Who wouldn’t want technology to solve some of mankind’s gravest problems to make the planet a better place? That’s why it’s important to understand what these “problems” are and slot them under distinguishable cohorts. Towards meeting this end, the UN’s Sustainable Development Goals have helped in direction-setting. There are 17 of them: No poverty; Zero hunger, Good health, and well-being; Quality Education; Gender Equality; Clean water & sanitation; Affordable & clean energy; Decent work & economic growth; Industry innovation & infrastructure; Reduced inequalities; Sustainable cities & communities; Responsible consumption & production; Climate action; Life below water; Life on Land; Peace, justice and strong institutions; and, Partnerships for goals.

Phew! That was an exhaustive list and pretty much covers everything. In 2018, which seems so far away now, Google announced the AI Impact Challenge and as expected, the applications came pouring in (more than 2600) from all the six continents, and cumulatively they essayed a rich compendium of insights. 

Is AI the silver bullet to all of mankind’s problems? Apparently not, as the compendium reveals. There are instances where rules-based systems can provide cheaper solutions to complex problems, and the study goes on to add, why the demand for technical talent has expanded from specialized AI expertise to data & engineering expertise as well.  

Cotton Plantation - Supporting 6 million Indian Families 

That’s right, 30 million Indian lives (approximately) are supported – as per an India Today story, last year – through cotton plantation. We are the largest cotton producer in the world – 29 million bales in ’19 – 20 (United States Department of Agriculture). Despite these impressive figures, productivity per hectare is very low – 2.47 bales per hectare or a little over 420 kilos (avg. global - 775). One of the reasons is because 50% of the yield is destroyed by pink bollworm (2017 estimates). And, it is not due to inadequate usage of pesticides. On the contrary, farmers’ usage account for 55% of pesticide sales. 

Then, why?

Pest traps are just that – sticky paper to trap pests. Farmers with small landholdings (75% of the output comes from this category) depend on the field staff or the government-appointed extension workers for advice. They go about manually counting the pests trapped on paper, and the data is sent across to experts for advice on the extent of the infestation. Accordingly, the framers would then be advised on how much pesticides are to be used. 

Does this idea seem straight out of the dark ages? It is! Time-consuming, unreliable, and all sorts of doubts cloud the mind. Climatic conditions, coupled with the direction of the wind including the soil composition, affect the kind of pests that swarm in and there’s great variance in the type of attack. Indiscriminately spraying pesticides – as some farmers are wont to – can, and often does, lead to collateral damage. One is of course, to the environment, and two, it leads to change in soil composition, and that would mean more fertilizer usage later on. There has to be an economic threshold, or else the efforts can be counter-productive. Pesticides do not come for free, neither do fertilizers.  

Enter Wadhwani Institute of Artificial Intelligence

The organization is supported by Google as part of the AI Impact Challenge, mentioned earlier. Aided by them, the Wadhwani AI model enables great speed and accuracy to complement the human in the loop. The workers can take a picture of the pest-trap through a multi-lingual app, and the solution monitors pest density by detecting and counting the pests using AI and tracks pest density changes over the cotton crop cycles based on the routes the pests take. These inputs combine with auxiliary farm-level environmental parameters to provide timely advice to individual farmers. This data is processed further for a deeper understanding of the problem – at a macro level. And, like any other AI engine, it goes on learning with every new input and continues to build a higher level of accuracy – hitherto unheard of. 

It also works on a “basic” smartphone that puts paid to doubts on connectivity issues. “We are quite excited to announce that we have just achieved model compression that makes this AI model small enough to put on a really basic smartphone and hence this can now actually work offline,” said Raghu Dharmaraju, Vice President, Products and Program, Wadhwani Institute for Artificial Intelligence during the launch. To add, it’s available as an open-source model that agricultural programs can latch onto any time they want. And, with millions of farmers who are in this segment, the model is immensely scalable. 

The solution can be integrated with existing government systems to serve 80% of the six million cotton farmers in India. 

The Idea of Shared Learning 

Out of the 2600+ applications received by Google for this project, more than thirty were about using AI to identify and manage agricultural pests. These technologies need great volumes of clean data which may not be so easy to come by - more so in the social sector. 

Duplication of efforts can be avoided if these best practices are made shareable. On the practical front, organizations may not be all that amenable to share learnings; after all, similar initiatives are vying for the limited pool of available grants. Again, this is a generic observation and should not be read with the Wadhwani AI example. Google, for instance, requires grantees to responsibly share open-source for their funded projects. But to scale and be company agnostic, it may be a good idea to have a third-party body that helps to develop and aggregate open-source best practices. 

According to the Food & Agricultural Organisation of the UN (FAO), plant pests and diseases can adversely impact 20 – 30% of global food production. That’s huge when you consider this stunning piece of statistics - 820 million people around the world do not have enough to eat. As part of the AI-powered capabilities, Computer Vision can play a very big role in narrowing this gap.