Systems Thinking for Agentic AI – Building Robust, Reliable, and Explainable Systems for Healthcare

Value of the Information & Strength of the Argument

The talk provides a compelling argument for applying systems thinking to agentic AI, moving beyond a purely technical focus to consider socio-technical dynamics. The speaker effectively uses analogies (squirrel vs. thermostat, evil corporation) and a concrete use case to illustrate abstract concepts. The argumentation is logical and builds a case for a structured, multi-agent approach with explicit modeling of feedback and delays. However, the talk is largely conceptual and lacks empirical evidence or case studies demonstrating the effectiveness of the proposed framework. The speaker acknowledges this is a ‘sounding board’ for ideas, which limits the strength of the argumentation.

Scientific Rigor, Source Quality, Title Accuracy

The speaker demonstrates scientific rigor by grounding his arguments in known limitations of LLMs and systems thinking principles. However, he does not cite specific sources or studies during the talk, and the only reference provided in the description is a link to the symposium page, which does not contain detailed references. The title accurately reflects the content, and the talk is well-structured. The speaker’s credentials and the context of an academic symposium lend credibility, but the lack of explicit citations and empirical validation reduces the overall rigor.

The title accurately reflects the content, which focuses on applying systems thinking to agentic AI in healthcare.

7/10

The speaker is a domain expert with a strong academic background, and the talk is well-structured with practical examples. However, it is a conference presentation without peer-reviewed sources or empirical validation, and the content is largely conceptual.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 02:00Introduction of the squirrel vs. thermostat analogy to define agentic behavior.
• 05:00Discussion on LLM limitations, including contextual fragility and the 'strawberry problem'.
• 10:00Proposal to decompose large models into smaller, specialized SLMs with tool use.
• 15:00Introduction of systems thinking principles: stocks, flows, feedback loops, and delays.
• 20:00Presentation of the heart failure home care app use case, defining environment, boundaries, and sensors.
• 25:00Explanation of the world model and the importance of modeling delays to avoid oscillations.
• 30:00Discussion on the orchestrator agent and the need for explicit causal loop diagrams.
• 35:00Emphasis on involving stakeholders and the importance of boundaries and leverage points.

• ADIA Lab Symposium — Event page for the symposium where this talk was presented.

• ADIA Lab Symposium — Event page for the symposium where this talk was presented.