Introduction
Nuru is Swahili for “light”–it is also the name of the mobile app cassava farmers are now using to ensure the health of their crops. A starchy root, cassava is a “versatile staple” across regions of the African continent, used for food, glue, animal feed, and pharmaceuticals.[i] Unfortunately, this key crop is constantly caught between diseases, such as mosaic and brown streak disease, as well as green mites. Furthermore, farmers only accurately identify the aforementioned diseases 18-31% of the time.[ii] With a growing global population (and several stomachs) and several of the afflicted African regions undergoing rapid development, securing a plentiful harvest is becoming more and more critical.
Developed by Penn State researcher David Hughes in 2018, Nuru is a part of a larger project, PlantVillage, organized to assist rural farmers through the novel intersection of AI and agriculture.[iii] Through careful and extensive machine learning, Nuru constantly improves its object detection capabilities in the field. Initially delivering 65% accuracy in identifying plant diseases and infestations, the app has neared 90% when one expands the number of cassava leaves exposed to the app.[iv] Not only does it inform the user of their crops’ health, but it also gives low-technology recommendations for disease management and treatment.
Object detection is one of the simpler applications of AI. While technology has only begun to appear on the farming scene, the average American consumer is quite used to this basic AI by now. Image and video data are fed into programs to accurately identify license plates from CCTV, recurring faces in social media posts, and to follow the soccer ball in your favorite team’s match–although it may just mistake the referee’s bald head for the ball.[v] Nuru is one of several ambitious AI technologies PlantVillage has deployed to aid developing regions’ farming efforts. The project has also delivered drones and mobile spectrophotometry technology equipped with AI to accomplish the same crop security.
Questions arise despite the altruistic nature of PlantVillage’s technologies and similar AI in agriculture. Several are found in other applications of AI and ML, such as employment and discrimination issues. However, the overarching question is: how will this technology be evenly distributed and specifically applied to its fullest potential?
While questions focus on the ethics and governance of the Nuru apps, practical obstacles also present themselves due to the novelty of the technology and the condition of the recipients. How will sufficient, quality data be gathered due to inaccessibility and the prevalence of extraneous features and images? Relatedly, the actual deployment of models must consider the diversity of each peculiar scenario and factor in the likelihood of unskilled workers and hardware/platform limitations.
Object detection plays an important role in our lives, and it is Hughes’ and PlantVillage’s hope that the same technology plays a similarly large role in the lives of developing region farmers. To determine, then, whether the application of AI in agriculture is more beneficial than obstructive, we must begin with the state of object detection in 2022. From there we may approach the earlier ethical questions and practical obstacles to decide on a final recommendation for Nuru and related apps and technologies.
State of object detection
Object detection has been around as long as modern AI. Pattern recognition systems came about in the early 1960s to aid office automation tasks. This supervised learning technology has since evolved to accept an enormous amount of labeled pictures and make rapid, accurate classifications to improve the user’s experience. Such an effect, especially in PlantVillage’s case, is amplified by deep learning: three or more layers of artificial neural networks simulating the human brain allow Nuru to “learn” from copious amounts of data to deliver more accurate results. The deep learning “revolution” initiated in the early 2000s resulted in an explosion of applications that shifted perspectives on fields from healthcare to video games.[vi] Object detection did not only become better, it grew better. The major groups that built a name in other sectors of electronic and Internet technology turned their eyes toward deep learning and the practical applications of object detection. Google was naturally one of these groups. Their public release of the free software library TensorFlow sparked their role as a central part of many AI applications, including PlantVillage’s Nuru app.
Tensorflow was released in 2015 as a second-generation system, succeeding the 2011 DistBelief system.[vii]TensorFlow bundles ML and deep learning models and algorithms, with object detection as the goal of many of these. It is an open-source library that is easily deployed across a variety of platforms. Its object detection application programming interface (API) is a versatile and efficient system that makes annotating, training, and monitoring accessible for several projects. High school students have even used TensorFlow to create their own AI agriculture app, “PlantMD,” inspired by PlantVillage.[viii]
While the sophistication of object detection technology through its history is nothing short of impressive, lingering obstacles remain before any may claim its ubiquitous accessibility and equal usage.
Obstacles to use
Practical issues abound in the target areas of Nuru and its users. There are few individuals able and willing to deploy AI models, and such models require extensive resources and time to develop from scratch. Rural areas, especially in developing regions of Africa, also do not lend themselves to accessible or orderly data. Obstacles to Nuru and related technologies’ application then fall into two related categories: model-related and physical/environmental.
Tanzania’s local farmers just got smartphones, and PlantVillage has helped supply them. Nuru could be the most useful AI application in agriculture, but it needs suitable hardware to deliver its outputs. Thankfully, Hughes, in tandem with the International Institute of Tropical Agriculture (IITA), has solved at least one of the glaring questions on readers’ minds: How will this app work in areas certainly outside the range of Wi-Fi towers? Once downloaded, Nuru does not require cellular data or remote computing power, working instead from a pre-set library of labeled images to correctly identify diseased plant leaves.[ix] Hughes and his team at PlantVillage have also been working on solving the other problem of individual usage. Penn State researchers tied with the project and other team members are onsite to onboard and familiarize the farmers with the technology. These answers, however, do not eliminate all deployment questions: every farm and region is different, and this diversity potentially hinders Nuru’s data collection.
Data are the backbone of any AI model and algorithm, and when access to them is hindered, the technology flounders. The main question related to the environment in which agricultural AI garners its images is whether the app can and will learn to overcome physical obstacles. Poor quality images and extraneous features within the images are only a couple of considerations for which the technology must account. The supervised, annotated images fed into such an app must also be accurate and there must be enough of them for the application to develop and learn.
Ethics
Superior supervised learning will likely enhance Nuru’s performance, but this very improvement may be problematic when one considers a range of ethical questions. There are three main categories: employment, equality, and culture.
Employment is a typical issue one considers when AI has moved into a new industry and application. The first question then asked is how many workers will be replaced by the technology. Nuru and related technology are not seeking a replacement, but rather an enhancement of workers and farmers in target regions in 2022. Farmers are required to both identify diseases and perform the harvesting themselves, detection AI only works to aid the former task. Apps such as Nuru pose little replacement for the latter. However, the consideration of employment with the advent of AI in agriculture must continue to be present in programmers’ and farmers’ minds; with AI already able to identify crop diseases and infestations far better than the average farmer, it may not be long until the harvesting task becomes widely mechanized. Also important are ethical considerations of technology equality in both access and politics.
Which farmers and regions are in most need of Nuru and other AI applications in agriculture? Are these same groups the ones which will work best with the technology in its state as of March 2022? There are several approaches to these questions, each with advantages and drawbacks. One could deliver the AI to the farmers which need it most, lifting the poorest and those most susceptible to crop damage. Those who arguably need it the most, however, may not supply the most amount of people. It would also be up to someone’s discretion as to who “needs” it the most, and questions would there arise on the metrics by which that someone defined the need. On the other hand, one could present it to those for whom the technology would benefit the most people, a more utilitarian argument. This may lead to a potentially increasing gap between the poorest/“neediest” despite aiding the most people. Finally, one could argue that such technology should (first) benefit those who developed it. This last approach, however, is the most susceptible to counterarguments: once one considers the similarities between this case and that of the COVID-19 vaccine diplomacy strategy of several developed countries, it becomes apparent that a home-first policy for globally applicable inventions is not a global solution.[x] Variants in crop disease and pests would not arise as mutations did in the Coronavirus, but a similar disparity in benefits would emerge if this technology were restricted to those who developed it and were sold at an unreasonable price.
Despite PlantVillage’s status as an altruistic NGO project, local and national governments or groups may not look as favorably upon the partnerships and actual application of the technology. Less stable authorities in such regions may not appreciate the gesture Hughes and his team seems to be making. They will take the donations of Nuru as undermining their power or a sign that they cannot provide for their people. Similarly, how will regions in conflict react if one of them receives focus from AI-delivering groups and the other does not? Without fair distribution of such technology, even farms within the same country will experience varying crop yields, and some areas may be stressed for food while others reap ML-driven benefits. What appears to symbolize a great leap forward in developing agricultural regions may unintentionally lead to different and potentially greater human costs.
Finally, some cultures may be hesitant about receiving AI technology for various reasons. Traditional farming practices may be an obstacle to application and individual farmers can be uncomfortable switching older methods of disease identification for high-tech ones. In a similar vein, the low-tech treatment recommendations suggested by Nuru and other AI apps may conflict with familiar, traditional methods. The province of Bali in Indonesia is an example to consider for this ethical question. Bali was a prime target for the Western-initiated “Green Revolution” in the mid to late 20th century. Proponents introduced chemical fertilizers, and pesticides, and ostensibly improved cropping patterns. These “improvements,” however, were made without considering the sway and practical advantages the Balinese religion held over consistent irrigation and planting patterns. As a result of swift and ignorant modifications, the initially galvanized rice crop faltered and sank to the region’s detriment.[xi]
Traditional farming and agriculture have changed shape many times throughout history and to various degrees according to groups of people. Diversity in individuals, physical land, politics, and traditions have all proven to be potential hindrances to the application of AI in agriculture. To better understand AI’s impact on the industry, one must consider the state of agriculture and its influences in March 2022.
Zooming out: agriculture in the context
One can understand how influential the development of organized agriculture was for humanity, but more important are recent movements that have modernized and mechanized agriculture to sustain an exponentially growing global population. In this discussion, the impact of the Green Revolution cannot be understated. From the 1960s to the 1980s crop genetic modifications were made available to the developing world.[xii] This was made possible by large public investments, allowing countries like Turkey, China, and India to apply this technology to staple crops such as maize, wheat, and rice. The “post-GR period” into the early 2000s, however, saw a decline in investments in economically and ecologically sustainable agriculture. As a result and despite renewed interest in the mid-2000s to the 2020s, “low-income countries and lagging regions of emerging economies continue to rely on agricultural productivity as an engine of growth and hunger reduction.”[xiii] The Green Revolution was catalyzed and sustained by Western practices for maximizing crop yields. Although it did not always have the intended effect, per the Bali example in the previous section, it did revolutionize many regions and their approach to crop production. AI’s application in agriculture was not a factor of the Green Revolution, but its splash in the industry was made around the height of the same movement.
One of the first applications of AI in agriculture was the identification of soybean diseases in Illinois in 1983.[xiv] 40 years later and PlantVillage took full advantage of the development of deep learning and neural networks to improve the same object detection tasks that fed Midwesterners in the previous century. Although this study is mainly concerned with these and related tasks, scientific authors have observed other applications significantly impactful for the industry. AI sensors, drones, and robotic agents save excess water, and fertilizer, and maintain soil fertility by optimizing irrigation and soil management techniques.[xv] These technologies, albeit helpful, raise similar and more concerns outlined above regarding application. For one, they are also not evenly distributed among farmers of the world or specific regions. To assist in determining the ethics of PlantVillage’s Nuru app and stay within the scope of this study, however, it is helpful to consider the actual figures of crop and food production in concerned areas.
Agriculture is the dominant occupation in most countries of Africa, yet crops face constant natural threats of drought, disease, and fires. An estimated 40% of crop losses in Kenya alone come from diseases, pests, and weeds.[xvi] The country is also not alone in its inability to adequately feed its population–a third of Tanzania’s youth under age five are stunted and malnutrition runs rampant despite the country’s agriculture sector employing 75% of the population.[xvii] Private sector investments and policy initiatives are helpful, but they work against challenges such as low capital, skills, and resources, climate change, and a burgeoning population. With gloomy outlooks on the further effects of climate change and global population growth, technologies applied to mitigate the damage done to agriculture will be of great importance. It is then appropriate to view PlantVillage and its efforts under such constraints and conditions, as well as the ethical and practical questions above.
Identifying PlantVillage’s impact
How effective are PlantVillage and its AI applications? The organization advertises the three aforementioned projects: Nuru, drones, and mobile spectrophotometry. Established in 2013, PlantVillage reached almost five thousand individual farmers with the Nuru project in 2019 alone.[xviii] Furthermore, the same report by Self Help Africa, a partner group of PlantVillage, found that over 95% of households interviewed have “diversified their farming approaches by adopting one or more farming practices either as a result of the program intervention or through their initiative.”[xix] What may be more impactful than PlantVillage’s direct involvement in their database that contains relevant information on over 150 crops.[xx] A story on PlantVillage’s website describes the first aggregation center for collected cassava in Amukura, Kenya. The introduction of PlantVillage’s apps delivered an unexpected surplus which the community treats and sells as dried cassava chips to companies. It has also created jobs for women and children to peel and wash tubers.[xxi]
Success stories such as these display the ability of farmers to overcome some of the obstacles mentioned in earlier sections. With more farmers using Nuru and downloading images, the technology can only be improving its accuracy in identifying diseases and pest infestations. Perennial ethical questions remain, however: Can AI technology such as Nuru be distributed evenly and effectively in critical areas? As the applications improve, how disruptive will they be to farmers and related laborers? It also remains an open consideration for PlantVillage and other groups to navigate relevant politics and culture that may conflict with their goals.
Commercialization and partnerships also exist as options to expand PlantVillage’s reach and reduce inequality in targeting specific areas over others. As the organization grows, Hughes and his team(s) will have to decide how best to partner with public and private groups and donors to stay true to their first and current mission of benefiting rural farmers.
Conclusion
While revolutionize is a term that should be used sparingly, one could argue for applying it to Nuru’s impact on the agriculture industry and the cassava farmers in Kenya, Botswana, Tanzania, and more. This impact must be deemed ethical enough for such widespread application, however. AI in Agriculture is an easier topic to discuss on moral grounds than other AI applications with murkier implications, such as facial recognition. Agriculture is always “good” and the AI in the case above has relatively fewer opportunities to be “bad.” With little room for direct discrimination or disparate treatment, related obstacles are limited to those concerning available AI for users and how to integrate them into foreign situations. The AI in these instances is going towards a common human goal: (global) food security.
My own opinion is in favor of AI object detection in agriculture. I would, however, be cautious about deployment. As shown above, one must be careful in the decision to distribute Nuru and related apps–to ensure effective use, extensive training, annotating, and sheer time are needed at all points of application. PlantVillage’s a steady increase in team members is likely the most productive move to make at this juncture with the technology and target regions in mind. With more people in the area equipped to distribute hardware and train others in using the services, growth should be rapid. With more users, Nuru’s ML base will continually improve and reduce crop loss. Outside of securing greater crop yield on the same amount of land, such AI will also secure farmers’ livelihoods, an important point for African countries in which rural farming contributes to a plurality of the GDP.
The future of agriculture is also, in part, the future of AI. This intersection began with disease and pest detection applications that have greatly improved since their origins. With parallel advancements made in robotics and drones, the questions that did not explicitly arise when discussing object detection become relevant. Employment issues and discrimination become more prominent. It is then paramount to maintain the altruistic course David Hughes and the team at PlantVillage have established to benefit the parties that require the most aid.
Endnotes
[i]. “Cassava: Africa’s Most Versatile Staple,” Demand Africa, 2022.
[ii]. Mrisho LM, Mbilinyi NA, Ndalahwa M, Ramcharan AM, Kehs AK, McCloskey PC, Murithi H, Hughes DP, and Legg JP, “Accuracy of a Smartphone-Based Object Detection Model, PlantVillage Nuru, in Identifying the Foliar Symptoms of the Viral Diseases of Cassava–CMD, and CBSD,” Front. Plant Sci, 2020.
[iii]. PlantVillage home site.
[iv]. Mrisho et al.
[v]. Sanyja Jain, “AI Camera Ruins Football Game By Mistaking Referee's Bald Head For Ball,” NDTV, 2 November 2020.
[vi]. Jeffrey Dean, “The Deep Learning Revolution and Its Implications for Computer Architecture
and Chip Design,” Google Research, 2019.
[vii]. Konstantinos Katsiapis, “Towards ML Engineering: A Brief History Of TensorFlow Extended (TFX),” TensorFlow Blog, 25 September 2020.
[viii]. Lucas Schmidt, “AI Improves Farming with Google’s TensorFlow,” The Borgen Project, 29 July 2019.
[ix]. Steve Lohr, “From Agriculture to Art — the A.I. Wave Sweeps In,” New York Times, 21 October 2018.
[x]. Tom Wright and Colin Kahl, Aftershocks, St. Martin's Press, 24 August 2021.
[xi]. Lansing, J. Stephen, Priests, and Programmers: Technologies of Power in the Engineered Landscape of Bali, Princeton, N.J.: Princeton University Press 2009, p. 52.
[xii]. Prabhu Pingali, “Green Revolution: Impacts, limits, and the path ahead,” Proceedings of the National Academy of Sciences of the United States of America (PNAS), 31 July 2012.
[xiii]. Ibid.
[xiv]. “The Methods and Applications of Artificial Intelligence Used in the Technologies of Precision Agriculture,” The International Society of Precision Agriculture, 13th International Conference on Precision Agriculture, 31 July-4 August 2016.
[xv]. Tanha Talaviya, Dhara Shah, Nivedita Patel, Hiteshri Yagnik, Manan Shah, “Implementation of artificial intelligence in agriculture for optimisation of irrigation and application of pesticides and herbicides,” Science Direct, 22 May 2020.
[xvi]. Crops, Kenya Agricultural and Livestock Research Organization.
[xvii]. Tanzania Agriculture and Food Security, USAID, 1 December 2021.
[xviii]. Annual Report 2019, Self Help Africa, p. 65.
[xix]. Ibid.
[xx] Allyn Rosenberger, “Penn State’s PlantVillage Forum is Making Crop Information Free and Accessible,” October 2016.
[xxi]. Amukura Cassava Aggregation Center, PlantVillage blog post, 31 January 2022.