on a set of twelve dialogue groups conducted across metropolitan and rural Australia to understand how members of the public view the social and ethical impacts of AI and robotics in agriculture. We presented participants with several hypothetical scenarios of AI and robotic technologies – including the possibility of ‘farmerless farms’ – in the horticulture and poultry industries to gauge public attitudes toward these technologies. By mapping these sentiments, we reveal the ethical approaches publics take in assessing the value of agricultural AI and robotics and highlight their importance for research and policy regarding the future of farming.

The paper is organised as follows. First, we briefly introduce the existing literature on the ethics of AI and robotics in agriculture. In the Methods section, we outline our research method and introduce the idea of dialogue groups. We then list the key findings of the study. Specifically, we explore how participant values are expressed in their deliberations about the consequences of the adoption of AI and robotics across the areas of agricultural labour, corporate power, rural identity, and animal welfare. In our discussion, we highlight the significance of our research and point to several areas for further research.

2. The ethics of AI in agriculture: Debates and public views

AI and robotic technologies for agriculture range from ‘intelligent’ farm management software, yield prediction algorithms, and smart cameras to machinery such as drones for crop-dusting and monitoring, automated packing and handling systems, and self-driving tractors [11,12]. John Deere now sells a fully autonomous tractor, which can plough a field on a farmer’s behalf and uses AI to collect, stream, and sort images of the areas of its operations. Robotic fruit picking technologies are also in development, with some in trial stages, but are yet to be commercially viable on a large-scale [13,14].

There is a burgeoning scholarship examining the ethical issues raised by AI and robotics in agriculture. Scholars have highlighted issues of data and trust [9,15,16], environmental sustainability [11,17], farmer livelihoods and autonomy [18,19], animal welfare [12,20,21], rurality [22], and labour [23–25].

Our paper builds upon a small body of work that explores public views and perceptions of AI and robotics in agriculture to understand the factors that lead to acceptance or rejection of these technologies [26–29]. It is novel in using the methodology of dialogue groups to generate rich debate among members of the public and in drawing on hypothetical scenarios to induce members of the public to consider the type of agricultural future they value. Our research moves beyond the framing of AI in agriculture as having risks and benefits that create a “trolley problem” [8], and instead draws attention to the meanings and values members of the public ascribe to these technologies when they think about a future in which AI and robotics might be widely adopted in agriculture. Our results add depth and nuance to the existing, mostly quantitative, literature on public attitudes towards agricultural robotics and AI, and constitute a valuable resource for policymakers, or other stakeholders, who want to engage with public opinion regarding these technologies.

3. Research methods

The study presented in this paper is drawn from dialogue groups conducted as part of a project examining the social and ethical impacts of AI and robotics in agriculture. Dialogue groups have been used extensively to explore the moral intuitions and ethical evaluations of lay members of the public regarding new technologies [30,31]. Dialogue groups promote facilitated discussions that are centred around hypothetical scenarios in order to elicit participants’ moral positions, and importantly, their reasons for holding such views [31].

To explore public attitudes towards the use of AI and robotics in agriculture, we conducted twelve dialogue groups between August and December 2023. The dialogue group study was approved by the Deakin University Human Research Ethics Committee, approval number HAE-23–068. Participants from metropolitan and regional Australia were recruited by a professional recruitment service (Taverner) using selection criteria based on socio-economic indicators to ensure a diversity of people that represented the communities in the study. Potential participants were then sent an email providing a Participant Information Statement and link to an online Qualtrics consent form to which they provided written consent to participate in the study.

Dialogue groups were held via Zoom and facilitated for 90 minutes. Each dialogue group consisted of four to seven participants. Six dialogue groups involved participants from Metropolitan areas of Melbourne (Victoria), Sydney (New South Wales), and Brisbane (Queensland), with the remaining six groups involving participants from regional farming areas of the Goulbourn Valley (Victoria), New England (New South Wales), and Gladstone (Queensland). All dialogue groups were audio-recorded and professionally transcribed, with participant identities remaining anonymous in the transcription process.

To facilitate discussion, participants were presented with a hypothetical scenario focused on fully automated farms in horticulture and poultry (see S1 File). Scenarios were developed with the aim of provoking discussion of what we took to be key questions about the ethics of AI and robotics based on previous research [11,32], and an extensive literature review. As discussion in each group developed, we introduced complicating factors – such as the shift from semi to fully-automated farms, the removal of farmers from decision-making processes, and biosecurity threats – to tease out why participants held the beliefs they were expressing.

Participants were encouraged to offer explanations for their responses throughout to provide insight into the assumptions, imaginaries, and judgements guiding their positions. We also presented participants with three poll questions at the end of the horticulture and poultry scenarios to produce a quantitative record of their opinions about AI and robotics in agriculture. Dialogue group discussions were audio-recorded and transcribed. Participants were compensated for their time with a $100 e-gift voucher.

We used the Framework Method to analyse the content of the discussions that took place in the dialogue groups [33]. Three researchers from the team read the first three transcripts to develop an initial coding framework. The texts were hand-coded and categorised by deductive (pre-defined codes including the pros and cons of AI, trade-offs, and tensions) and inductive codes. The remainder of the transcripts were then split and coded independently, with an audit being conducted. The process was iterative with discussions held weekly between the three coders and fortnightly with the larger team, with additional themes constructed and added. Codes were charted on an Excel spreadsheet, summarizing the data from each transcript and including references to illustrative quotations [33,34].

4. Findings

Dialogue groups were conducted with a total of 67 participants (Table 1). While there was a reasonably good gender split, there were slightly more women than men. Most participants had no direct connection to agriculture although some rural participants had familial connections to agriculture or worked in agricultural and adjacent industries; metropolitan groups were composed of individual consumers. Participants had a range of levels of educational attainment, although, considered as a whole, the average level of educational attainment was higher than that of the average Australian.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Participant characteristics.

https://doi.org/10.1371/journal.pone.0332461.t001

This section reports and unpacks participants’ intuitions about the acceptability of AI and robotics in agricultural applications. First, we outline how participants understood the risks and benefits of these technologies. We then explore participants’ ethical positions in relation to four key areas of concern: 1) the future of agricultural labour; 2) rurality and identity; 3) corporate power and; 4) animal welfare and biosecurity. We go on to note the existence of a widespread fatalism amongst participants about the introduction of robotics and AI into agriculture. The section concludes with a presentation of survey results.

4.7. Survey findings

Participants were presented with a series of survey questions at the conclusion of discussions (Table 2). The only items that secured more than 50% support were the use of AI in horticulture justified as a business measure and as a sustainability measure and in poultry production for the sake of biosecurity. Interestingly, despite just having spent time discussing each example, a sizeable number of participants remained neutral about the appropriateness of using AI and robotics for profitability, biosecurity, and animal welfare when they were required to state a final position in their response to the questions in the poll.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Poll results.

https://doi.org/10.1371/journal.pone.0332461.t002

5. Discussion

Society has always contained those who have “praised” and “glorified” new technologies and those who are critical of new technologies and their associated social, ecological, and political impacts [7]. The participants in the dialogue groups convened for this study did both of these things.

One thing our study shows, then, is how deep and pervasive disagreement about the ethics of the application of robotics and AI and agriculture is. Agricultural robots and AI are almost always portrayed in a positive light in both popular and academic contexts [42,43]. A recent study [29], conducted in Germany, suggested that the public is generally positively inclined towards agricultural robots, despite the long history of scepticism towards new technologies in Germany on environmental grounds. In contrast, our study suggests that a significant portion of participants were deeply sceptical of the use of robots and AI in agriculture. Moreover, the extensive, and often enthusiastic, discussion between participants that took place in the groups did not alter this or produce consensus. As the poll results showed, somewhere between (roughly) a quarter and a third of participants disagreed with the full automation of agriculture in horticulture and/or poultry production for any reason other than to improve biosecurity. A similar, if slightly smaller, proportion remained neutral on each topic. Our results therefore suggest that, at least in Australia, public opinion is more divided than other studies have suggested.

Another significant finding was that different participants often mobilized the same values in support of arguments both for and against the use of robotics and AI in agriculture. That is, even when they agreed on the importance of a particular value – for instance, environmental sustainability or animal welfare – participants often disagreed about whether it constituted a reason to be enthusiastic about the adoption of robotics and AI in agriculture, or a reason to try to be opposed to their adoption. Appeals to the same value in service of arguments to different conclusions occurred in the context of discussions of the environmental impacts of robots and AI, of their implications for labour, and the contribution robots and AI might make to biosecurity. Some participants held that robots and AI would reduce the environmental impacts of farming by facilitating precision watering and application of fertilisers, while others thought that they would threaten biodiversity by promoting monoculture. Some held that robots and AI will free farmers from having to rely on an increasingly scarce pool of seasonal workers, while others thought they would generate technological unemployment in rural areas. According to some participants, these technologies will be good for workers by relieving them of the need to do “dirty” jobs and creating a need for employees with technical skills; according to others, they will bad for workers by rendering them more vulnerable to surveillance and exploitation [44,45]. Further, participants believed both that robots might improve the welfare of poultry by reducing their vulnerability to cruelty from human workers and eliminating interactions with people that are stressful for birds and that they might threaten the welfare of poultry by confining them en masse in highly artificial environments: each of these claims finds some support in the growing body of literature exploring the implications of AI for the interests of nonhuman animals [46–48]. Even the value of biosecurity, which appeared as the motivation for a hypothetical full automation of poultry production in one of the scenarios that the participants in the dialogue groups were asked to consider [12], was mobilised by some participants as a reason not to proceed down this path.

Thus, rather than serving to motivate consensus, these shared values served as sites for disagreement about the merits of different ethical and technological futures. These observations suggest that it will be difficult for either advocates or critics of automation in agriculture to win the public over by framing the issue in terms of the importance of any particular ‘trump’ value. They also suggest that much sociological and philosophical work remains to be done in identifying, untangling, and evaluating different arguments that appeal to each of these different values.

Interestingly, one attitude – if not quite a value – which did unite (most) participants was scepticism about the real corporate agenda behind the push for the use of robots and AI in agriculture. A surprising number of our participants might be characterised as what Langer and Kühl identified as “skeptical citizens” [26]. That is, they were sensitive to the risks posed by the introduction of AI and robotics – particularly when it came to its implications for farmers on small farms – and inclined to question the contribution technology more generally was making to human progress. Participants expressed this scepticism by means of cynical remarks about corporate agendas, and by questioning whether the goods that were being promoted as being served by the introduction of robots and AI were really what was motivating those developing and advocating for these technologies. It has been suggested that many of the digital technologies being touted as ‘solutions’ to the problems facing agriculture are being developed by Silicon Valley entrepreneurs with little understanding of the sector, and are often rebranding versions of products developed for other purposes [49]. Even when agri-tech startups set out to develop products with the goal of promoting sustainability or social goods, commercial imperatives often come to dominate the final version, especially if the startup is bought out by a more established market player (for an example and useful account of this dynamic, see [50]. The confluence of ‘big Ag’ and ‘big data’ produces new corporate conditions that risk exacerbating the uneven flow of benefits throughout the food system, further disenfranchising farmers and strengthening multinational corporations [9,38,51]. Thus, the serious concerns expressed in the dialogue groups about the possibility that these technologies might lead to further consolidation of ownership of agricultural enterprises and disempower farmers in relation to decisions about how they farm, receive some support from the literature. The discussions that occurred in our dialogue groups suggest that, in Australia at least, manufacturers of agricultural robots and AI suffer from a significant trust deficit in the eyes of the public.

Our results also foregrounded questions regarding how, and when, the interests of certain groups become centred in the ethical evaluation of AI and robotics, and how and when others are rendered marginal. It was notable, for instance, how participants referred to the interests of Pasifika workers in discussions about the implications of the adoption of robotics and AI for the exploitation of labour, but to the interests of Australian citizens when it came to talking about the potential of these technologies to create opportunities for skilled work in rural areas. There are genuine concerns regarding the conditions of employment of Pasifika workers in the Australian horticultural industry which are usually precarious and sometimes verge on a modern form of slavery [52,53]. However, the distinction in types of labour – that is, the skilled ‘farmer’, and unskilled and exploited ‘migrant worker – that was mobilised by some participants plays into the ongoing racialisation and ‘production of precariousness’ that shapes the experiences of, and the opportunities available to, Pasifika and other migrant agricultural workers [54,55]. Our findings therefore highlight the importance of interrogating not only whose interests are considered in the contexts of debates about any transition to AI and robotic agriculture, but also how these interests are constructed and to what ends. They also speak in favour of including immigrant farm workers, whose experience and working lives will be reshaped by technological change and who are often “invisibilised” in this context [56], in deliberations about the future of farming.

Participants also drew on longstanding nationalist narratives about the role played by agriculture in securing Australia from the “threat” posed by foreign powers and values. This was, perhaps, unsurprising given that concerns about foreign ownership of Australian land and financial investment in agriculture in Australia have received considerable media attention in Australia over the last decade [57–60]. Nevertheless, appeals to the importance of maintaining national sovereignty had considerable rhetorical force in the context of participants’ discussions of the implications of adopting AI and robotics for control over agricultural land, technology, and data. These appeals sat uncomfortably next to calls for Australian farmers to make the transition to a more automated agriculture in order to maintain their global position in providing reliable, safe, and nutritious foods to a growing global population.

Relatedly, it was clear that participants were often influenced by a powerful and nostalgic imaginary regarding the nature of rural life and the role of the farmer in discussions. In Australia, agriculture and farming – and thus, the ‘bush’ – plays a significant role in the shaping of a national identity that is grounded on the valorisation of the (settler) farmer as a hardworking, steadfast, and resilient embodiment of stewardship and good citizenship [61]. This social imaginary played a significant role in shaping discussions, with a number of participants suggesting that the introduction of robotics and AI into agriculture would impact on the nature and quality of rural life and increase the vulnerability of rural towns to external economic, political, and social pressures. Implicitly – and sometimes explicitly – life in rural areas was figured as different from, and superior to, life in big cities: this ‘idyllic’ rural way of life was what needed to be defended from the impacts of new technologies [62]. Despite many of the participants having no direct experience of farming, they mobilized a figure of the farmer as someone with a distinct social role, a historical tie to a particular area or piece of land, and a strong moral character in order to think about the future of farming [37,63]. Again, this idea – this imaginary – of the farmer often functioned as something that needed to be preserved and protected against the encroachment of a new, more technical, more technological vision of food production. Whether this idea of the farmer and farming is realistic and/or worthy of preservation strikes the authors as being and important ethical and political question when it comes to the larger debate about the future of farming.

Insofar as food systems affect and implicate all of us, the case for public, democratic, input into the future of the food system, including decisions about the technologies that are likely to shape it in the future, is especially strong [64]. Calls for public deliberation, citizen engagement in science, and democratic input into technological trajectories are rife in the Responsible Research Innovation and Science and Technology Studies literature, including the literature on agricultural robotics and AI [32,65,66]. Several of our findings – that the public is divided in its attitudes towards the use of AI and robotics in agriculture; that this difference persists even when those with opposing opinions appeal to the same values; the diversity of different ethical and political concerns expressed by participants; and that members of the public are sceptical about the motives of those developing these technologies – lend force to the idea that there is a strong moral, as well as a pragmatic and political, case for including the public in decision-making about the future of agricultural robots and AI. The detail of our findings, especially regarding the arguments that participants put forward in the discussions, constitute a valuable resource for policymakers or other stakeholders who want to engage with public opinion on these technologies.

In this context, it is, perhaps, unfortunate that our participants seemed to have little faith that consumers and citizens would have any say regarding the future shape of food systems. Despite their seeming enthusiasm for discussing the ethical issues of AI and robotics in agriculture, participants seemed resigned that the sector was already heading in the direction of fully automated farming. Given the strength of the case for public input into the future of food systems, this struck the authors as both depressing and disappointing.

Inevitably, this study has some limitations. Insofar as the dialogue groups method relies on the introduction of various scenarios by the research team, concerns about ‘framing’ have some valence here [67]. Given that participants did not necessarily know much about the current state of AI and robotics in agriculture when commencing discussions, there was some risk that their attitudes were significantly shaped by the scenarios we presented, which postulated capacities of robots and AI that have not yet been seen at commercial scales. Regardless of the role – if any – played by the scenarios in shaping participants’ responses, the fact that most people have little experience with robotics and AI means that their ideas about these technologies tend to be shaped by their representation in science fiction [68]. This is an unavoidable limit of any contemporary research regarding public attitudes towards AI and robotics. The authors believe this risk was justified by virtue of the method revealing participants underlying moral and political values, which they brought to a discussion of scenarios that were explicitly flagged as hypothetical.

Similarly, it is possible that the poll results exaggerate the extent of public hostility to the use of robots and AI in agriculture by virtue of asking participants about their attitudes towards “full automation” in the context of horticulture and poultry production. We did this in hope of eliciting strong responses to the prospect of increased use of robots and AI in agriculture, which might in turn move participants to reveal the values they brought to the question. Wording the survey questions differently, for instance to ask participants about their attitudes towards an “increased” use of AI and robotics, might have generated a different set of responses.

As we observed above, a significant number (between 13% and 35%, depending on the question) of participants chose to record their attitude towards the use of robots and AI in various forms of agriculture as neutral. A limitation of the survey instrument is that, by itself, it does not allow us to distinguish between participants who might have been undecided on, or felt insufficiently informed about, the use of these technologies, and those who might have reached a confident conclusion that the considerations speaking for and against the use of these technologies in agriculture were precisely balanced. That said, the dialogue group methodology used in our study facilitated wide-ranging discussions about the merits of the use of AI and robotics in agriculture, which provided ample opportunities for participants to ask questions of each other and/or the facilitators. We believe it would, therefore, be a mistake to assume that those who chose to record their attitudes as neutral were not reporting a considered conclusion.

Obviously, there is also a question as to how much our findings, based on research conducted in Australia, generalise to public attitudes towards the use of robotics and AI in agriculture in other nations. Given the similarities between the economic and environmental challenges facing farmers in Australia and in other nations that practice forms of agriculture that are amenable to automation, we suspect that our findings may well be relevant to discussions of the ethics of the use of robotics and AI in agriculture in other nations as well: this awaits confirmation by the results of the work of others.

Finally, while we have attempted to attend to some of the ethical and political issues arising out of the history of race in the context of agriculture in Australia, we did not explicitly refer to Indigenous perspectives in our scenarios. Further research, on the intersections between agriculture, race and coloniality, and emerging technologies, will be necessary to explore the implications of the history of invasion in Australia, and the attitudes of contemporary First Nations people, for the future of agriculture.

6. Conclusion

Despite media and industry actors celebrating the promise of AI and robotics for agriculture, our study finds that many members of the public remain sceptical of these technologies. Participants were concerned about how the food system ought to be structured and drew on longstanding agrarian imaginaries in defence of the moral value of farming: racial and nationalistic determinants of this imaginary also influenced how and when participants considered the interest of various groups who might benefit from, or be harmed by, the use of robotics and AI in agriculture. Although participants could agree on the values that should be brought to the discussion about the future of agriculture, they disagreed on whether these values spoke in favour of, or against, automation. When required to express opinions about the prospect of farmerless farms, our participants were divided, even after lengthy discussions.

By revealing the extent of differences in opinions regarding the wisdom of adopting AI and robotics in agriculture, our research provides support for the idea that there is a significant question concerning the democratic legitimacy of policies regarding the development and use of these technologies. If – as we believe is appropriate – policymakers and technology developers want to ensure that future applications of robotics and AI in agriculture have social license, then they must engage with, and take seriously, the range of public opinions on this topic. In reporting the diverse concerns expressed by participants in their own words, our study provides a valuable resource to policymakers, or other stakeholders, who wish to do so.

Supporting information