16/10/19

Q&A: AI for developing countries must be adaptable and low-cost

Wadwani in the field 1 - Main
Wadhwani AI has developed a smartphone-based solution that classifies pests based on photos provided by cotton farmers and offers localised advice on pesticide use. Copyright: courtesy of Wadhwani AI

Speed read

  • Broad partnerships are crucial for low-cost artificial intelligence development
  • Problems as diverse as pest control and infant mortality can be helped by AI
  • AI solutions must be intuitive for users

Send to a friend

The details you provide on this page will not be used to send unsolicited email, and will not be sold to a 3rd party. See privacy policy.

Even tech outsiders have heard the hype about artificial intelligence — whether it facilitates a new drug discovery or early disease detection, it seems a promising artificial intelligence (AI) application debuts every week.

As an intervention that relies on powerful computing capacity and highly educated software engineers, AI might easily deepen the divisions between affluent and lower-income areas of the world.

But organisations such as Wadhwani AI, an India-based nonprofit, seek to avert that outcome. The group is working on three projects: helping cotton farmers to reduce crop losses through smarter pest management; providing rural healthcare workers with computer vision-equipped smartphones to screen newborns for dangerously low birth weights (without the need for data or network connection); and tackling tuberculosis through more accurate estimations of risk factors and regional caseloads.

“Governments, implementation agencies and programmes need an understanding of what AI can do and what problems cannot be solved through AI,”

Neeraj Agrawal, senior programme director, Wadhwani AI

Neeraj Agrawal, senior programme director at Wadhwani AI, talks to SciDev.Net about how his organisation approaches AI solutions for low-income communities, and how those solutions might be scaled elsewhere.

Emerging technology is usually seen as a high-expense solution. How can Wadhwani’s work serve as a model for cost-effective AI solutions?

We pay a lot of attention to making our AI-based solutions potentially adaptable to a variety of contexts and believe that AI-based solutions must work within existing broader systems and programmes, with minimal additional input.

We continuously interact and engage with all stakeholders — beneficiaries, users and decision makers — to identify big problems worth solving. Most importantly, our solutions are available free of cost for implementation and scaling-up in contexts where these are needed.

What needs to happen for AI costs to be scaled down to make it more of an approachable, usable option in developing regions?

Several things. Governments, implementation agencies and programmes need an understanding of what AI can do and what problems cannot be solved through AI. Systems and programmes need to be able to prioritise their problems. Stakeholders including donors, development partners, research organisations and governments need to come together not only to develop partnerships, but also to identify and develop sustainable business models around AI solutions.

How did Wadhwani select maternal/neonate health, cotton farming and tuberculosis as priority areas of focus?

We interacted with a variety of stakeholders in the areas of health and agriculture to gather insights around current status, data, issues around service delivery, client perspectives and programmes. For example, along with looking at a lot of data, we engaged and gathered insights from 30 global tuberculosis experts for deciding to take on TB as a thrust area.

We typically ask seven questions that serve as a set of anchoring principles to help us decide if a project can benefit from the use of modern AI. Is this a big problem? Does it have an AI solution? Will solving the AI part make enough of a difference? Will the solution be accepted by stakeholders? Does the data exist, or can it be created easily enough? Are there partner organisations that can co-create and pilot the solution? Are there existing programmes and pathways to scale?

Have you seen any results or measurable differences in each of the focus areas?

Initial results for our work in cotton farming and smartphone-based anthropometry [making measurements of the human body] are quite promising and we are moving ahead with our field experiments. We were recently named as the official AI partner to India's Central Tuberculosis Division (CTD) [part of the Ministry of Health and Family Welfare]. We are excited about using artificial intelligence and machine learning to help the CTD in its goal of ending the threat of tuberculosis in India by 2025.

Beyond cost, what are the challenges to getting local users to adopt or test AI solutions in the field? How does Wadhwani intend to overcome these challenges?

With our field experiments, the initial challenges we are facing include availability of mobile and data networks, hesitancy in using new technology, and tech illiteracy. Several community-based healthcare providers do not own a smartphone, and are unable to use app features due to poor literacy. “Trainability” of farmers and frontline health workers was also quite challenging initially.

To overcome these challenges, we are working on making the solutions as intuitive as possible, so that a minimum amount of orientation can help end-users start using them. We are also working to make these solutions work without network connectivity.