- AI adoption in public sector and healthcare carries unique responsibilities.
- Evaluation-driven development is key to building trustworthy AI.
- The cost of inaction on AI innovation is a growing threat.
- Partnership and transparency are crucial to overcoming fear and driving adoption.
In an era where artificial intelligence is rapidly transforming industries, its application in the public sector and healthcare presents both immense opportunities and significant challenges. Rebecca Nissan, a Delivery Solutions Architect specializing in healthcare and life sciences, highlighted the critical need for 'trustworthy and trusted' AI. This means not only building systems that are technically sound but also ensuring that end-users, from clinicians to citizens, have confidence in their capabilities. The session delved into practical frameworks and real-world perspectives on how to responsibly innovate with AI.
Nissan emphasized that unlike making casual recommendations, AI in the public sector often involves high-stakes decisions impacting people's futures, resource allocation, and patient diagnoses. The core philosophy for building trustworthy AI, she explained, is 'evaluation-driven development.' This iterative process integrates evaluation at every stage, from initial prototyping and 'eyeballing' results to using LLM judges for quality checks and generating synthetic data for comprehensive testing. Crucially, it also includes collecting continuous feedback from stakeholders through tools like review apps, ensuring that the AI evolves in response to real-world needs and concerns.
The discussion extended to the importance of providing AI agents with the right context. Using a healthcare example, Nissan demonstrated how an AI agent could leverage specific tools and data, even unstructured clinical notes, to provide detailed patient admission information. This capability moves beyond general intelligence to 'data intelligence,' where AI is equipped with the specific knowledge needed to be truly productive. The Databricks platform, with its integrated tools and model serving endpoints, facilitates this process, allowing for rapid deployment and continuous monitoring of AI performance in production.
A panel discussion featuring Charles Abraham from Gainwell Technologies and Dr. Tom Osborne from Microsoft Federal Division shifted the focus to the 'risk of not using AI.' Dr. Osborne painted a stark picture of the healthcare landscape: an aging population with complex needs, unsustainable costs, and a growing shortage of providers. He argued that the traditional methods of throwing more money or people at these problems are failing. AI offers a powerful alternative, providing clinical decision support, extracting new insights from vast datasets, and automating administrative burdens that contribute to clinician burnout. The consensus was clear: thoughtful, responsible AI adoption is not just an advantage, but a necessity to address these 'titanic forces colliding' in healthcare.
Overcoming the 'fear of AI' emerged as a central theme. Charles Abraham noted that while some states are eager to adopt AI for efficiency and budget tightening, others are hesitant due to concerns about data quality and the unknown. Dr. Osborne likened this fear to the 'shadows' in a scary movie – the unknown is often scarier than the reality. Both panelists stressed the importance of a 'crawl-walk-run' approach, starting with well-defined problems and bringing all stakeholders (IT, security, business, clinicians) into a collaborative partnership. This transparency and shared understanding are vital to mitigating risks, building trust, and ensuring that AI tools are not just powerful, but also appropriately implemented to drive better outcomes and improve citizen services.
“There is as you suggested significant risks of not using powerful tools at least not using powerful tools appropriately.”
- Rebecca Nissan, Sr. Delivery Solutions Engineer
