Before opening an AI tool, a simple question can be surprisingly useful: Should I use AI for this?
Benita Olivier and Michael Rowe's decision guide for doctoral researchers turns that pause into four practical questions. Is the task mechanical or intellectual? Is it a capability I am meant to develop? Could I defend every part of the output? Do I know enough to identify an error?
Those questions protect learning and individual accountability. In health professions education, however, they are only the beginning. An AI decision can also affect patients, learners, colleagues, institutions and communities. The newer Health CARE-AI framework helps us see that wider responsibility.
The first decision: should I use AI for this task?
The original guide makes an important distinction between mechanical and intellectual work. AI may be a reasonable aid for routine reformatting or reorganising information, provided the result is checked. It deserves greater caution when the task is intended to develop reasoning, synthesis, writing or professional judgement.
A polished output is not evidence that learning has occurred. If I am developing a capability, I should make a meaningful attempt before asking AI to review, challenge or refine it. If I cannot explain the result or recognise when it is wrong, I am not ready to rely on it.
Why individual judgement is not enough
Imagine that a physiotherapy student can defend an AI-assisted answer and identify obvious errors. That still does not tell us whether patient information was uploaded appropriately, whether consent was needed, whether the system disadvantages a particular group, or whether the university provided an approved and accessible tool.
Responsible AI therefore cannot be reduced to better student behaviour. Educators, clinical teams, institutions and technology providers also shape what responsible use is possible.
What Health CARE-AI adds
Sonnenberg and colleagues developed the Health CARE-AI framework through a three-phase modified Delphi consensus study involving 303 unique participants. The framework addresses AI professionalism across health professions education, research and clinical care.
Health CARE-AI stands for Contextual, Accountable, Responsible, and Equitable Artificial Intelligence. Importantly, it is not merely a four-letter checklist. The published framework contains ten principles organised across four interdependent domains: values, competence, accountability and structural equity.
- Responsible AI use is both an individual and collective duty.
- Users should act with honesty and integrity and build role-appropriate AI literacy.
- AI should complement, rather than replace, human judgement.
- Use must respect law, consent, privacy, intellectual property and ethical information governance.
- Bias and inequity should be actively addressed in outputs, design, access and governance.
- AI use should be proportionate to its benefit and attentive to social and environmental sustainability.
A two-stage decision for health professions education
I now see the two resources as consecutive stages rather than competing frameworks.
- Stage one — Should I use AI for this task? Protect the person's learning, authorship, subject knowledge and ability to defend the work.
- Stage two — If I use it, is the use professionally defensible? Examine the context, who is accountable, whether human judgement remains central, and how benefits and harms are distributed.
The first stage might permit AI use while the second stage rules it out. A tool can be educationally helpful yet unsuitable because it requires identifiable patient information. It can produce an accurate answer yet remain inequitable because some students cannot access it. It can save an educator time while creating an unexamined privacy or sustainability cost.
Four practical lenses
- Contextual: What is the purpose, setting and level of risk? Who could be affected, and which policies apply?
- Accountable: Who owns and verifies the decision? Could the use be disclosed and defended? What duties belong to the learner, educator, team and institution?
- Responsible: Does AI support rather than replace human judgement? Are consent, privacy, intellectual property, data governance and sustainability addressed?
- Equitable: Could the system reproduce bias, exclude particular people or distribute benefits and harms unevenly?
These four lenses are my practical bridge between the two resources; they should not be mistaken for the complete published framework.
What this changes in teaching
For curriculum design, AI professionalism should not sit in a single introductory lecture. It should return in classroom work, simulation, research and clinical learning. Students need opportunities to make and defend decisions about disclosure, verification, consent, bias and appropriate reliance.
Faculty development matters just as much. If lecturers are expected to design AI-supported assessment, approve tools and guide students, they need role-appropriate AI literacy, protected development time and clear institutional support. We cannot hold students to standards that staff and systems are not prepared to meet.
A framework, not proof of outcomes
Health CARE-AI offers a carefully developed consensus about professional expectations. It does not yet show that adopting the framework improves learning, professional practice or patient outcomes. Participants were also drawn predominantly from Canada, the United States and other English-speaking, high-resource settings, so application elsewhere requires contextual judgement.
That limitation does not make the framework unhelpful. It tells us how to use it: as a strong starting point for curriculum, faculty development and governance, accompanied by evaluation and adaptation rather than treated as a finished answer.
The question I would ask now
‘Should I use AI for this?’ remains an excellent pause before action. In health professions education, I would follow it with a second question: Can we use AI here in a way that is context-sensitive, collectively accountable, professionally responsible and structurally equitable?
That shift moves us from permission to professional judgement. It also moves responsibility beyond the student. Responsible AI depends on the learning activities educators design, the tools institutions provide, the safeguards teams maintain and the values the profession chooses to protect.
References
- Olivier, B., & Rowe, M. (2026). Should I use AI for this? A decision guide for doctoral researchers [Decision guide]. Still Yours; Springer Nature. Original guide.
- Sonnenberg, L. K., Wiljer, D., Mamdani, M., Do, V., Tang, B., Haroon, B., Jaffer, B., & Maniate, J. (2026). Principles for responsible AI in health professions education, research, and care: Health CARE-AI Framework Delphi consensus study. JMIR Medical Education, 12, e91626. https://doi.org/10.2196/91626.