AI for Social Good in Maternal Health

AI for social good is emerging as a significant focus amidst the buzz surrounding generative tools and profit-driven innovations. At the forefront of this movement is Dr Milind Tambe, who leads the AI for Social Good initiative at Google DeepMind. Together with ARMMAN, a non-profit organisation in Mumbai dedicated to maternal and child health, Tambe’s team is using advanced machine learning models to address a critical public health issue in India: maternal mortality.

Dr Tambe explained the initiative during a discussion, highlighting the complexity behind their work with a calm clarity. He recalled, “The journey started with a conversation at a Starbucks in Mumbai,” where Dr Aparna Hegde, the founder of ARMMAN, shared the significant challenges they faced, particularly concerning participant drop-off rates in programs like mMitra. This dialogue ignited a partnership that’s transforming lives in India.

The Drop-Off Dilemma

The mMitra programme from ARMMAN has been providing essential health information to expecting and new mothers through regular voice calls. However, data indicated a troubling trend: 30-40% of women disengaged early, particularly those most vulnerable. Tambe’s team intervened by developing a predictive AI model capable of identifying these high-risk participants and facilitating targeted interventions.

“We managed to decrease the drop-off rate by approximately one-third for the most at-risk mothers,” Tambe stated. “These women also showed a marked increase in healthier behaviours, such as taking iron and calcium supplements, attending medical check-ups, and accurately tracking their infants’ birth weights.”

A recent impact assessment corroborated this improvement. Among women identified by the AI model for further support, there was a 22% increase in the likelihood of taking iron supplements, a 28% rise for calcium intake, and a 9% improvement in the recording of infant health statistics.

Scaling Up with Kilkari

Encouraged by these findings, the collaboration extended to ‘Kilkari’, a government-endorsed initiative that sends out weekly health messages to pregnant and postpartum women across India. With over 60 million beneficiaries across 27 states, Kilkari stands as the largest mobile-based maternal health information programme worldwide.

In contrast to mMitra, Kilkari lacked detailed demographic data about its users. The AI team had only one behavioural indicator to work with: whether a woman answered her phone.

“Initially, it appeared to be minimal information,” Tambe remarked. “However, we discerned patterns from this data. Some women consistently answered calls in the morning, while others did so in the evening. By grouping these behaviours, we customised call times to align with when each group was most likely to engage.”

The results from a pilot study in Odisha were impressive. Adjusting call times based purely on behavioural patterns achieved a 12% improvement in pickup rates during specific time slots.

“It’s a minor adjustment, but it could have significant ramifications,” he noted. “Higher pickup rates lead to more women receiving potentially life-saving messages.”

Lessons from the Field

For Tambe, the Kilkari and mMitra initiatives represent more than technical achievements; they exemplify how AI can be judiciously and responsibly incorporated into public health frameworks.

He summarises the lessons learned into three key takeaways for AI practitioners:

1. Deep Partnerships: “AI cannot be introduced into communities without careful consideration. We relied on ARMMAN’s local knowledge, their call centre workforce, and their medical insights. This exemplifies genuine collaboration.”
2. Interdisciplinary Teams: “We combined computer scientists, public health specialists, and social scientists. It is people, not AI, that resolve issues.”
3. Frugal AI: “We had to work within the constraints of limited data and computing power. This is not just a Silicon Valley challenge; we developed lean and effective models suitable for resource-limited settings.”

This commitment to frugality and respect for context also enhances data privacy measures. “In Kilkari, we do not utilise any personal data such as age, income, or education. The information is strictly behavioural, anonymised, and confidential,” Tambe asserted.

A Glimpse of AI’s Broader Potential

While the team is focused on expanding Kilkari, Tambe perceives immense possibilities for similar AI initiatives on a global scale. From vaccine distribution to emergency response planning, he envisions AI profoundly transforming public health.

Dr Aparna Hegde, ARMMAN’s founder, echoed this perspective in an official media statement: “AI has demonstrated to be a force multiplier. We are addressing a health challenge while also contributing to India’s national objectives to lower maternal and child mortality. This fills me with great hope.”

Tambe shares this optimism, stating, “There is still progress to be made, but we have already witnessed the potential of thoughtful technology combined with compassionate fieldwork. When used correctly, AI becomes an incredibly human-centric tool.”

As India strives to achieve its Sustainable Development Goals, the integration of AI and public health could emerge as not only beneficial but essential.