Sketching AI Concepts with Capabilities and Examples: AI Innovation in the Intensive Care Unit

Area of Science:

• Artificial Intelligence
• Human-Computer Interaction
• Critical Care Medicine

Background:

• Innovation teams struggle to envision desirable and buildable artificial intelligence (AI) concepts.
• A disconnect exists between data science (unwanted innovations) and domain experts (unbuildable innovations).
• Effective ideation is a critical bottleneck in AI development.

Purpose of the Study:

• To explore effective approaches for AI concept ideation and problem formulation.
• To present a firsthand account of ideating AI concepts for critical care medicine.
• To identify methods for multidisciplinary teams to generate buildable and desirable AI use cases.

Main Methods:

• Conducted a series of design workshops with a multidisciplinary team (data scientists, clinicians, HCI researchers).
• Explored effective approaches to AI concept ideation and problem formulation.
• Documented the process, challenges, and effective practices and artifacts.

Main Results:

• Identified challenges in AI concept ideation across different expert groups.
• Developed and tested practices and artifacts to facilitate effective AI ideation.
• Demonstrated a structured approach to bridge the gap between technical feasibility and user desirability.

Conclusions:

• Effective AI ideation requires structured collaboration between diverse stakeholders.
• Human-Computer Interaction (HCI) can play a crucial role in reducing AI innovation failure rates.
• The presented methods offer a pathway for improved stakeholder engagement and AI development in critical care medicine.