by Muhammad Aurangzeb Ahmad, Hemant Purohit, Oshani Seneviratne
This symposium was centered around the broad vision of how AI could be used for social good through supporting solutions toward the United Nations’ sustainable development goals that touch every aspect of human, social, and economic development. The symposium identified the critical need for responsible AI solutions for these goals, which demand holistic think- ing on optimizing the trade-off between automation benefits and their potential side-effects.
The motivation behind the AAAI symposium was to have a forum on how AI technology can have an incredible impact on social good. We organized the symposium around two themes that are crucial concerning social good in the age of burgeoning AI: (1) Responsible AI in healthcare, and (2) Humanitarian Relief and Development.
Healthcare data, in general, is characterized by missing data, lack of data standardization, and data incompleteness, which hinders the deployment of solutions which are relevant to real-world use cases. More- over, AI in healthcare is characterized by the last mile problem, where delivering a practical solution that is reliable, robust to errors especially in ”break glass in case of emergency” situations, and graceful degradation has proven hard to implement. These have broader implications in the context of fairness, explainability, and transparency. There- fore, implementation and deployment of AI systems in production in healthcare bring up challenges that go far beyond model building and scoring. In this symposium, we focused on a broad range of AI health-
care applications and problems encountered in the context of deployment of AI in health- care.
Similarly, in the humanitarian relief and development space, we believe that detect- ing and predicting how a crisis or conflict could develop, analyzing the impact of catastrophes in a cyber-physical society, and assisting in disaster response as well as re- source allocation is of utmost importance, and the advances in AI can be utilized in many such tasks. These techniques can allow better preparation for emergencies, help save lives, limit economic losses, provide adequate disaster re- lief, and make communities stronger and more resilient. The symposium has several papers that highlighted where the utility of AI technology would be useful in this regard.
The symposium program involved a mix of activities from two keynotes and invited panels to breakout groups and paper presentations across three days. Two keynotes illustrating the scenarios of where AI could benefit the most were given by the well- accomplished speakers with extensive experience in lead-
ing, researching, and developing AI-infused solutions for diverse applications from healthcare to humanitarian operations. The keynote by Dr. Ankur Teradesai from the University of Washington emphasized the importance of creating responsible AI systems that are accountable, transparent, ethical, robust, and explain- able, with specific applications in the healthcare domain. The AI in ”AI in Healthcare” is the easy part; delivering insights that are usable to healthcare us- ing AI is the hard part. Several common themes emerged dur- ing the panel discussions and breakout group discussions that can be broadly categorized into three areas. First, the suc- cess of AI in many domains in the last few years has been observed in narrowly-defined problems with limited generalization capabilities. It presents a danger that overhyping of AI can lead to a backlash against the usage of AI and may re- tard its integration in the real world. It is partly also due to the media coverage of inflated claims about AI; thus, responsible communication about the limitations and expected out- comes is necessary. Second, the human-centric systems of the world have well-defined roles and responsibilities, for which a person could be made accountable. However, the cur- rent thinking of AI solutions as the perfect stand-alone solutions for real-world problems would be dangerous. Especially due to the lack of evaluation methods and the cost of machine errors that are not very well-characterized when such AI algorithms are integrated into the real-world systems. Third, researchers need to worry about the dan- ger of AI solutions leading to unintended consequences when considering one to one mapping between a problem in the real world and an AI model. For instance, consider the problem of reducing the risk of readmission in hospitals; creating a machine learn-
ing model that can predict the risk of readmission and then making decisions based on its insights can lead to a re- duced risk of readmission, but it might also likely result in the prolonged length of stay in hospitals.
This symposium brought together AI researchers, do- main scientists, and policy- makers to exchange problems and solutions, to identify synergies across different application domains, and to lead to future collaborative efforts. One of the outcomes of the symposium was to organize the ’AI for Social Good in South Asia’, which would be targeted to the specific problems from that region that the co-chairs are passionate about. Given the success of this symposium and greater interest by both researchers and the practitioner community of government and non-government organizations and we plan to again organize the follow-on symposium for AI for Social Good.
This symposium was co- organized by Dr. Muhammad Aurangzeb Ahmad, Dr. Hemant Purohit, and Dr. Oshani Seneviratne.
Dr. Muhammad Aurangzeb Ahmad is an Affiliate Assistant Professor at the University of Washington Tacoma and Principal Data Science Researcher at KenSci Inc.
Dr. Hemant Purohit is an Assistant Professor of Informa- tion Sciences and Technology at George Mason University.
Dr. Oshani Seneviratne is the Director of Health Data Re- search at Rensselaer Polytechnic Institute.