ACP Position Statements on the Ethics of AI in Medical Practice Save

Know-it-now

ACP Position Statements

1. The patient–physician relationship is the core of health care. AI must be implemented clinically to support, not weaken, this relationship.
2. AI is affecting medical practice, research, and learning. Physicians and trainees must recommit to the competencies of caring and develop new competencies on the appropriate use of AI.
3. When physicians use AI to support medical decision making, that use should respect and promote — not replace — physician-independent clinical judgment and recommendations based on the patient's individual needs and values.
4. Physicians, in coordination with clinics, hospitals, and health systems, must seek to identify and mitigate the effects of algorithmic bias in the use of AI, including by addressing root social and structural causes.

Recently JAMA updated its guidance on artificial intelligence (AI) in medical publications. Now, the ACP has published a position paper regarding the use and ethical considerations for the use of AI in medical practice. 

AI use in daily life and medical practice is either pervasive or growing.  Prompted by the growth of ambient dictation (scribes), office chatbots, and AI-assisted diagnostics, the ACP's new position paper offers a formal ethical framework for practicing physicians built on three guideposts (relationality, self-governance, competence) rather than technology-specific rules.

The paper notes that 200 AI ethics guidelines and 17 consensus statements have already been published worldwide, yet practical, visit or patient centric guidance is lacking. This is particularly important for rheumatologists who regularly engage complex patients, diagnostic/therapeutic demands, and nuanced shared decision-making at the point of care. 

The Three Guideposts

1. Relationality: anchors AI use in beneficence, nonmaleficence, and respect for autonomy — the patient–physician relationship must remain the organizing principle, not an artifact AI displaces. 

2. Self-governance: preserves physician independent judgment and clinical integrity even as algorithmic outputs enter the decision stream. 

3. Competence: obligates physicians to apply expertise equitably, avoid bias, and critically resist deskilling.

Deskilling refers to the erosion of a physician's clinical skills (diagnostic reasoning, physical examination, pattern recognition) by increasingly relying on AI (an easily accessible tool) instead of honed, time-honored skills.  Clinical judgment is deminished and this ACP paper treats this as an explicit ethical issue, and not a competency or training concern. Deskilling could result in:  

• Patient harm (with MD atrophied skills) - forced to have shorter visits, will you cut corners by doing labs/US/MRI over skilled assessments?

• Erosion of the patient–physician relationship.  AI depersonalizes and does not build trust accrued from physician attention. (Don't outsource your patient or your skills)

• Loss of self-governance. Clinical care can be compromised when clinical judgements become algorithmic.

Practice Points to Consider

• Ambient documentation: Ambient scribes may reduce EHR burden during dense follow-up visits (e.g., reviewing DAS28 trends, medication toxicity monitoring) but carry hallucination and bias risk that requires physician verification. Surveys show ~60% of Americans are concerned AI will worsen the patient-physician relationship.
• Diagnostic augmentation: The paper cites mammography and skin-cancer detection as domains where AI improves accuracy. This is already happening in rheumatology where AI-assisted interpretation of referral notes, labs, and imaging (musculoskeletal ultrasound, capillaroscopy) is maturing.
• Deskilling risk: Framed explicitly as an ethical (not just competency) issue
• Bias and equity: AI trained on historical data risks reproducing known disparities.  Rheumatology relevant given documented age-related, racial and socioeconomic gaps in access to biologics and delays in diagnosis for many rheumatic diseases.
• Transparency without over-disclosure: ACP recommends erring toward disclosure of AI use (e.g., informing patients when ambient scribes are active) but does not require exhaustive technical disclosure — judgment should be individualized, as with any clinical communication.