ORMS INFORMS Today | By ORMS INFORMS Today Staff
In a Monday afternoon keynote session, “Multiagent Reasoning for Social Impact: Results from Deployments for Public Health and Conservation,” Milind Tambe of Harvard University and Google Research, proved his talk was worth the wait. (Thanks to the understanding audience and INFORMS PR Manager Ashley Smith for filling in the gaps while technical issues were resolved.)
If you missed the action live, Tambe spoke about using artificial intelligence (AI) to create research solutions for complex societal problems, truly illustrating how INFORMS members are saving lives, saving money and solving problems.
Tambe’s work is centered around the problems of public health and conservation, addressing one key cross-cutting challenge: how to effectively deploy limited intervention resources in problem domains. He presented results from his work around the globe using AI for HIV prevention, maternal and childcare interventions, tuberculosis prevention, COVID-19 modeling and wildlife conservation.
Tambe credits his work in large part to the essential partnership with nonprofits and communities. He says achieving social impact in these domains often requires methodological advances. The goal is to facilitate local communities and nonprofits to directly benefit from advances in AI tools and techniques.
For his HIV prevention work, Tambe looks at preventing the disease among youth experiencing homelessness. He says they’ve identified an AI system that can find the people within a social network who can most effectively promote information about HIV prevention to their peers. A field trial analyzing more than 700 homeless youth found that the algorithm significantly reduced key risk behaviors for HIV transmission within the population.
Tambe’s work on maternal and childcare interventions looks at a nonprofit in India called ARMMAN that provides resources and information to pregnant women and mothers. Part of their resources include a service called mMitra, an automated service that leaves messages for women about resources to use based on age of gestation or the age of their child. Mothers who listen to the messages see significant benefits. Tambe’s work has been to increase the efficiency of the calls, who they go to, and increasing the number of women who in fact listen to the messages.
Some of Tambe’s other work consists of social networks and green security games, more directly, predicting where poachers put traps to catch wild animals illegally. This work combines machine learning and game theory. He also has recent work in rethinking COVID-19 testing. His modeling found that rapid tests can help better contain COVID-19. The results demonstrate that effective surveillance, including time to first detection and outbreak control, depends largely on frequency of testing and the speed of reporting, and is only marginally improved by high test sensitivity. He says delays in diagnosing COVID-19 cases using the PCR method can offset its effect, a shortcoming that a less accurate but faster test like the rapid tests can help with.
There are obvious research challenges in AI for social impact, most obviously lack of data and uncertainty of a key feature of AI for social impact. But the benefits obviously outweigh the challenges, and from Tambe’s work we can clearly see the positive impact of his work on these complex societal problems.