AI for Social Good - AAAI Fall Symposium 2020

Welcome & Opening Remarks (slides)

Keynote: AI and Machine Learning in Understanding and Predicting COVID-19 (slides)

COVID modeling and prediction is a crowded area, but in this instance, more crowd is better. We study the fundamental AI and Machine learning challenges relevant to COVID prediction, and how we addressed those challenges. One has to pin down what we would like to learn from the data, and reconcile that with what we can learn from the data. What we can learn from the data tells us what is happening as opposed to what we think ought to happen. Focusing on what we can learn drove our focus toward simple local models with robust change-point analysis and model identification. Our approach allows data at the county-level to dictate the learning for a county, which allows us to infer quantities such as the number of infectious and immune individuals at a county-level. Such quantities are not easy to deduce from other models or without extensive infection and/or antibody testing. Our algorithms are fully automated and in a team effort, we have produced two useful tools: COVID War-Room and COVID Back-to-School. Most of our results will be drawn from these tools.

Robust Lock-Down Optimization for COVID-19 Policy Guidance (slides)

Reinforcement Learning for Optimization of COVID-19 Mitigation Policies (slides)

COVID-19 in Spain and India: Comparing Policy Implications by Analyzing Epidemiological and Social Media Data

Measuring and Visualizing Social Distancing Using Deep Learning and 3D Computer Vision

Towards Context-aware Knowledge Entailment from Health Conversations

Emergency Department Optimization and Load Prediction in Hospitals (slides)

Towards Explainable Question Answering (XQA)

Asymptotic Cross-Entropy Weighting and Guided-Loss in Supervised Hierarchical Setting using Deep Attention Network

Keynote: Friends Don’t Let Friends Deploy Black-Box Models: The Importance of Intelligibility in Machine Learning

It’s more dangerous to deploy a black-box ML model than you think --- there are bad things lurking in all models. We have developed a machine learning method based on generalized additive models (GAMs) that is as accurate as full complexity models like boosted trees, random forests and neural nets, but even more intelligible than linear and logistic regression models on tabular data. This makes it easy to understand what a model has learned and to edit the model when it learns inappropriate things. In the talk I’ll present case studies where these high-accuracy GAMs discover surprising patterns in data that would have made deploying a black-box model risky or unfair.

Health Care Misinformation: An artificial intelligence challenge for low-resource languages (slides)

Lessons Learned from Problem Gambling Classification: Indirect Discrimination and Algorithmic Fairness

Artificial Intelligence and Resource Allocation in Health Care: The Process-Outcome Divide in Perspectives on Moral Decision-Making (slides)

Assessing Fairness in Classification Parity of Machine Learning Models in Healthcare (slides)

Direct Multi-Step Forecasting with Multiple Time Series using XGBoost: Projecting COVID-19 Positive Hospitalization Census for a Southern Idaho Health System

A Two-Step Framework for Parkinson’s Disease Classification: using Multiple One-Way ANOVA on Speech Features and Decision Trees

Socioeconomic and Geographic Variations that Impact the Spread of Malaria

Health Panel

• Chair: Dr. Muhammad Aurangzeb Ahmad (University of Washington Tacoma and KenSci)
• Carly Eckert, MD (Department of Epidemiology, University of Washington)
• Vikas Kumar, PhD (KenSci)
• Nicholas Mark, MD (Swedish Hospital)
• Oshani Seneviratne, PhD (Rensselaer Polytechnic Institute)

Symposium Day 1 Closing (5 mins)

Welcome

Keynote: AI for Social Good: Small Nations' Perspectives

As AI has increasingly integrated in every walk of life and affecting the lives of billions of people, more than a hundred frameworks have proliferated globally from various governments and large organizations. However, a perspective that combines the concerns of peripheral actors in the AI world and small nations is largely missing. In this talk I shall give a brief overview of the benefits that AI can offer to small nations and marginalized groups. I will also use this as an opportunity to implore the AI community to focus their efforts towards a big tent approach that respects such concerns.

AI-assisted Accurate School Mapping in Developing Countries

Human Trafficking Screening in Humanitarian Relief and Development

Clean Water: How the AI community can contribute to accessing water sources in developing countries (slides)

AI against Modern Slavery: Digital Insights into Modern Slavery Reporting - Challenges and Opportunities (slides)

Using AI to Identify Optimal Drilling Locations for Sustainable Irrigation for Subsistence Agriculture

Keynote: Intelligence for Peace: AI in UN Field Operations and Cyber-peacekeeping
(slides)

The peacekeepers of the United Nations are deployed to over a dozen war-torn areas of the world to monitor ceasefires, create buffers between combatants, provided humanitarian assistance, protect civilians, rebuild societies, and occasionally enforce the peace. Information is the life blood of UN peacekeeping and the world organization has made progress in incorporating sensors in the air (e.g., drones) and on the ground (remote cameras) but it still lacks AI tools to analyze this vast and increasing pool of information. Some possible AI applications and pilot initiatives are described, especially for image analysis and natural language processing. Given that conflict has now spread into cyberspace, what kinds of peacekeeping activities can be done in cyberspace? Surprisingly, many of the functions of peacekeeping operations in physical space have potential analogies in cyberspace. The possibilities and obstacles are many but there is a realistic hope for more intelligent peacekeeping in both physical space and cyberspace.

Uncertainty Aware Wildfire Management (slides)

Exploring AI-Enabled Use Cases for Societal Security and Safety

Designing Emergency Response Pipelines : Lessons and Challenges (slides)

Data Collection and Annotation Pipeline for Social Good Projects

Arguments as Social Good: Good Arguments in Times of Crisis

Keynote: Initial steps towards Assistive Augmentations

The overarching topic of this seminar is centered on the design and development of novel human computer interfaces that seamlessly integrate with a user’s mind, body and behavior, providing an enhanced perception. We call this ‘Assistive Augmentation’. Creating such Assistive Augmentations poses a twofold challenge as they require a novel input and interactions as well as a holistic design approach. Using modern biological understanding of sensation, emerging electronic devices, computational methods and design thinking approach, we now have an opportunity to design a new generation of Assistive Augmentations. This talk will present several proof of concept Assistive Augmentations for enhancing human I/O in the focus areas of assistive technologies, novel input strategies, smart health and well-being, and interactive learning technologies.

Humanitarian Panel

• Chair: Dr. Hemant Purohit (George Mason University)
• Dr. Jennifer Chan (Northwestern University)
• Dr. Walter Dorn (Royal Military College of Canada & Canadian Forces College)
• Mr. Steve Peterson (National Institute of Health)
• Oshani Seneviratne, PhD (Rensselaer Polytechnic Institute)

Symposium Closing (15 mins)