Keeping the Human in the AI Loop with SPARC

As Artificial Intelligence (AI) tools move from novelty to integral parts of teaching, learning, and health care, medical and health professions educators face two critical challenges: How do we harness the power of these tools for education such as curricular design, item writing, content presentation, without compromising quality, rigor, or pedagogical integrity? How do we maintain human agency as we incorporate these tools? While AI may have strengths as a thought partner, its limitations make intentional human oversight a foundational component of teaching and learning with and about AI.  

AI SPARC: A Reflective Framework for Prompting AI When Used as an Educator's Thought Partner is part of the AAMCs collection of tools for educators. AI-SPARC provides a practical and reflective framework for keeping the human in the loop. The framework is a five-step cycle that guides educators through the process of using AI as a true thought partner, ensuring outputs are grounded in professional standards and evidence informed pedagogical practices. Based on Kolb’s experiential learning cycle, AI-SPARC provides explicit opportunities for reflection and human input on the process.  

Before prompting:  

S–Self-Reflect: Thinking about your thinking. What are your core learning goals? What is your teaching philosophy? What is your perspective about AI in teaching and learning? Answering questions like these inserts a human in the loop from the start.  

During prompting:

P–Prompt with a model: To prompt, we recommend using a structured method like TRACI to ensure clarity. TRACI is an acronym that stands for: T) Task: The general type of output you want ("Draft a rubric"), Role: The persona the AI should adopt ("Act as a program director"), Audience: Who the content is for ("For first-year residents"), Create: The desired format ("in a 5-point Likert scale table"), and Intent: The overall goal (e.g., "to assess communication skills").  

Adding a request for AI to “ask you questions before it starts” within your prompt is another strategy to improve the context provided to AI and to insert a human in the loop.  

A–Academic Requirements: Including official standards provides AI with the language of your discipline or specialty. Use academic standards like Accreditation Council for Graduate Medical Education (ACGME) milestones, specific competency statements from your discipline, or learning objectives from your program or course.  

R–Research on Pedagogy: Grounding AI with established educational theory like Miller’s Pyramid, Kolb’s Experiential Learning cycle, or Pangaro’s RIME model provides AI with the language of evidence informed teaching practices.    

Starting with self-reflection, then providing context with a prompt, and grounding AI with academic requirements and research on pedagogy can improve your prompt, resulting in more efficient and effective use of AI.  

After prompting:

C–Critique: The final and most crucial step is C–Critique. Do not accept the response at face value. The "After" phase enables you to critically compare the AI output back to your self reflection as well as the academic requirements and research on pedagogy. It is another point during which you can intentionally put the human back in the loop.  

Some questions to consider are:  

• Did the AI output reflect your teaching philosophy (S)?  
• Did it genuinely align with the accreditation standards (A)?  
• Did the response accurately reflect the pedagogical literature you chose (R)?  

The critique completes each cycle, allowing you to revise the output to meet your purposes or refine your prompt for the next iteration.  

An example of SPARC in action

Use AI as a thought partner to draft multiple choice questions. Start by reflecting on your goals for the learners in your course. What competencies should they develop, or what does it mean to trust them to do a professional activity? As you prompt to provide context, add specific descriptions of your course and learners. Ground AI with your course objectives or accreditation standards, as well as literature on good multiple choice questions. Instead of asking for a complete question, try asking AI to develop three to five clinical vignettes in the NBME style. Making sure your vignette is appropriate before asking for questions inserts the human in the loop. Revise the vignette and use that as part of your next SPARC cycle when drafting multiple lead-in questions, answer options, and the rationale for why options are correct or incorrect. Ground these prompts in clinical research and guidelines on item writing.  

Finally, SPARC can keep a human in the loop as a teaching and learning tool. Teach your learners to use SPARC and turn in their reflections, prompts, grounding resources, and critiques as an AI Audit Trail. This will provide opportunities for discussions about each component, expert guidance on its application in health care, and the importance of keeping a human in the loop on this journey to a future with AI.