Rockin' HIT Sales Podcast
Standards That Scale: Why AI Measurement and Trustworthiness Matter in Healthcare
Ram D. Sriram, Chief of the Software and Systems Division, National Institute of Standards and Technology’s Information Technology Laboratory (ITL)
Disclaimer: The views expressed by Ram Sriram do not represent the official policy or position of NIST.
Watch or Listen
Choose your preferred platform:
Also available on all major podcast platforms.
Episode Summary
In this episode of Rockin’ HIT Sales, David Hacker sits down with Ram D. Sriram of NIST for a practical conversation on why trustworthy AI in healthcare requires much more than a strong model or compelling demo.
Ram explains why measurement matters, what “metrology for AI” actually means, and why standards, interoperability, and uncertainty quantification are becoming essential for any Health IT company trying to earn trust and scale in clinical or operational settings.
The discussion also explores how startups should think about existing standards, what the NIST AI Risk Management Framework looks like in healthcare terms, and why knowing when a model does not know can be just as important as accuracy itself.
If you build, sell, or invest in healthcare AI, this episode is a practical look at what real-world readiness requires beyond product claims.
Why This Matters for Health IT Companies
Healthcare AI companies cannot establish trust simply by demonstrating that a model produces accurate results. Providers also need to understand what the system is designed to do, how its performance is measured, what standards support its operation, how it works across different data sources, and how uncertainty is handled when the model encounters circumstances outside its training environment.
For Health IT companies, measurement and standards therefore become part of product and go-to-market readiness. Vendors should be prepared to define the purpose of the solution, identify the metrics that matter for the specific healthcare use case, demonstrate interoperability, and explain how the system recognizes and manages uncertainty before asking clinicians or patients to rely on its output. Ram repeatedly emphasizes context, purpose, measurement, existing standards, and trustworthiness throughout the discussion
What You’ll Hear in This Episode
- Why AI measurement and calibration matter in healthcare
- What “trustworthy AI” looks like beyond buzzwords
- How standards can accelerate adoption instead of slowing innovation
- Why interoperability remains foundational in multi-source healthcare environments
- How uncertainty quantification changes clinical risk and decision pathways
- What early-stage companies should understand before putting AI in front of clinicians or patients
Questions This Episode Answers
What does “metrology for AI” mean in healthcare?
Metrology is the science of measurement. Applied to AI, it means determining how AI systems should be measured, evaluated, and calibrated so that their performance can be understood and trusted in the environment where they are being used. Ram notes that the appropriate measurements depend on the purpose and context of the system.
Why isn’t accuracy alone enough to establish trustworthy healthcare AI?
Because healthcare AI operates within a specific use case, population, workflow, and data environment. Trustworthiness can also depend on precision, privacy, protection, interoperability, ground truth, and whether the system can identify uncertainty when it encounters data outside what it was trained to recognize.
What should an AI company measure before putting a solution into a clinical environment?
The metrics should reflect the actual healthcare use case. For a diagnostic system, for example, a provider may need to monitor false positives, false negatives, accuracy, and other performance measures rather than relying on a single headline metric. The NIST AI Risk Management Framework organizes this broader work around govern, map, measure, and manage.
Why should startups use existing healthcare standards instead of creating their own?
Existing standards can simplify integration across the many systems and data sources that make up healthcare. Ram’s advice to startups is to first understand what standards already exist, use them where possible, and focus carefully on the specific use case rather than unnecessarily reinventing foundational infrastructure.
How does interoperability affect healthcare AI readiness?
Healthcare information increasingly comes from multiple sources—EHRs, devices, wearables, monitoring systems, and other applications. Those sources need to exchange information accurately and consistently, making interoperability and shared standards foundational to solutions that are expected to operate across real healthcare environments.
What does uncertainty quantification mean for a clinical AI system?
It means understanding when the model may not have enough confidence to support the same action it would take under more certain conditions. Ram uses a diagnostic example to show that uncertainty can change the clinical pathway, including whether additional human review is needed before acting on an AI-generated result.
What are AI companies commonly underinvesting in?
Ram specifically points to standards as an area often overlooked during AI development. He also emphasizes the importance of starting with the purpose of the system—understanding what the company is actually trying to achieve before deciding how the technology should be built and measured.
What is one question every AI company should ask before putting a feature in front of clinicians or patients?
The company should ask whether it can convince the clinician or patient that the system is trustworthy. That requires more than simply asserting that the technology works; the user needs a reason to have confidence in the output.
3 Brief Takeaways
About the Guest
Ram D. Sriram is Chief of the Software and Systems Division in the Information Technology Laboratory at NIST. He has worked across multiple eras of AI development and brings a long-range perspective on knowledge-based systems, neural networks, standards, interoperability, and trustworthy AI.
Transcript
Prefer to read or download the conversation?