The ketchup you get by growing tomatoes for others

What happens when AI is harvested in the right way and the right soil by farmers? What happens when your software talent resources are shared for community work? Wells Fargo has been trying to improve the lives of farmers in Telangana and Karnataka, as well as for some upskilling work with other NGOs. Its technology team dedicated 2,200 volunteering hours to redesign the FarmPrecise app, developed by Watershed Organisation Trust (WOTR). The idea was to equip farmers with actionable insights for better decision-making, covering weather forecasts, crop management advice, fertiliser application and spraying techniques and real-time market prices and government scheme updates. This redesign was noteworthy for the integration of AI-powered computer vision used for precise crop disease detection and recommendations for organic solutions. Turns out that AI seeds blossom in these pastures. Recent metrics show upticks in feature usage by 25%, and savings of 20% of input costs for over 70,000 farmers (with more than 100,000 downloads). Here’s an interview with Harish Mohan, Head – Digital Technologies and Innovation - Wells Fargo India and the Philippines, to understand this interesting wealth creation strategy—growing knowledge by generously sowing some seeds in volunteer work. He also explains why precision farming and wealth-creation, and fraud detection patterns are branches of the same tree. And what rocks still come in the way of new digital furrows?

How many developers and resource hours are committed to such community work? Any other examples like WOTR that you can share?

More than 36 developers contributed their time to refine the FarmPrecise app. Apart from this, Wells Fargo has partnered with the NGO Anudip Foundation to streamline their placement process for underprivileged students. Anudip Foundation empowers under-privileged students with industry-ready skills and provides job placement in corporates. The NGO faced a challenge of manually reviewing and matching 12,000 job opportunities for 40,000 students. This was not just time-consuming, but also error-prone, as incorrect tagging of skill requirements was as high as 30 per cent. Wells Fargo volunteers helped streamline this process by using Python-based Fast APIs, and a pre-trained model, SkillNer, and spaCy was chosen as the open-source library for NLP. This resulted in accurately delivering job skills categorisation in seconds, saving a 72+ person effort. Additionally, our cybersecurity assessment fortified Anudip’s critical applications by plugging in gaps such as encryption of the data, reducing excessive permissions, etc.

We also learned how to scale, generalise, incorporate feedback, and optimise models on a large scale in a short span of time.

Does such work also help in testing or improving your own AI edge?

Such work has significantly enriched our knowledge and skillset on AI, UI/UX, product management and other advancing technologies. Also, our team members got first-hand exposure to the real-world issues that farmers and the larger farm-based community face, sensitising them to the needs of society. We have gone the extra mile to implement certain models to which we had no prior exposure, such as the app’s computer vision feature. We also learned how to scale, generalise, incorporate feedback, and optimise models on a large scale in a short period.

How evolved are computer vision and real-time data in such apps in Indian regions, given the lack of infrastructure, poor data adequacy, adoption complexity, etc.?

The application of computer vision and real-time data in Indian agriculture is not growing as fast as one would expect, owing to several factors. Reliable internet connectivity, especially in rural areas, is crucial for real-time data transmission and processing. Power outages are also a significant hurdle. Then there is the issue of poor data adequacy. High-quality, labelled datasets are essential for training robust AI models. The scarcity of such data in India, particularly for diverse cropping patterns and regional variations, limits the accuracy and generalisability of AI solutions.

Is digital literacy a hurdle as well?

The complexity of adoption can be a challenge. Farmers, particularly smallholder farmers, often lack the digital literacy and technical skills needed to effectively use AI-powered tools. The complexity of the technology and the perceived risk of failure can also deter adoption. Plus, the upfront cost of hardware (sensors, drones, etc.) and software can be prohibitive for many farmers.

Is there a need for synthetic data and RAG in creating better and faster agri-AI models?

Yes, there’s a strong need for Synthetic data, and Retrieval Augmented Generation (RAG) can accelerate the development of agri-AI models in India. Synthetic data can augment scarce real-world data, allowing for the training of more robust and generalisable models. It can simulate various scenarios (e.g., different weather conditions, pest infestations) that might be difficult or expensive to capture in real-world datasets, also owing to reduced volumes of real-time data availability. RAG, on the other hand, can leverage existing knowledge bases and research papers to improve the accuracy and explainability of AI models. This is particularly important in agriculture, where context-specific knowledge is crucial for effective decision-making.

What are your thoughts on Moravec’s paradox in the context of AI and farmers?

Moravec’s Paradox highlights that tasks easy for humans are often difficult for AI, while tasks difficult for humans are relatively easy for AI. In the context of Indian farmers, this means that AI might excel at tasks like image recognition for disease detection (difficult for humans to do consistently across vast fields), but struggle with tasks requiring nuanced judgment and contextual understanding (easy for experienced farmers). Bridging this gap requires AI systems that can integrate human expertise along with farmer knowledge.

Are IoT or ground sensors, or drones, helping here in any way?

Data-driven technologies like remote sensing, smart sensors, and IoT-based devices constructed over AI/ML algorithms have become a fundamental aspect of agriculture that assists farmers with critical decision-making. This revolution is supporting farmers in terms of farm management by improving crop yield, pest control, soil health, etc.

Scarcity of high-quality, labelled datasets limits the accuracy and generalisability of AI solutions.

What challenges and strides have been made in the area of soil testing?

Challenges in soil testing include accessibility, cost and turnaround time. Laboratories may be far from farms, making testing inconvenient and expensive. Soil testing can be costly for smallholder farmers. And results may take days or weeks, delaying crucial decision-making.

How does the app ensure the insights translate into a usable/swift way for farmers?

Apps such as FarmPrecise can address these challenges by integrating with mobile devices, allowing farmers to easily upload images or data, providing quick, localised insights, using AI to analyse data and provide immediate recommendations and connecting farmers with local labs while facilitating convenient and affordable testing.

Have you ever faced issues like weak digital literacy or high digital divide in India’s hinterland? Also, challenges with upfront costs, etc. Encountered by small farmers.

Digital literacy and the rural digital divide remain challenging. Many smallholder farmers lack access to smartphones, the internet, and the necessary skills to use AI-powered tools. The upfront costs of technology can also be prohibitive. Addressing these challenges requires targeted training programs, affordable technology solutions, and government subsidies.

What next can such initiatives and apps shape into? How is this app different from other efforts/solutions in this space? Like E.L.Y AI model by Bayer&MS, John Deere’s spray system, Kisann e-mitra chatbot, Baramati AI project, and farming start-ups in India?

Future agri-AI apps can evolve into comprehensive farm management platforms, integrating various data sources (weather forecasts, market prices, etc.) and providing personalised recommendations for each farmer. They can also incorporate features like Predictive analytics (Forecasting yields, pest outbreaks, and other risks), Financial management tools (Helping farmers manage their finances and access credit) and Market linkages (Connecting farmers with buyers and ensuring fair prices). The key differentiator for a successful app is its ability to address the specific needs and challenges of Indian farmers, focusing on simplicity, affordability, and accessibility. A successful app needs to go beyond replicating existing features and offer a unique value proposition.

Is a precision farming model not too different from fraud detection/market aanalyticstrade ops in finance? What other fields of research are you exploring? Can AI used in the financial industry be reproduced/replicated in community areas and vice versa?

Precision farming shares similarities with fraud detection, market analytics, and trade operations in finance. All involve analysing large datasets to identify patterns, predict outcomes, and optimise processes. AI techniques developed in one domain can often be adapted to others. The company’s AI research might explore applications in areas like risk management—predicting loan defaults and other financial risks, customer service—improving customer interactions through AI-powered chatbots and operations optimisation—streamlining internal processes and improving efficiency.

What’s the company’s stance and pace on AI- slow/cautious/leading-the-pack? Especially as a bank that has roots dating back to 1852? And why?

As a bank that operates within a strict regulatory environment and is subject to high scrutiny, the adoption of AI, like any new technology, prioritises risk management and addresses ethical considerations. The pace of AI adoption is likely to accelerate, given the potential benefits in various areas, including compliance, fraud detection, and customer service.

pratimah@cybermedia.co.in