Empathy, ethical AI and breaking into the black box when creating data insights
When Kenneth Davis of Fraym first heard about the COVID-19, he was on a business trip in Nairobi and became very ill himself. Though he did not have the virus, his personal experience of being ill while news stations shared the news of the pandemic has informed his awareness of the importance of data and technology as a vital and reassuring part of the response to the pandemic.
“Logically, I knew that I had probably just caught the flu mixed with jetlag,” he explains in an email interview. “But sitting in the Frankfurt airport during my layover, shivering with the chills… there was this very primal, fearful part of me that kept asking — what if this is it? What if I don’t make it home? In retrospect, the most unsettling part of that memory is the image of me traveling, mask-less, while obviously sick.”
The irony of the situation was that he had spent several days previously working on a project involving a different widespread pandemic. “The world was already starting to feel very precarious,” he says.
Fraym uses machine learning-generated data on population characteristics and behaviours to solve large-scale challenges and help governments and organisations to make better decisions on how best to respond. Members of the company’s leadership team have held senior positions in The White House, the U.S. Treasury, the African Development Bank, the Center for Global Development, and the ONE Campaign and all of them faced a consistent and fundamental challenge — sourcing localized data to make informed decisions in emerging markets. “Like most innovations, Fraym was born out of frustration,” says Kenneth. “Many important decisions in developing economies are made with highly inadequate information. Often, analysis is limited to anecdotes, aggregate statistics, or gut instincts. Fraym was created to fill this gap.”
Notes from a pandemic
Over the past year as the pandemic spread and evolved, Kenneth and his team have observed important changes in the way that data is, can and should be used. While statistics and high-level trends are common, they are not as impactful as maps and locations. Spatial data is vital for politicians, city planners, public health experts as well as parents, teachers and small business owners, it helps everyone to visualise disease spread and impact on the economy and health infrastructure in order to make informed decisions.
“Decision-makers and families do not have the time to sort through and analyze hundreds of datasets — they need answers, alongside a summary of how those answers were derived from evidence so they know they can trust it.”
Granular data is also key. “Aggregate data is just not as useful at the region, county or district level,” says Kenneth. He says that the effects of COVID-19 on health, finances, employment, and education have been completely disparate and that without community level data, it is impossible to fully understand and respond to the needs of real people.
Data overload is also becoming a problem. “We need simple, strategic data,” Kenneth says. “The issue today is not a lack of data, but so many different sources of data in so many different locations. It’s overwhelming and impossible to make sense of. We need to be delivering insights rather than raw data. Decision-makers and families do not have the time to sort through and analyze hundreds of datasets — they need answers, alongside a summary of how those answers were derived from evidence so they know they can trust it.”
Machines and intelligence
Fraym uses machine learning (ML), a form of artificial intelligence, to produce spatial data on population characteristics at scale. “Before these technologies, we were making decisions based on broad indicators that covered large swaths of populations,” says Kenneth. “The discrepancies in health, education, and poverty between neighborhoods, cities, suburbs, and rural areas was completely obscured. This required extensive, often in-person data collection and ground truthing for every single project and initiative. This was costly, time consuming, and often overlooked entirely. ML allows us to provide deeper insight at a fraction of the time and cost.”
“Unfortunately, the term “black box” makes it seem as though those working in AI or ML do not thoroughly review and consider the inputs and outputs of their models.”
While new developments are helping to create insights artificial intelligence, whether this is machine learning or in other forms, has been dubbed a ‘black box’ technology — in other words, it can be hard to explain what happened between the time when the data went in and how it came out. Kenneth is thoughtful about how these technologies are handled best.
“Unfortunately, the term “black box” makes it seem as though those working in AI or ML do not thoroughly review and consider the inputs and outputs of their models,” he says. “At Fraym, we understand the inputs, outputs, and the research-backed process in-between. We carefully identify, clean, and harmonize the inputs that are critical to producing actionable outputs for implementers, donors, and researchers. Since we know why and how the outputs are created, the “black box” of these ML models does not create ethical issues for us.”
Fraym’s use of machine learning follows rigorous ethical standards and its community level data works to protect individual privacy. Harmonization of data from different sources can be labour intensive but it is always worthwhile to obtain better results.
“It is a significant amount of work to standardize and harmonize such a wide variety of data,” says Kenneth. “Our data scientists closely examine representativeness, sampling frames, questionnaire coverage, periodicity, and a range of other factors. A lot of time goes into just cleaning the data. The main issue is the cost associated with that time — so we have invested heavily in automating this process where possible and designing robust standards for every other aspect.”
Forward thinking
For the past several years, Fraym has consistently grown its capacity to develop this hyperlocal insight more quickly and efficiently and deliver it through more comprehensive products across an expanding range of sectors. “Ultimately, our goal is to make this level of granular, spatial data the standard best practice in informing decision making across development sectors. Whether you are focused on the elimination of poverty and hunger, the improvement of health, education, and gender equality, combating climate change, or building a more peaceful world — you need real information on populations, wherever they are. We want to help organizations and agencies have a deeper, more nuanced understanding of the people they are working to improve the lives of.”
To continue this work, the company needs support in getting data into the hands of decision-makers. “We could use support in building partnerships with additional international and government agencies so they know this data and these products are available and how they can support their work,” Kenneth says.
“We need to move quickly to mitigate the long term, downstream effects of the pandemic. It is critical that we invest the time and resources today, or we will need to invest far more in the future.”
While keeping an eye on positive future work, Kenneth still has concerns about what may happen next, “I worry a lot about backsliding,” he says. “As a global community, we have made tremendous strides in combating disease, reducing poverty, and improving education. The COVID-19 pandemic has threatened that progress, putting many communities at significant risk. We are anticipating a significant increase in HIV, TB, and Malaria because of interruptions in health services and delivery of prevention products. We need to move quickly to mitigate the long term, downstream effects of the pandemic. It is critical that we invest the time and resources today, or we will need to invest far more in the future.”
While it is easy to look at the strong technical work being done, that fateful trip to Nairobi where Kenneth was so unwell, still resonates in the work he does. “I think it made me more empathetic and sensitive to the very real threat disease outbreaks pose to humankind, and the havoc they can wreak for generations. I felt incredibly privileged to return home to a city that was shortly afterwards shutdown, to a now entirely remote job. That is a luxury that so many others do not have — here in the U.S. as well as in other countries around the world. I hope this shared, global experience reinforces how reliant we all are on one another. How important it is that we ensure all our basic needs are taken care of, as a priority. Poverty, hunger, disease — they make all of us more vulnerable, but some of us more than others.”
