Together We Stand

[ 93 ] Supporting decisions in crisis response: Artificial Intelligence for Digital Response and MicroMappers Qatar Computing Research Institute, Social Computing M inutes after the Nepal earthquake in 2015, people flocked to the Internet — particularly social networking services — to gather and share infor- mation. Digital content included expressions of concern, updates on recent actions and logistical details of earth- quake effects, including photos of damage and locations. In the humanitarian space, one common mantra is ‘communica- tion is aid’. However, given the massive amounts of information generated during a crisis, how can humanitarians and responders find strategic insights, or actionable data, in the massive flood of digital content? Can stakeholders improve citizen engagement and local humanitarian response with digital tools? From discov- ery to delivery, Qatar Computing Research Institute (QCRI) connects the science of machine learning and human computing with digital partnerships. QCRI, part of Hamad Bin Khalifa University, conducts world- class, multidisciplinary computing research that is relevant to the needs of Qatar, the wider Arab region and the world. QCRI is a member of the Qatar Foundation for Education, Science and Community Development. The QCRI Social Computing group is a multidisciplinary team of social scientists, computer scientists, program managers and software engineers. 1 We are interested in understanding various phenomena of societal interest through the analysis of social media, and in using social media data to fuel time-critical applications. Our goal is to support decisions made by policymakers, emergency respond- ers, journalists and everyday citizens, who can benefit from knowledge and information generated in online media. QCRI partners with the United Nations Office for the Coordination of Humanitarian Affairs (OCHA) and a digital volunteer community, the Standby Task Force (SBTF) to help curate large digital datasets into various usable information prod- ucts. We collaborate with OCHA to define problems of mutual interest, explore various computer science research questions, and prototype tools to potentially address the opportunities. OCHA and agencies use common operational datasets (CODs) for strategic decisions. Generally CODs are the baseline files/ data (admin boundaries, population numbers, etc) thus for our research these act as a denominator to help provide context or to enable the creation of a map. There is one COD which deals with the Humanitarian Case Load type numbers – number dead, number affected, etc. QCRI engaged SBTF, who provide humani- tarian organizations with real-time crisis mapping and situational awareness support. They are a global volunteer-based network of trained digital humanitarians who represent the first wave in online community emergency response teams. SBTF is part of the larger Digital Humanitarian Network (DHN). The aim of the DHN as a ‘network of networks’ is to form a consortium of volunteer and technical communities and to provide an interface between formal, professional humanitarian organizations and informal yet skilled and agile volunteer and technical networks. With OCHA serving as a field partner and SBTF serving as a digital partner, QCRI explored how research and technology could help solve several signal-to-noise equations. The proposed solution, Artificial Intelligence for Digital Response (AIDR) and MicroMappers, combines machine learning and human comput- ing. Each partner provides ongoing feedback both during and in between emergencies to improve the tools. Over the course of three years, the partnership has been activated for a number of large-scale emergencies: typhoons Haiyan (2013) and Hagupit (2014) in the Philippines, cyclone Pam (2015) in Vanuatu, and the Nepal earthquake (2015). The experience of the Nepal earthquake demonstrates how our partnership functions. Once the earthquake struck, OCHA requested SBTF’s help to provide social information insights and situational awareness in response to the earthquake. SBTF used AIDR and MicroMappers as the primary tools to deliver the request. With over 2,300 unique contributors for the Nepal earthquake response, the SBTF and QCRI teams collectively aggregated, curated and identified tweets and images about damage assessments and needs in Nepal. The result was a highly curated geolocated dataset of 410 images and 219 text items to provide overall insight into the damage assessments. QCRI created map products with image data described as severe or mild, and with text data tagging infrastructure damage, urgent needs and response efforts. SBTF incorporated these insights into information products given to over 250 aid organizations in their remit, which also included a 200-page document of all responding agencies, what they were doing and where. We recently conducted research on the efforts of contributors for the Nepal earthquake response 2 , and found that the majority were from northern regions of the world, that there were some ‘super-users’ who contributed extensively, and that there may be a correlation between news items about MicroMappers and the peak contributions to the activation. For the Nepal earth- quake, a BBC interview about AIDR and MicroMappers led to a subsequent peak of contributions. We see this opportunity T ogether W e S tand