Artificial intelligence has become a core part of Google’s products in recent years and occupies a major segment of the I/O keynote each year. While the company has been aggressively pushing for the use of AI and Machine Learning models in its products, it has also showcased what happens when AI is used in other cases with healthcare being a prominent example.
Previously, Google has shown how its machine learning models, which are the stepping-stones for AI technology, have helped with detecting diabetic retinopathy more accurately. In this year’s keynote Google also showcased how one of its machine learning models could help with detecting lung cancer at an early stage, potentially saving more lives.
During the keynote, Google’s head of AI Jeff Dean spoke about the company’s AI Impact challenge, “Our AI for Social Good programme brings together our efforts to use AI to explore and address some of the world’s most challenging problems.” For the Google AI Impact challenge, the company has selected 20 organisations, which are using artificial intelligence to solve problems across the world.
Google has chosen Wadhwani AI from India, which is using AI to solve the problem of pest infestation in cotton crops, one of the major challenges faced by farmers in the country. The company claims that the lack of right knowledge and paucity of resources means that farmers end up ruining their crops by using the wrong pesticide at the wrong time or too much of it.
The organisation wants to solve this problem using artificial intelligence and machine learning models which can identify different kinds of pests using photos from pest traps set by farmers.
Padmanabhan Anandann and Raghu Dharmaraju launched the project last year in India in September around the monsoon season with some field experiments in Tamil Nadu.
“If we talk about India, 60 per cent of the population is small farmers and their families. Now, how do they know what to do about their crops, water plants, when to spray pesticides, and things like that? It is predominantly through the large agricultural programmes,” Raghu told indianexpress.com.
He explained that right now in Maharashtra alone, there are over 10,000 extension workers who are out visiting demo plots, keeping track of number of pests and counting them.
“They also use apps but they have to manually count different types of pests and enter that in. That goes into a data centre where experts now have this manually counted pest data that is often not reliable because these folks do not have the expertise to do that in time,” he explained. That where AI could solve some problems by giving timely action to farmers.
Their AI works by analysing the image from pest traps, which in turn showcases the kind of pest and also issues a warning to contact local officer if the pest infestation is serious or depending on how harmful a pest is.
Right now, only once a farmer connects to experts, they can guide them through the safety measures. But the AI app gives out initial information and thus lets the farmer take action quickly. “So you are talking about timely, localised, and actionable advice,” he pointed out.
Wadhwani AI claims to have developed algorithms that can detect three major pests. A field demo in partnership with the Better Cotton Initiative and the Government of Maharashtra has been completed as well. The company is looking to develop a more comprehensive tool for testing by October this year and will open source its APIs and SDK kits for others to use.
The project is in its nascent stage and is currently cloud-based. For it to be used extensively, it needs to work offline, something that is in the pipeline, he says. “These models will be compressed to the point that actually work offline on the phone,” Raghu added.
So far, AI models for three pests have been created, but the company aims to add detection for three more pests, that have been identified as doing the most damage in the last few years. This includes the Pink Bollworm that is said to have wiped out 50 per cent of cotton crops in the 2017-18 crop season in India.
Another challenge is the timely data collection for the machine learning models to train and Anandann says they receive around 800 photos of pest traps from farmers everyday. The pest traps are a part of existing workflow, but farmers use manual count to identify pests and not AI, which is often inaccurate. That’s where the duo are hoping they will make the biggest change with their tools.
Disclaimer: The author is in California on the invite of Google India.