FLASH TALKS, ROUND 1
Andrew Beam, Harvard T.H Chan School of Public Health
Machine Learning and Artificial Intelligence in Medicine: What’s Old, What’s New, What’s Real?
In this talk, Andrew Beam will present an overarching conceptual framework that relates the recent breakthroughs in AI to previous and well-known approaches in clinical decision support. He will conclude with a summary of the key ways in which current AI differs from previous approaches to provide a reasonable set of expectations for what is possible with AI in healthcare.
Leo A. Celi, MIT Laboratory for Computational Physiology
The Reproducibility Crisis in the Age of Artificial Intelligence
As databases of medical information are growing, the cost of analyzing data is falling, and computer scientists, engineers, and investment are flooding into the field, AI in healthcare is subject to increasingly hyperbolic claims. Every week brings news of advances: superior algorithms that can predict clinical events and disease trajectory, classify images better than humans, translate clinical texts, and generate sensational discoveries around new risk factors and treatment effects. Yet the excitement about AI poses risks for its robustness: we must take steps to avoid a reproducibility “crisis” of the kind that has engulfed other areas of biomedicine and human science in the last decade and shaken public confidence in the validity of scientific work.
John Halamka, Beth Israel Deaconess Medical Center
The Potential for Machine Learning in Healthcare
The speaker will review digital health examples from around the world, highlighting the challenges and urgency to implement new solutions, and the technologies used. He will share lessons learned from the Gates Foundation and his work as Harvard’s International Healthcare Innovation Professor. He will also reflect on the real world possibilities of emerging technologies from his experience as a CIO for over 20 years.
Merce Crosas, Harvard Institute for Quantitative Social Science
Responsible and FAIR (Findable, Accessible, Interoperable, Reusable) Data Sharing
Data sharing is becoming a key part of the research lifecycle. Funders require it to make public assets available to the public. Journals require it to enable replication of published research results. But data sharing is not simply posting your data in a website or sending them by email to other researchers; proper data sharing must support findable, accessible, interoperable, and reusable (FAIR) data. This talk will introduce the FAIR requirements for data sharing and show how the Dataverse repository is aligned with these requirements.
Adam Landman, Brigham and Women’s Hospital
We have an unprecedented opportunity for digital health and artificial intelligence to improve healthcare delivery, addressing the Quadruple Aim (improving health, patient experience, efficiency, and clinician experience). While new technical advancements and digital applications are being developed daily, evidence which demonstrates their effectiveness and impact on health is scarce. Pairing the latest technical advances with academic medical center experts to test and refine solutions, will help us identify valuable solutions. We will describe how an innovation accelerator can bring more patient-centered, efficient and safe care through use, development, evaluation and commercialization of digital health solutions.
John Brownstein, Boston Children’s Hospital
Machine Learning and the Future of Precision Public Health
Through social media, forums and online communities, wearable technologies and mobile devices, there is a growing body of health-related data that can shape our assessment of human illness. Collectively, this data comprises an individual’s ‘digital phenotype’ – unique, unsolicited and real-time information about a person’s health. Our current research focuses on using digital phenotypes for population health surveillance, specifically to identify and analyze specific sub-populations over space and time with the goal of better understanding patient behavior and disease dynamics. Some current research topics include foodborne illness, insomnia, autism, febrile illness, and patient experience.
Peter Szolovits, MIT Computer Science & Artificial Intelligence Lab
The Omni-Ome: Heterogeneous Data Promises Better Predictions
Modern machine learning methods can make accurate prediction and classification decisions. Access to enormous data sets helps, and so does availability of different kinds of data: not only imaging, but also vital signs, lab values, prescriptions, procedures, signal data, and many kinds of narrative text. We also need to identify the most useful targets for prediction and how to develop trust in our models.
Ashley Nunes, Harvard Law School
Can AI Cut Costs? It’s Complicated
The view that AI cuts costs – while common – isn’t necessarily true. Drawing on examples from the transportation industry, our work shows that while technological progress can improve performance, it also – contrary to mainstream opinion – can cause costs to rise. This suggests betting the wholehearted embrace of AI may be unwarranted.
FLASH TALKS, ROUND 2
I. Glenn Cohen, Petrie-Flom Center for Health Law Policy
AI in Health Care: Legal and Ethical Issues
This talk will give a very quick overview of the legal and ethical issues raised at the various stages of building and implementing an AI model in medicine from data collection to broad dissemination.
Ryan Budish, Berkman Klein Center for Internet & Society
International AI Governance: New Risks and Old Law
International AI governance has moved quickly over the last few years, with many countries, international organizations, and even companies releasing broad AI strategies and ethical principles. But how do we move from high-level principles to rules that can be operationalized? And how do we merge those principles with existing rules particularly in highly regulated fields, such as health?
Eva Guinan, Laboratory for Innovation Science at Harvard
Moving Toward Smarter and More Strategic AI Initiatives
There is huge capacity for artificial intelligence approaches to multiple aspects of global health. Because this work is costly in many dimensions, optimizing strategies, processes, and applications within specific projects can help improve outcomes and generate evidence-based practices to accelerate productivity more generally. An example of a multipronged approach is provided as a basis for discussion.
Rifat Atun, Harvard T.H Chan School of Public Health
AI, Health Systems and the Innovation Challenge
AI has tremendous potential to transform health systems by improving efficiency, effectiveness, equity and responsiveness of care, and enhancing health outcomes. There are also risks of unintended consequences of AI. Yet, as with many innovations, the adoption and diffusion of AI at scale could be hampered by health systems constraints, which need to be overcome to harness benefits of AI.
Mark Michalski, MGH & BWH Center for Clinical Data Science
From Local to Global – Challenges and Strategies for Making Clinical A.I.
Phuong Pham, Harvard Humanitarian Initiative
Digital Data Collection and AIs
Since launching the public version in late 2014, KoBoToolbox has sustained exponential user rates with users in many countries and 6 million submissions per month. KoBoToolbox has been formally adopted by the United Nations Office for the Coordination of Humanitarian Affairs (OCHA), United Nations High Commissioner for Refugees (UNHCR), and many major international, national, and local NGOs. The presentation will focus on the experiences, both strategy and challenges, to develop KoBoToolbox for all users especially those with little digital resources and the role of AI in advancing the next generation of digital technology.
Rich Fletcher, MIT D-Lab
AI in Global Health: Applications and Pitfalls
Several generations of information and communication technologies (ICT) have been used in Global Heath over the past 30 years, with the recent emergence of mHealth 1.0 (simple phones), mHealth 2.0 (smart phones), and now mHealth 3.0 (phones+AI). Artificial intelligence (AI) now provides enormous opportunities to implement health diagnostics, disease screening, and decision support in places where resources and skill levels are in short supply. However, deployment of AI solutions must include proper knowledge of the local culture and stakeholders. AI algorithms and must also be implemented properly to avoid problems with fairness and bias that might exacerbate problems of health disparities.