Artificial intelligence has the potential to revolutionize healthcare and mitigate clinicians' administrative burden. Many health systems, however, are still only in the earliest stages of changing their workflows to incorporate AI capabilities.
To learn more about AI and how leading healthcare organizations are leveraging the technology, Becker's Hospital Review recently spoke with V. "Juggy" Jagannathan, PhD, AI evangelist at 3M Company.
Note: Responses edited for length and clarity.
Question: How did you become an expert on artificial intelligence?
Dr. Jagannathan: In 1981 I earned my PhD researching expert systems to analyze electroencephalography signals. Since then, my career has been in industry creating natural language processing solutions for clinical documentation at M*Modal, now part of 3M. At M*Modal, I have designed AI systems that can improve clinical documentation and extract quality measures by analyzing and understanding the clinical text. I now work at 3M researching solutions that understand conversations between clinicians and patients.
Q: We hear the term "AI" everywhere. What does artificial intelligence actually mean?
VJ: If you are communicating with an external system and you can't determine whether that system is a computer or a human, then the system is exhibiting intelligence. This definition of AI goes all the way back to 1952 and Alan Turing's Turing Test, partly the subject of Hollywood movie, "The Imitation Game."
AI has several subcategories including text understanding which Columbia Professor Naome Sagar pioneered using early computer systems in 1959 for medical language processing.
DARPA (Defense Advanced Research Projects Agency) funded a huge amount of research on speech understanding, another form of AI, at Carnegie Mellon University where the founders of M*Modal got their start.
Q: Is machine learning the same as AI?
VJ: Machine learning is one of many approaches to creating a system under the umbrella of artificial intelligence. In the late 1990s, ML-algorithms utilizing feature extraction became popular. Feature extraction is a process by which, if you want to predict an image of a dog, the developer will first extract different aspects of the image such as the dog's ears, tail, eyes and then feed these to a statistical learner to classify the image.
Starting in 2010, increases in memory capacity and CPU speeds allowed analysis of large volumes of raw data in a type of machine learning referred to as deep learning or neural networks. In the dog example, we simply feed the image of a dog, and state it is a dog, and the deep learning model learns how to identify the animal on its own. Now, these deep learning systems are the engines behind advances such as automated driving, creating realistic art and even writing novel stories.
These systems don't exhibit common sense and their decisions are opaque. This has led to a resurgence of hybrid AI systems that combine human knowledge with deep neural networks to inject explainability, eliminate bias and provide common sense guard rails.
Q: Which hospital workflows have been most impacted by AI?
VJ: I think the real question is what aspects of healthcare haven't been impacted by AI. A recent blog post I wrote explores how AI affects every aspect of healthcare, including hospital systems. One place where AI is having a huge impact on hospital workflows is in clinical documentation because of AI's ability to monitor documentation for critical omissions.
Hospital revenue cycles are also being impacted by AI. Coding is an extremely complex task required for a hospital to get paid in which they translate a provider's narrative into a set of codes from a list of thousands of ICD-10 codes. With AI and natural language understanding solutions, it's possible for systems to understand what clinicians have documented and code it effectively.
Q: How are care providers leveraging AI today?
VJ: Clinical diagnostic support is an AI-based solution that alerts physicians if, for example, they didn't order a medication for a particular patient condition. Initially, these systems were fairly crude and were constantly alerting physicians, resulting in alert fatigue. Today, virtual assistant technologies are improving. They know when to get the physician's attention, nudge providers to do the right thing, and can save providers' time. Virtual assistants are developing capabilities, similar to those of a human scribe, to understand the conversation between patients and providers and summarize it. These solutions streamline workflows and reduce physician burnout.
Q: How can people get smarter about technical topics like these?
VJ: Here are a few great options:
1. Introductory Accenture article "What is Artificial Intelligence, Really?"
2. Take an online course like "AI for Everyone" or if one is interested in a deeper dive — "Deep Learning Specialization" both offered by Stanford University on Coursera.org. To understand solutions related to healthcare, consider the course, "AI in Healthcare."