ACP sets ethical guideposts for AI in medical practice

Artificial intelligence is rapidly becoming part of everyday medical practice, but its use should remain grounded in clinical integrity, professional responsibility and the patient-physician relationship. The American College of Physicians (ACP) has outlined three ethical guideposts intended to help physicians navigate the growing use of AI: relationality, self-governance and competence.

The position paper, published in Annals of Internal Medicine, comes as AI is increasingly being used for clinical decision support, documentation, education and other medical tasks. While these technologies may improve efficiency and quality, the ACP argues that physicians need practical ethical guidance as questions around privacy, disclosure and fairness remain unresolved.

Supporting patient-physician relationship

A central principle of the paper is that AI should support rather than weaken the relationship between physicians and patients. The ACP approaches medical AI as a form of augmented intelligence: technology that assists physicians and patients instead of replacing human clinical judgement.

The first guidepost, relationality, reflects the importance of the human relationship in medicine. AI may provide information, recommendations or assistance, but healthcare involves more than producing technically correct answers. Physicians must consider patients' individual circumstances, values and preferences when making decisions.

This becomes particularly relevant as AI systems become more closely integrated into clinical workflows. Greater automation should not result in less attention to communication, empathy and shared decision-making. According to the ACP, responsible implementation should ultimately strengthen patient-centred care rather than allow technology to become a barrier between physician and patient.

Physicians remain responsible

Self-governance forms the second ethical guidepost. As AI systems become capable of generating increasingly sophisticated clinical recommendations, physicians need to retain their independent judgement and ability to assess whether an AI-generated conclusion is appropriate for an individual patient.

The ACP therefore positions AI as an additional source of support rather than an autonomous decision-maker. Physicians should remain capable of reasoning independently instead of automatically accepting an algorithm's output. This principle is closely connected to professional accountability: introducing AI into a clinical process does not eliminate the physician's responsibility to provide appropriate care.

The position paper also addresses fairness. AI systems may perform differently across patient populations, depending partly on the data and assumptions used in their development. Responsible use therefore requires attention to whether technology contributes to equitable care rather than reinforcing or creating disparities.

Questions about transparency and disclosure remain particularly complex. The ACP notes that consensus has not yet been reached on issues including when and how patients should be informed about the role of AI in their care. The ethical framework is intended to help physicians navigate such questions while more detailed standards continue to evolve.

Changes in clinical competence

The third guidepost, competence, addresses another consequence of AI adoption. Physicians increasingly need sufficient understanding of AI tools to use them responsibly. Clinical competence in an AI-supported environment therefore involves not only medical expertise, but also the ability to critically evaluate technological outputs and recognise their limitations.

At the same time, the ACP warns against allowing dependence on AI to undermine physicians' own clinical capabilities. AI can augment decision-making and medical education, but physicians still need the knowledge and reasoning skills required to assess whether its recommendations make sense in a particular clinical situation.

This creates a balancing act for healthcare organisations and medical education. Physicians need opportunities to learn how AI works and how to interpret its output, while maintaining the clinical skills that allow them to recognise errors and make decisions independently.

Framework for physicians

The ACP's position paper does not attempt to resolve every ethical question surrounding medical AI. Instead, the three guideposts provide a professional framework for physicians working in an environment where technology is developing faster than consensus on its governance.

The organisation sees substantial potential for AI to improve healthcare efficiency, clinical decision-making and patient outcomes. Realising those benefits, however, requires implementation that preserves physician judgement, supports equitable care and keeps the patient-physician relationship at the centre of medicine.

As AI becomes more deeply embedded in clinical care, the challenge will therefore extend beyond determining whether an algorithm performs well. Healthcare organisations and physicians will also need to determine how these systems can be incorporated without compromising the ethical and professional principles on which medical practice is based.

Ethical use of AI

Responsible use of AI in healthcare is not only a technical challenge, but also an ethical responsibility, according to Vanda Vitorino de Almeida, Responsible AI Clinical Lead at Philips. Last year, a tech event in the Netherlands, she argued that AI should augment healthcare professionals rather than replace them, with human oversight remaining essential.

Responsible AI requires attention to privacy and security, transparency, explainability and fair representation of different patient groups. Bias is a particular concern. When AI is trained on limited or unrepresentative datasets, its recommendations may be less reliable for certain populations, potentially contributing to unequal care. Vitorino de Almeida therefore stresses the importance of critically examining data, consent, research methods and conclusions.

The Philips Future Health Index similarly identifies proven effectiveness, prevention of data bias and appropriate legal frameworks as important conditions for building trust in healthcare AI. According to Vitorino de Almeida, responsibility for ethical AI should be shared across organisations rather than assigned solely to technical teams.