Generative AI — A Real Game-Changer in Healthcare
Generative Artificial Intelligence (AI) is technology’s hottest topic of 2023. Since the launch of ChatGPT in November 2022, and its achievement as the most rapidly adopted technology in history, it has rapidly gained traction among businesses, professionals, and consumers. There is a lot of “hype” about this technology, but it is a rare true game-changer and one of the most important advances in technology in decades. This is especially true for the use cases in healthcare. A recent McKinsey analysis, “The State of AI in 2023: Generative AI’s Breakout Year”, underscores the rapid change that “gen AI” is catalyzing across every industry. In the healthcare industry, there are several key data points that executives should keep an eye on. Even at this early stage, one-third of survey respondents said they are using gen AI in at least one business function, with 40% planning to increase investment in AI due to gen AI. C-Suite executives are also engaged. Nearly 25% are using gen AI tools, and it is on the board agenda for over 25% of companies. It is important to recognize, from a competitive standpoint, that the rich will get richer. Organizations that had already embedded other forms of AI in their operations are outpacing others in adoption, with 79% of respondents experimenting with gen AI and 22% regularly using it. The competitive moat that effective use of gen AI could create will begin to shake out winners and losers. I predict that industry winners and losers will emerge in 2–3 years, depending on how effectively an organization leverages AI in general (and related tools like Robotic Process Automation) and Gen AI in particular. There will be a small window of opportunity for laggards to leapfrog if they can accelerate the effective adoption of AI. The game will be over for those who are slow to adopt this breakthrough technology.
Generative AI is expected to have the most impact on industries that depend on knowledge and services from a business model and differentiation perspective. Healthcare fits this equation perfectly. And more importantly, healthcare has spent the last ten years digitizing the content of administration, operations, and care delivery (e.g., Electronic Health Records, radiology images, claims, etc.). The challenges with interoperability may be lessened, as gen AI can help users to “understand” and correlate concepts in natural language and translate between different systems based on enhancing semantic interoperability. This game-changing technology should be adopted quickly, because it will solve many intractable problems, where many other technologies and approaches historically have failed or created unintended negative consequences.
A Bain report titled “Generative AI Will Transform Health Care Sooner Than You Think” underscores the speed of adoption and innumerable use cases for the application of gen AI in healthcare. Gen AI algorithms can process vast amounts of medical data to create new, reorganized, or summarized content according to natural language specifications or prompts within seconds. This is functionally the Star Trek computer. What sets generative AI apart and helps it overcome some of the earlier challenges to AI implementation in healthcare is its need for less data and its adaptability to new and unfamiliar situations. Unlike many healthcare applications, such as EHRs, its usability, and utility are readily apparent, making adoption easier. It is not ready, yet, for clinical decision support, but reducing administrative burden is low-hanging fruit. For example, DeepScribe, which offers AI ambient scribing services, has been able to decrease the time providers spend on administrative tasks by three hours a day. One use case has the potential to affect the longstanding problem of clinician burnout.
The enormous potential of gen AI must be tempered by the risks. There are concerns regarding privacy, security, bias, and ethical considerations. However, the most significant concern in healthcare is the quality and accuracy of the output. The open-source model of gen AI was intentionally tuned to be more creative than accurate. As a result, the output can be amazing and unexpected. On the other hand, the tool has a propensity for “hallucinating,” which means it will make up responses to a prompt that sound very reasonable but are totally false. This capability is a great resource for brainstorming, but not for medical diagnosis and treatment. This is why the output should always be validated by an expert if used for any clinical purposes. This fault will be corrected over time as the gen AI capability is tuned more for medical uses and accuracy. For example, Google’s Med-PaLM 2 harnesses the power of Google’s LLMs, aligned to the medical domain, to answer medical questions more accurately and safely. Med-PaLM 2 was the first LLM to perform at an “expert” test-taker level performance on the MedQA dataset of US Medical Licensing Examination (USMLE)-style questions, reaching 85%+ accuracy. Google and Mayo Clinic have partnered to use gen AI to make it easier for doctors to get access to relevant medical notes, research papers, or clinical guidelines and also to help patients more easily find the information they need.
Many other technologies and policies have threatened to “transform” healthcare, including EHRs, Value-Based Care, and Digital Health. They are all now grist for the mill of gen AI. We are at the beginning of a true tech-enabled transformation in healthcare, and clinical care will never be the same.
