How to shape a data solution for social impact, according to a data scientist

AUSTIN, Texas — Even as more international development organizations understand how data science can help them achieve their goals, they often don’t know where to begin. “Just add a data scientist!” joked Jake Porway, founder and executive director of DataKind, which pairs volunteers with expertise in artificial intelligence and data science with nonprofit organizations, governments, and social enterprises that could benefit from their support. “Throw on some AI,” he said in a talk at South by Southwest this week, alluding to the reaction he often gets from social change organizations. “Just sprinkle it on. We’ll save lives!” Since its launch in 2011, DataKind has grown to 30,000 volunteers, 250 projects, and five global chapters. Earlier this year, the organization received the first grant from the new Data Science for Social Impact collaborative, funded by the Rockefeller Foundation and the Mastercard Center for Inclusive Growth. With $20 million over the next five years, DataKind will shift its focus, supporting more organizations working in several high impact areas while also working to build more capacity in data science for social good. “When you think about social change problems, they’re actually a specific type of problem,” Porway said. They are “wicked problems,” which are difficult or impossible to solve, in part because of their interconnectedness with other problems, he said. “If we’re trying to solve these problems on the backs of volunteers, something is wrong,” Porway said. Porway describes DataKind as “Doctors Without Borders for data geeks.” As more tech companies provide their own pro bono services, the organization seeks to turn its projects into catalysts, so that nonprofits see the value of data science for social good and continue to invest in this work either directly or through partnerships. Porway is on a mission to see “more data science and AI in the service of humanity,” and spoke at South by Southwest about ways to scope data science projects for impact rather than for profit. During his talk, Porway showed a video of a “racist soap dispenser,” which detected light skin to dispense soap, but failed to do the same for dark skin. “This is a visceral example of the data bias that comes up in so many conversations today,” he said in an effort to convey how a design choice is made anytime data is collected. Porway was making the point that “your results are only as good as your data,” one of five key principles for applying data science to social good. Others included: • Finding problems can be harder than finding solutions. • Your system will optimize for what you tell it to — and nothing else. • Good data science follows good design. • Data for good must mean data for all. Speaking with Devex after his talk, Porway pointed to a hackathon DataKind organized with the World Bank. The finance institution wanted to figure out ways to identify fraudulent proposals for grants more quickly. Previously, one person went through piles of papers, but data scientists wrote a script to geolocate the phone numbers on each of these applications and surface information on whether the companies were registered. “They just missed that because it was too much to do without a machine,” Porway said. He said DataKind volunteers often serve as data therapists for the organizations they work with: “Most people, they just want to talk. They’re embarrassed. They’re ashamed. They say, ‘We don’t have good data,’” he said. “It’s okay. It’s a safe space.” According to Porway, in order to shape a data or artificial intelligence solution for social impact, organizations should ask themselves the following questions: • What is your mission and theory of change? • What are your biggest bottlenecks? • What data do you have available, if any, how was it collected, and what assumptions were made? • What does success look like? • Who will use it and how? • Who will it affect? • Who can change it? • What’s the worst that happens if we fail? • What’s the worst that happens if we succeed? “The only world where this works is one in which everyone affected by this technology has access and agency to creating and changing it,” Porway said. Not only can international development organizations better leverage data science for their own objectives, but they can also ensure the people they aim to serve develop these skills, in order to build solutions closer to the problems. During his presentation in Austin, Texas, Porway mentioned Global Witness, an international NGO that campaigns against human rights abuses and corruption, which worked with DataKind volunteers from Google Brain and Foursquare to build a model to predict mining sites, and is now building out a machine learning team. “We almost always get contacted by the lone wolf, the advocate,” Porway said. Often, they say something like, “Hey, we both know what this can do. Help. These people aren’t focused on it,” he continued. “These projects are trojan horses for what you need to build your own capacity,” Porway said. His call to action for organizations that might be considering how they can apply data science to their social change work is to go ahead and start experimenting: “You can't make it more tangible like that until you actually get people on the ground to start building just to test,” Porway said. Give and take is required on both sides, with technologists needing to be iterative, and international development professionals needing to understand that it will take time. “We’ve had the conversation enough,” he said. “We’re going to have to go to something hands on so you can touch and feel it.”