High Stakes and Hard Choices: Life Science Leaders on the Intersection of Health and AI
AI in Health and Life Sciences panel at Discover Experiential AI. (Photo by Heratch Ekmekian)
Ask a life scientist about the biggest challenges in their field, and chances are their answer will have something to do with data, namely what to do with it. Patient records, clinical trials, lab results, protein structuring, molecular screenings, medical inventory — buried in all that data are insights that can transform patient outcomes and redefine what it means to be healthy. But unearthing those insights can only be done with artificial intelligence (AI), which poses a new suite of challenges.
At Discover Experiential AI, the inaugural event of the Institute for Experiential AI (EAI) at Northeastern University, representatives from industry and academia came together to discuss the role of AI in advancing drug discovery, medical research, and clinical practice, among other fields.
Translating Theory into Practice
Moderator Eugene Tunik, Director for AI + Health at EAI, opened the discussion with a challenge to define the state of AI in the life sciences. For Rai Winslow, director of Life Science and Medicine Research at the Roux Institute, the answer is visible in the healthcare professions.
Medicine, he said, is becoming a computational discipline, and algorithms are becoming a part of healthcare teams. Whether to recommend a patient care plan or to analyze unstructured data sets, AI systems can help doctors, technicians, nurses, and other practitioners make more informed decisions from the information available to them.
But the biggest challenge is how to translate theory into practice: How can real-world applications of AI be quantified to assess their impact on medical outcomes? To answer that question demands a careful approach, one which places the algorithm within the context of a broader, human-driven system. AI, in that sense, best serves as a recommendation tool for clinicians and practitioners rather than a replacement.
Sifting Through the Hype
AI is not exactly short on hype. In the life sciences, diagnostic pathology is perhaps the most hyped-up application for AI.
Laurent Audoly, founder and CEO of Parthenon Therapeutics and director for AI + Life Sciences at EAI, expressed some reservations about whether AI, which is inherently context-dependent, is ready to make the leap to generalizable, disease-agnostic recommendations. While there has been tremendous progress in using AI to predict the structures of disease-causing proteins, these contributions are not exactly game-changing — and certainly not “disease-agnostic.”
What about using AI to help design novel medicines? Audoly is not convinced we’re there yet either, as there isn’t enough data to suggest an end-to-end drug discovery solution. Encompassing all of these hurdles is a need to understand increasingly complex, multi-parametric data sets, and there’s really no way to do that without the help of AI.
Using AI to Make Better Drugs
Jared Auclair, director of the Biopharmaceutical Analysis Laboratory at Northeastern, touched on other challenges in the life sciences sector, including how to help life scientists better understand the AI systems they use. Machine learning algorithms, for example, are black boxes to doctors with no background in data science, which is most of them. And that context is important for understanding how and why a particular AI system produces a given result.
That said, there’s an enormous opportunity for AI in drug delivery. Biotechnology, according to Auclair, is essentially “using cells to make drugs.” That poses a number of storage and transportation challenges that simply don’t exist with traditional pharmaceuticals. But AI can be used to oversee and make sure cell or gene therapy products are grown safely and sustainably, producing information that can be fed back into the development process.
Why Drugs Are so Expensive and How AI Can Help
Melissa Landon, chief strategy officer at Cyclica, helped put the pharmaceutical challenge in a business context: Over 90 percent of drug discovery programs fail; accounting for those failures, it costs, on average, roughly $2 billion to bring a drug to market. And we ask why these drugs cost so much.
“When you are paying for your prescription drug,” Landon says, “you are paying for all the failures.”
The hope with AI is to bring that attrition rate down. A problem there is that people generally don’t want to test things that could be wrong, which limits the availability of negative data that would help develop a robust algorithm. Rendering better, more useful data will require a cultural shift that maximizes learning and incorporates failure across disciplines.
Building Bridges
Aileen Huang-Saad, director of Life Sciences and Engineering at the Roux Institute, sought to answer what it will take to create a more interdisciplinary workforce. One avenue is to build bridges between academia and industry. AI itself shows how interdisciplinary the world has become, so it’s imperative that the design and development of AI reflect the landscape in which it’s used. But we also need to do a better job teaching the context of the problems AI is meant to solve; that includes not just the depth of the technology but also its impact on downstream users.
As is often the case with AI, we return to the question of ethics and the need to center the human experience. Huang-Saad likened traditional higher education to that of a mosaic: Students are being taught how to work within individual tiles, yet they are expected to be able to see the entire picture. The only way to combat this tendency is to recognize the common threads that unite all the different disciplines.
For Huang-Saad, one of those threads is user-centric design. Another is ethics. But it’s not enough to simply make students take a class on ethics and expect them to make more ethical decisions. Competing incentives are at play, many of which have little bearing on the people most impacted by AI. Progress may require a fundamental shift in how we understand problems and how we make decisions to solve them.
In that world, a human-driven model of AI, one which recognizes the vital role of people in combatting the worst outcomes of biased data, is the only way to fulfill the promise of artificial intelligence.
Contact the Institute for Experiential AI to learn more about AI in the Life Sciences or how we can help you achieve your research and business goals.