I have written several articles about the adoption of technology in healthcare, including Electronic Medical Records (EMR), data analytics, point-of-care decision tools, Artificial Intelligence (AI) and Machine Learning (ML). Each and every technological innovation has the potential to impact healthcare. Why? Because there are so many areas to improve to enable clinicians to deliver quality-driven healthcare while being cost-conscious.
I first wrote about technology in healthcare for Forbes.com in a March 2019 article. Blockchain Technology May (Eventually) Improve Healthcare: Just Don't Get Your Hopes UpAfter 35 years working in healthcare, there's one thing I've noticed: the next best technology will “solve everything” (and still…) This is my third article on AI, so it feels a bit timely.
Starting point: AI is not an intangible object, a sentient being, or a master decision maker somewhere just below God. Rather, it is a computer algorithm (programming) built by humans and deployed to enable computers to “learn” and create advanced outputs based on that “learning”. However, AI can, will, and should play an expanded role in healthcare and healthcare delivery (optimizing patient encounters, faster and better data utilization/analysis, managing revenue cycle, reducing prior authorizations, etc.). However, all these decision-based tools require initial inputs and human intervention as a baseline and core code. While this may produce great results in the future, we need to be aware of the components in coding AI models, especially when it comes to care delivery.
An interesting recent lesson on hubris is the debacle surrounding the launch of an AI product (not specifically for healthcare) by a major international tech platform. After an inauspicious launch and rigorous (one might even say “brutal”) social media “vetting” and trials, the product was found to have built-in racial bias and historical factual flaws that inevitably led to provably erroneous output. The program was summarily halted with some flowery language that further product development was required. The episode was comical and expected, but also frightening as we look to a future where we rely even more heavily on AI and ML to help compute and understand vast amounts of data. One hopes that the underlying programming is mediocre, free of bias of any kind, so as not to taint the final product/output. But this recent big tech “product” is a classic example of how AI can go horribly wrong. Racial bias doesn't just “happen”. As mentioned before, AI is not a vat where inputs are loaded and the computer gets super smart and provides a high-quality output. No, AI, as I said before, is like a pyramid. To structurally stand the test of time, the foundations must be well-built and provide a solid base for the remaining components. A poorly-built foundation will lead to detrimental outcomes (such as the collapse of the structure). Arguably, the same is true for AI. Ultimately, the value of the output is determined by the underlying structure (the bottom of the pyramid). Poor foundations will result in a bad product. Poor inputs (coding/configuration) will produce questionable outputs that are inaccurate, low quality, and unreliable.
AI is growing, and that is a good thing, a good thing. We should embrace these technological advancements to empower us and help us. This growth is evidenced by Nvidia's recent phenomenal, and for some, insane, earnings reports and growth (just ask any member of Congress how much Nvidia is adding to their portfolios. Let's just say that many of these people are better stock pickers than Wall Street analysts. But I digress). Healthcare delivery can certainly benefit from the adoption of AI models. There are cost savings and potential care enhancements that can be realized based on proper adoption, and that's a good thing.
Someone asked me the other day, “Are you doing AI?” The answer is, of course, “How do you define AI?” So, I have a brief understanding of AI, I understand the fundamentals, Large Language Models (LLMs), parts of ML, and I've coded a little. I know that it has a role to play in bridging the gap in care delivery between data scientists, their output, and clinicians/operators. I know that AI can and will one day bring value to healthcare. But I also know that there are very real landmines defined and built by coders/programmers that can incorporate bias. This is dangerous. As I said before, garbage in, garbage out.
As people jump on or run away from AI and its applications in medicine, I want to warn you not to hate Frankenstein. Dr. Frankenstein. In other words, the output is not the monster; the monster is the creator.
