In 2018, Venture Beat posted this on what was then called Twitter:
AI is like teenage sex: “Everyone talks about it, nobody really knows how to do it, everyone thinks everyone else is doing it, so everyone claims they are doing it.”1
As healthcare lags the rest of the economy in adopting innovation, what was true in 2018 about artificial intelligence (AI) for the non-healthcare economy is true in 2024 for most health economy stakeholders. Next week, it is certain that many of the thousands of meetings held at and around the 42nd Annual J.P. Morgan Healthcare Conference will include enthusiastic discussions of AI in healthcare.
Unfortunately, that enthusiasm is often misguided, focused on Star Trek-like clinical applications. One underappreciated challenge of clinical AI use cases is that most published research findings are false, as John P. A. Ioannidis, M.D., of Stanford demonstrated in 2005.2 Automating false clinical knowledge can quite literally be fatal.
Happily, if paradoxically, the entire health economy is replete with numerous repetitive, mindless and bureaucratic processes that offer abundant opportunities to deploy AI for its highest and best use: automation.
Reasoning by first principles about deploying AI first requires a clear definition of AI, and our Chief Data Officer offers some useful insights:
AI is the general ability of computers or systems to emulate human thoughts and behaviors, encompassing an extremely broad category of applications, such as self-driving systems, recommended playlists, online shopping carts, etc. What is considered AI changes constantly, and numerous things that were formerly considered to be AI no longer are.
Machine learning (ML) is a sub-field of AI. Instead of developing code to enable a machine to solve a problem, ML is an iterative process enabling a machine to solve the problem itself.
Consider, for example, the problem of identifying a cat in a photograph. Using a traditional software engineering approach to solve the problem requires a comprehensive definition of all the things that a cat is, as well as the things that a cat is not. Using ML, an algorithm learns to identify a cat by being exposed to thousands of positive examples, i.e., pictures with cats, and thousands of negative examples, i.e., pictures without cats. If you have ever used CAPTCHA or ReCAPTCHA to prove that you are not a bot, then you have helped train an algorithm to identify a bridge, or a stop sign, or a school bus.
Natural language processing (NLP) is another type of AI, which IBM defines this way:
NLP combines computational linguistics—rule-based modeling of human language—with statistical, machine learning, and deep learning models. Together, these technologies enable computers to process human language in the form of text or voice data and to ‘understand’ its full meaning, complete with the speaker or writer’s intent and sentiment.3
ChatGPT is currently the “cool kid” of AI, even though ChatGPT merely delivers “answers” (or, more accurately, a predicted string of words) from large language models (LLMs). LLMs are models developed from the deployment of NLP against massive – and undisclosed – data sets sourced primarily from the Internet. Caveat emptor.
In 2024, NLP is no longer sexy. Plumbing isn’t sexy either, but bringing water into the house via pipes is markedly more efficient than drawing it from the well and carrying buckets into the house. The BMJ, f/k/a the British Medical Journal, notes plumbing’s impact on clinical quality:
More than 11 300 readers of the BMJ chose the introduction of clean water and sewage disposal—“the sanitary revolution”—as the most important medical milestone since 1840, when the BMJ was first published. The work of the 19th century lawyer Edwin Chadwick, who pioneered the introduction of piped water to people's homes and sewers rinsed by water, attracted 15.8% of the votes, while antibiotics took 15%, and anaesthesia took 14%. The next two most popular were the introduction of vaccines, with 12%, and the discovery of the structure of DNA (9%).4
The innovation of indoor plumbing was the automation of transporting water into and out of the house.
In 2016, I had the opportunity to lead a team of data scientists and engineers who developed and implemented an extremely compelling NLP solution commissioned by a large health system. By deploying NLP algorithms to read dictated clinical reports and logistic regression models to highlight the “critical” information in those reports, the health system automated clinical workflows - not diagnoses - to identify thousands of patients with previously undiagnosed cancers. This automation enabled the health system to reduce the time to treatment for those previously undiagnosed patients and save tens of thousands of hours of nursing labor, resulting in reduced labor expense and increased revenue and market share.
Last November, I moderated a panel about AI for a gathering of revenue cycle management executives from more than a dozen of the most prestigious U.S. health systems, representing almost 30% of all U.S. hospital spending. When I asked the attendees to share the average wage for a patient access representative, the answers ranged from $17-$19/hour. When I asked about the work history of the typical patient access representative, the answers were McDonald’s and Walmart cashiers. When I asked how many of the attendees provided their patient access representatives with automated scripts to collect patient co-pays and deductibles, fewer than half of the executives raised their hand.
AI offers every health economy stakeholder the opportunity to automate – to apply a set of rules consistently and uniformly – the numerous repetitive, mindless and bureaucratic processes in healthcare. Scripting the interactions of people who are paid $19/hour to ask patients in their community to pay as much as a $12,000 deductible requires a minuscule amount of AI. In 2024, any health system that has yet to do something so simple is wasting time thinking about more advanced AI applications.
Keynote speakers at the J.P. Morgan Healthcare Conference and Health Evolution Summit don’t talk about AI innovations like scripting interactions in the patient access department, and, in any event, most attendees of innovation conferences could not find that department with a map. However, deploying AI to automate healthcare’s numerous mundane processes can generate millions of dollars in improved cash flow, reduced costs and employee efficiency.
Make this your New Year’s resolution: Automation is innovation.