Artificial intelligence systems may improve monitoring, coordination, and task performance in healthcare, but they do not assume moral responsibility or perform the relational and evaluative work that defines caregiving, according to a conceptual analysis published in BMC Medical Ethics. 

Prevailing approaches to artificial intelligence (AI) ethics—including principles, deontology, consequentialism, and virtue ethics—are insufficient to evaluate how AI alters caregiving because they focus primarily on duties, outcomes, or character rather than caregiving as a structured moral practice, according to Rachel Wangari Kimani, clinical associate professor at the Decker College of Nursing and Health Sciences at Binghamton University in New York.

“As AI becomes integrated into healthcare, we must look beyond technical accuracy and potential outcomes and evaluate how it impacts the relational, empathetic core of clinical practice: the ethics of care,” Kimani told AACE Endocrine AI. “Rather than focusing solely on abstract rules or data optimization, care ethics emphasizes that true healing relies on human connection, mutual interdependence, and responsiveness to the patient's vulnerability. As healthcare becomes more digitally advanced, we must ensure it does not become morally detached.” 

Kimani examined ethical theories and literature on healthcare AI and social robots used in nursing care systems for companionship, rehabilitation, mobility assistance, and telepresence. Examples include Paro, a robotic seal used in dementia care, and Pepper, a humanoid robot that interprets facial expressions and vocal cues.  

She then applied Tronto’s care ethics to compare human caregiving with AI-mediated care. Tronto’s framework assesses care through four phases: attentiveness, responsibility, competence, and responsiveness. 

The analysis concluded that AI does not so much replace caregiving as reshape it. According to Kimani, attentiveness becomes data-driven monitoring, responsibility shifts toward oversight, competence is increasingly defined by optimization, and responsiveness is reduced to adaptive feedback rather than genuine reciprocal engagement.

Kimani noted that AI-mediated attentiveness differs fundamentally from human attentiveness because it detects and processes signals without recognizing vulnerability as a moral claim. For example, robots may respond to touch, tone, or facial expressions, but their responses are generated from programmed stimuli rather than an understanding of a specific patient’s circumstances. The analysis further suggested that AI monitoring systems could gradually become the surrogate judge of need, shifting caregivers from perceiving needs directly to interpreting data and potentially narrowing what counts as need.  

In AI-enabled care, accountability may be distributed among clinicians, designers, institutions, and technology developers. According to Kimani, this diffusion of responsibility can create what has been described as a “moral crumple zone,” in which accountability remains focused on the nearest human.  

Regarding competence, the analysis distinguished operational performance from moral judgment. Although AI systems may excel in speed, accuracy, and reliability, they lack the contextual awareness and interpretive judgment needed to determine what constitutes good care for a specific patient in a particular circumstance. Kimani illustrated the distinction using automated medication systems. While such systems may dispense the correct drug and dose, they cannot determine whether a patient appears unusually confused, distressed, or otherwise different from the previous day—judgments that rely on observation, context, and clinical discernment. 

She similarly concluded that machines adapt behavior without experiencing vulnerability or being morally affected by patients’ experiences. As a result, AI systems can simulate empathy through behavioral adaptation though lack the reciprocal engagement that care ethics considers central to caregiving. 

 Kimani said her analysis should challenge clinicians to consider how AI tools might shift the provider-patient relationship by creating a "functional resemblance" to care without true moral engagement. “Algorithms can process data—like calculating an insulin dose—but cannot comprehend human suffering,” she said. “If clinicians offload their relational duties to machines, healthcare becomes a hollow performance of empathy. True caregiving requires a human being to remain morally present, listening, and accountable to the person behind the data.” 

Kimani reported having no conflicts of interest.