As a keynote speaker at HealthWorld 2022, Michail Bletsas, research scientist and Director of Computing at the MIT Media Lab, talked about the role of AI in life sciences. Building on those thoughts, he talks to T.C. Lowrie about the potential of AI in healthcare and beyond, the role of trust and human input in realizing this potential, and the importance of ensuring that effective learning is not only limited to machines.
AI is becoming increasingly sophisticated, but its uses are still largely centered around e-commerce, in applications that convenience many but only truly benefit a few. Yet in fields such as healthcare and life sciences, its benefits are manifold and its potential could prove game changing. What are your thoughts on the growing role and impact of AI in healthcare?
AI can have a strong impact in any data intensive activity and healthcare is one such activity where the volume of data involved increases constantly. Although AI’s first foray into healthcare proved to be unsuccessful, it was mostly because the technology was both immature and not very effective, and on top of that and most importantly, it was hard to integrate into existing processes. The technology is much better today, and we have realized that it is not there to replace humans but to augment human capability like pretty much every other form of successful technology. Furthermore, in healthcare, we can have good results without exotic forms of AI or large foundational models, such as GPT-3, that seem to be all the rage these days. The trick is to have good data management practices (so that we can feed high quality data into our models) and to pay a lot of attention to making those AI models trustworthy and well-integrated into established medical practice.
AI is not there to replace humans but to augment human capability
Capabilities aside, we have a complicated history when it comes to utilizing AI in healthcare—a mix of jumping on the bandwagon to exploit the marketing potential of new applications and taking a hostile approach when their efficacy is cast as a threat. As digital transformation sweeps over everyday life, is our newfound familiarity with new technologies changing attitudes toward AI?
Machine learning, the prevalent form of AI today suffers greatly from a lack of explainability; i.e. we can’t really tell why a model (trained AI system) produces a specific output. This makes it hard to trust, and significant effort is being placed these days in making AI easier to trust. Also we should never take humans out of the decision loop, especially when we are talking about medical decisions. AI is very good at picking up details that are easy for humans to miss and totally clueless when it comes to connecting cause and effect, which still remains an exclusive human capability. The trick is to make the combination of human and AI better than either of the two, and I believe we are going to achieve that and that will change completely our attitudes towards AI.
Surging computational power, sophisticated algorithms and quality big data have opened up a whole new world of possibilities for AI, and machine learning is advancing in leaps and bounds. Is human learning keeping pace? With Greece set to become a regional tech hub, are we adequately preparing the country’s younger generations to succeed in the emerging landscape?
The more sophisticated our tools become, the more demands are placed on their operators, and when it comes to AI, there is a global shortage of qualified people who understand how these tools work. This is especially important for AI because unless you understand how these tools operate, you won’t be able to trust them and use them productively. Human learning is not keeping up pace for the majority of kids, and Greece is a particularly good example of this phenomenon. The PISA scores clearly demonstrate that radical changes are required at all levels of our educational system so that quality education is accessible to everybody and not only to a privileged minority. Unlike bigger countries, Greece doesn’t have the luxury to leave any kids behind. So the time to start making the required changes so that the educational system develops these unique human-only soft skills such as creativity, analytic ability, communication, problemsolving, and collaboration that machines find impossible to replicate—at least yet ;-)—is yesterday.
Speaking at HealthWorld, you likened AI to hot sauce, noting that much like America’s favorite condiment, it is applied liberally in the understanding that it makes everything better. With hot sauce, sometimes it’s the most unlikely pairings that work the best. What are some outlier AI applications that you think might prove particularly promising?
The current progress in large foundational models is proving to provide a boost in creativity. There are already multiple tools that write snippets of computer code for programmers from simple plain language descriptions, as well as many ways to produce useful text that can accelerate a writer’s creative process. We can use AI to simulate complex environments and reduce the iterative steps required to optimize them. When we can simulate complex environments, we can use them to generate synthetic data that can be used to train AI models to operate in these environments; for example, to create synthetic data of a car accident scene to be used to train autonomous driving systems.
Tools like DALL-E can be used to explore one’s artistic tendencies and to help them quickly explore many different directions that their work could take. By the way, the picture in this article was generated by asking DALL-E to create a picture of a 3D rendering of a medical robot in a hospital.
And like hot sauce in desserts, for example, we shouldn’t apply AI in places where analytical forms are much more suitable.