MD Anderson’s chief innovation officer, Dan Shoenthal, says hospitals aren’t uniquely "behind" on AI. He urges a balanced approach that pairs strong governance with real-world experiments, prioritizes workflow changes before technology selection, and protects patient data (no direct data transfers for model training). Early pilots—ambient recording and patient-facing AI agents—show improved engagement, but vendors must better demonstrate long-term value and collaborate closely with providers.
MD Anderson Innovation Chief: Hospitals Aren’t 'Behind' on AI — Focus On Governance, Workflow And Patient Trust
Dan Shoenthal is the chief innovation officer at MD Anderson Cancer Center, based in Houston.
Dan Shoenthal, chief innovation officer at MD Anderson Cancer Center in Houston, told Newsweek in Chicago on Sept. 15 that health systems are not uniquely "behind" on artificial intelligence. Instead, he said, providers face a shared challenge: translating fast-moving AI capability into safe, useful changes for clinicians, staff and patients.
From Hype To Real-World Impact
Shoenthal noted that AI has dominated conversations for two to three years, but measurable impact has lagged. Health systems are accustomed to stable technologies that last five to 10 years; AI evolves far faster, requiring a new maturity in how institutions adopt and manage change. Clinicians and nurses who have practiced the same way for years must learn to trust automated tools and altered workflows — a process that takes time and careful collaboration between vendors and providers.
What Vendors Often Miss
Vendors frequently underestimate how long it takes to realize value from AI. Using the surge in ambient-recording vendors as an example, Shoenthal described a crowded, rapidly changing market where early claims focus on staff-time reduction. In many cases, time saved is reallocated to new tasks rather than straight cost cuts. Vendors need to work with providers to map the longer, subtler story of value — including how freed time is redeployed to improve care.
Balancing Governance With Experimentation
MD Anderson pursues a two-pronged approach: strong governance plus permission to experiment. Governance brings cross-functional stakeholders together to prioritize high-impact projects, while experimentation helps teams understand how roles and workflows actually change when tools are deployed. Shoenthal warned against seeking perfection before piloting: real understanding often comes only from hands-on trials.
How Their AI Governance Works
The center uses a layered governance model. A cross-functional weekly workgroup (technology, cybersecurity, clinical leaders, ethics, legal, contracting and compliance) performs risk analysis and reviews requests. Upstream, business teams collaborate with technology teams to align projects with strategy; requests pass through that upstream review before reaching the AI governance group. Shoenthal emphasized that the structure is evolving but exists to protect patients while enabling impact.
Clinician Attitudes And Change Management
Clinician reactions vary. Some clinicians have embraced ambient-recording tools enthusiastically; others prefer different capabilities. Across oncology specialties, providers share a common drive: they want institutions to move faster — though they also weigh cost trade-offs when budgets are discussed.
Oncology’s Patient Engagement And Consent Issues
Oncology patients are highly engaged: MD Anderson reports much higher MyChart usage than typical ambulatory care. Patients often arrive with research or AI-generated summaries derived from their records, which can lead to more informed but sometimes partial or inaccurate conversations. Shoenthal framed this as an opportunity for richer dialogue, while noting the need for transparency and additional consenting when AI tools are part of care.
Protecting Patient Data
MD Anderson prohibits direct patient data from being transferred to vendors for model training. Protecting data is considered part of patient safety; the institution enforces strict standards and communicates tailored messaging depending on the use case. Patients may opt out of AI-enabled tools, and workflows must adapt when that happens.
AI Agents As Part Of The Workforce
The center is piloting patient-facing AI features in Epic and evaluating partnerships such as Hippocratic AI. Shoenthal stressed that AI agents should be managed operationally like staff: they require monitoring, spot-checking and performance oversight because they can "hallucinate" just as humans make mistakes. Agents are viewed as tools to increase patient touchpoints, not as replacements for clinicians.
Technology That Restores Human Connection
Shoenthal is most excited about technologies that free clinicians from screen-bound tasks and restore face-to-face engagement. He sees automation — documentation-on-the-fly, mobile access to clinical data and robotic supply delivery — as ways to reduce tethering to machines and let clinicians focus on patients. Early ambient-technology pilots have correlated with higher patient satisfaction and more engaged encounters.
Vendor Strategy: Workflow First
Given market fluidity, MD Anderson evaluates vendors by adaptability and alignment to workflows rather than betting on a single winner. Epic remains a major platform, but the landscape includes OpenAI, Anthropic, Google and many niche vendors. Shoenthal recommends designing adapted workflows first, then choosing the technology that best fits that vision and shared vendor-provider goals.
What’s Missing From The Conversation
"A lot of providers feel like they're behind. We're all in the same place," Shoenthal said, urging more sharing of successes and failures across institutions.
Shoenthal also called for more focus on patient expectations: too often discussions center on provider perspectives rather than on what patients want and worry about as AI becomes part of care.
Editor's Note: Responses have been lightly edited for length and clarity.
Help us improve.
