Intelligent Inquiry Training System (IITS): Generative AI Reshapes Clinical Skills Training

This article introduces the Intelligent Inquiry Training System (IITS) based on large language models, which aims to address core pain points in traditional clinical skills training such as scarcity of standardized patients, delayed feedback, and limited scenarios. Through generative AI technology, it enables personalized, scalable, and immersive learning experiences, bringing innovative changes to medical education.

Clinical skills training faces three major challenges:

1. Resource Scarcity: The number of trained standardized patients is limited, making it difficult to support large-scale repeated practice;
2. Feedback Lag: Physical simulators lack dynamic interaction, and feedback from standardized patients is affected by their state and lacks real-time guidance;
3. Scenario Limitations: The cost of simulating rare cases and complex complications is high, and key scenarios are hard to reproduce.

IITS uses a large language model as its foundation and builds a three-layer architecture:

• Pre-class Preparation Layer: Generates contextual preview materials and guiding questions to establish an inquiry thinking framework;
• Core Training Layer: Simulates virtual patients based on the Wenxin Large Model, supports open-ended inquiry in natural language, and generates real-time responses consistent with medical logic;
• Post-class Consolidation Layer: Automatically generates multi-dimensional evaluation reports and improvement suggestions.

Multimodal interaction capabilities include: emotional simulation (anxiety, confusion, etc.), dynamic feedback (adjusting cooperation based on the quality of questions), and role diversity (virtual patients of different ages/disease types).

IITS integrates the Problem-Based Learning (PBL) concept:

• Case-Driven Path: Centered on clinical cases, it collects information through inquiry, proposes hypotheses, verifies inferences, and cultivates clinical reasoning ability;
• Iterative Construction: Supports repeated training, fine-tunes cases and feedback strategies based on student performance, enables personalized learning, and strengthens training on weak areas.

Comparative experiments show:

• Skill Improvement: The experimental group significantly outperformed the control group in inquiry completeness, information acquisition efficiency, and diagnostic accuracy, with stronger performance in handling complex cases;
• Learning Experience: Students recognized its accessibility (training anytime and anywhere), safety (no pressure to trial and error in a virtual environment), and immediacy (real-time feedback loop);
• Teacher Transformation: The system undertakes repeated guidance and basic assessment, allowing teachers to focus on personalized tutoring and complex case design.

Technical innovations:

• Medical Knowledge Alignment: Optimizes LLMs to ensure accurate symptom descriptions, reasonable disease evolution, and diverse communication scenarios (emergency/outpatient, etc.);
• Evaluation Algorithm: Built-in multi-dimensional automatic scoring mechanism, including process-based (inquiry systematicness), communication-based (empathy and attitude), and result-based (diagnostic accuracy) assessments.

Future Directions:

1. Specialization deepening (extending to surgery, psychiatry, etc.);
2. Cross-platform deployment (integrating into existing teaching systems);
3. Long-term tracking research (verifying the transfer of virtual training to actual clinical abilities).

Implications:

• Generative AI can expand scarce educational resources and improve fairness;
• Virtual environments allow safe trial and error, strengthening the learning loop;
• LLMs make personalized teaching a reality, adapting to students' levels and paces.