This header image with an overly optimistic view of AI in 2025 and a random skeleton walking generated by AI is evidence that AI replacing everyone’s jobs is over blown 🙂
Jan Beger, Global Head of AI Advocacy GE HealthCare posed 10 questions on LinkedIn. Because they are such good questions (thanks Jan!) Instead of responding in a comment response, I wanted my subscribers to read my latest thoughts on AI.
1️⃣ Will we achieve aligned global standards for AI regulation in healthcare, or will conflicting policies create more challenges than solutions?
With an incoming Trump administration more likely to have a hands off approach and Europe going a more regulated route, I don’t expect alignment in global standards for AI regulation. Changes to regulation of healthcare AI in the USA may be more likely to happen because there is an existing regulatory regime (the FDA).
A lower hanging fruit around potential regulation would be the adoption of standards around evaluating the performance of AI when these systems are used in production. The requirement that incorporating metadata that tags data in a patient record as generated by a Practitioner, Patient, Caregiver or AI, so that the performance of AI can be evaluated on a continuous basis.
2️⃣ Are large foundation models the key to cracking the complexities of healthcare data integration, or are they too resource-intensive for practical application?
Foundation Models are frequently fine tuned on healthcare data to improve performance on tasks like healthcare data integration. While there is promise in this approach, the need for consistent responses to the same question will mean that hybrid AI approaches will remain important. Even if the semantic interoperability nut can be cracked, data governance, network and directory architecture and incentives will remain challenges for enabling scalable interoperability.
3️⃣ Generative AI is making waves in other industries, but will healthcare truly embrace it for tasks like clinical documentation, patient communication, or treatment planning by 2025?
Generative AI is already being used to document a patient’s conversation with their doctor. Shortages in doctors and nurses will continue to drive adoption of further use cases. If regulatory requirements regarding data standards adoption (e.g. FHIR) remain in effect, generative AI in conjunction with FHIR adoption will spur these additional use cases in areas like prior authorization and other documentation tasks, which can reduce clinician burdens, and staff burnout.
4️⃣ Can AI truly enhance diagnostics and decision-making to the point where we trust it with life-and-death decisions? Where do we draw the line?
This question deserves a book to answer it.
There are at least four different components that need to be answered:
- What are the lower hanging fruits, which can build trust and agreement from clinicians and the entire care team?
- As AI is evaluated for risk for factors that impact life and death, what are the measures like accuracy and bias for green lighting it, and are we comparing it against the same measurements of the status quo? Many organizations lack this data driven accountability in evaluating individual/organization performance even without AI being a part of their organization.
- How are healthcare organizations making decisions about technology, and what is the change management and training around AI adoption?
- While foundational level AI literacy is essential, we may need more fundamental retraining of clinical and social care team members so they can see AI as a trusted colleague, whose perspective is considered.
With respect to where we draw the line, while AI is perceived to be one thing, there are hundreds of task specific algorithms that have been in use for decades. What is new is that one particular type of AI, Generative Pre-Training Transformer (GPT) Large Language Models, are being shown to perform better on many tasks than many narrow AI algorithms—AI designed for one task. In healthcare in particular, we still need to draw the lines based on the performance of AI on each unique task, and “performance” needs to be clearly defined.
5️⃣ How will we tackle the challenge of bias in AI? Will 2025 mark progress toward equity, or will these systems widen existing gaps in healthcare?
Another book is needed to answer this in full.
For more, you can attend the HIMSS Health Equity Forum where I will be moderating on a panel that will be beginning to answer these questions and others on March 3, 2025. The panel is organized for clinical and social care staff that are not technical/knowledgeable for AI and are meant to equip you with practical questions for AI vendors.
We won’t see rapid progress on health equity in one year 2025–not because the right AI, built correctly, as part of a suite of products or tools, can’t be part of a Care Teams’ progress. Nor will AI widen gaps. The limiting factors will be:
a) Overly conservative adoption that won’t track capabilities.
b) And more importantly a lack of funding for cross sectoral investments to address the Social Determinants of Health.
c) Challenging collective action/Cross Sector Care Coordination challenges needed to address health equity.
Matt Bishop is CEO of Open City Labs, an award winning digital health company that develops software to streamline enrollment in government benefits, social services and clinical care, while unifying patient data in integrated Care Plans. Open City Labs was the first to implement in production the Gravity Project’s FHIR SDOH referral standard and is recognized as a Success Story by the Gravity Project. Open City Labs was the winner of the Administration for Community Living’s Social Care Referrals Challenge and Bonus Challenge, which tackled the need for referral interoperability across systems. He is a member of the HIMSS Social Determinants of Health Committee and HIMSS Health Equity Pre-conference active contributor to numerous ANSI accredited standards bodies, including Health Level Seven International, The Gravity Project, the co-author of the IHE International‘s 360x-SD (SDOH closed loop referral), National Directory for Healthcare Providers and numerous subcommittees within DirectTrust.