When Technology Becomes Obligation: AI and Duty

Saint Thomas Aquinas taught that ignorance of what one is bound to know, when it springs from negligence, is itself blameworthy.¹ The question belongs to another century, yet it carries a modern resonance. We live in a moment when knowledge arrives before illness announces itself, when patterns inside the body are revealed long before a human eye could notice them. Artificial intelligence has entered healthcare not as a distant promise but as a daily presence. It is being applied in radiology and pathology, tested in cardiology and dermatology, and extended into ordinary life through predictive analytics and continuous wearables.² These tools offer earlier detection and a kind of ever-present watchfulness that previous generations of medicine could not imagine.

Their rise invites a renewed form of Aquinas’s question. When reason is amplified by technology, and when that technology uncovers danger in its earliest traces, at what point does the ability to know become the responsibility to act? When does a tool that sharpens our sight turn into a duty that law can recognize?

This paper argues that artificial intelligence in early diagnosis and preventive medicine will expand the legal duties of healthcare providers and reshape the expectations placed upon patients. For providers, the central question is how technology may change the standard of care within an existing duty to a patient; for patients, it is how their conduct may affect recovery. Washington’s medical-malpractice decisions illustrate how an effective precaution may be required even when professional custom does not demand it, although expert testimony ordinarily remains necessary to establish reasonable care.³ When prevention becomes simpler, the law quietly raises its expectations.

The idea that advancing technology expands legal duty is far older than artificial intelligence. The common law recognized negligent medical treatment as an actionable wrong long before modern tort doctrine took shape; Blackstone described injury caused by a physician’s “neglect or unskilful management” as mala praxis, a breach of the trust the patient had placed in the practitioner.⁴ Duty has never been a static command. It has always been responsive to the practical abilities of ordinary life. When the means of avoiding harm become clearer or easier, the law grows correspondingly less tolerant of conduct that once fell within the boundaries of acceptable risk.

An early antecedent appears in the Case of Thorns. Choke J. reasoned that a defendant who cut thorns that fell onto another’s land could not rely on involuntariness alone: he needed to show that he could not have acted otherwise or had done all he could to prevent the fall. The discussion predates modern negligence doctrine, but it shows an early concern with the steps a person could take to avoid harming a neighbor.⁵

An eighteenth-century formulation appears in Judge Buller’s writings, which state that “every man ought to take reasonable care that he does not injure his neighbour.”⁶ The maxim stated a general duty of care rather than any rule about particular tools. Buller’s framing became part of the architecture of modern negligence doctrine: a duty of reasonable care toward one’s neighbor.

Medical disputes had raised related questions centuries earlier.

In the 1374 discussion collected under Stratton v. Swanlond, the text distinguishes negligent medical treatment from a cure that failed despite the surgeon’s diligence. The writ was ultimately abated because it omitted where the undertaking occurred. The discussion illustrates an early concern with the care a practitioner owed, although the case did not establish liability.⁷

By the nineteenth century, courts were defining a practitioner’s undertaking as a promise to bring the ordinary learning and skill of the profession, and treating the absence of that skill, no less than carelessness in applying it, as a ground of liability. That ordinary skill was measured against the learning and methods of the era. Those courts did not say that expectations rise as instruments improve; that inference is mine, and it is the argument this paper takes up.⁸ This historical pattern articulates that reasonable care is not a rigid formula. It is an evolving measure that grows with human capability. The more we can foresee, the greater the responsibility to prevent. The more precise our tools, the sharper the line between acceptable risk and preventable harm. It is a change in the field of vision. It transforms what people can know, and therefore what they ought to avoid.

These cases suggest a quieter principle, though it is a synthesis rather than a rule any court announced. Where a precaution is inexpensive, low-risk, and reliably effective against grave harm, a court may require it even though the calling has not adopted it. Under Washington medical-malpractice law, Helling v. Carey and Harris v. Groth illustrate that professional custom is evidence of reasonable care rather than its conclusive measure; expert testimony is ordinarily required to establish what reasonable prudence demands.⁹

Artificial intelligence enters this lineage naturally. Its power to detect patterns, predict risk, and reveal emerging danger is simply the newest expansion in human capability. The law has adapted to such expansions before, and it is being asked to do so again in the age of digital diagnosis. AI is the next chapter in a very old story of how law responds to the growth of human understanding. As knowledge arrives through the quiet grace that accompanies discovery, the excuses for preventable harm begin to diminish.

The evolution of neonatal medicine offers one of the clearest illustrations of how expanding capability reshapes the boundaries of duty. Survival at the earliest gestational ages was once vanishingly rare. In Bell and colleagues’ nineteen-center United States cohort from 2013–2018, survival to hospital discharge—or to one year for infants still hospitalized—reached 10.9 percent at twenty-two weeks, 49.4 percent at twenty-three weeks, and 69.9 percent at twenty-four weeks; among infants actively treated at birth, the corresponding figures were 30.0, 55.8, and 71.4 percent.¹⁰ This progress is reflected in the newest Guinness World Records recognition, which now lists Nash Keen, born in 2024 at exactly twenty-one weeks’ gestation, as the most premature infant ever to survive.¹¹ His case is an extraordinary individual result rather than a population-level threshold, but it shows how far the boundary can move. A central barrier at these gestations remains lung immaturity. Infants born this early are still in the late canalicular or early saccular stage of lung development, when the gas-exchange structures and the surfactant system remain immature, and the oxygen and respiratory support they require can itself contribute to chronic lung disease. Much of modern neonatology has been devoted to finding ways around this biological limit.¹²

In the earlier 1993–2012 cohort studied by Stoll and colleagues, antenatal corticosteroid use increased and less-invasive respiratory support became more common. Survival improved most markedly at twenty-three and twenty-four weeks, but the observational registry could not establish which changes in care caused that improvement; survival without major morbidity did not improve at twenty-two through twenty-four weeks. These trends have made active treatment at earlier gestations increasingly possible, but they do not establish a single population-level viability line.¹³

The most striking experimental development is the creation of external-womb systems. Research teams have supported fetal lambs in an artificial amniotic environment, often called a “Biobag,” for as long as twenty-eight days.¹⁴ These systems mimic the fluid-filled conditions of the uterus and allow continued development without relying on ventilators that can damage immature lungs. Parallel artificial-placenta efforts seek to transfer oxygen directly into the fetal bloodstream through a pumpless circuit. In one six-fetus ovine study, the circuit maintained systemic circulation for a mean of 250 hours and a maximum of 336 hours, though with reduced growth and unresolved complications.¹⁵

These systems seek to overcome the lung barrier by preserving fetal physiology rather than forcing immature lungs to breathe. The cited studies are preclinical: both used fetal lambs, and neither reports continued gestation of a human infant outside the uterus.¹⁶ Partridge and colleagues identified infants born at twenty-three to twenty-five weeks as the initial clinical population and expressly stated that their goal was not to extend viability below twenty-two to twenty-three weeks. Future refinements might affect that boundary, but the cited experiments do not establish that result.

As these capabilities grow, the meaning of viability itself begins to shift. When Roe v. Wade was decided, the Court placed viability at about twenty-eight weeks, though it noted survival might occur as early as twenty-four, reflecting the medical reality of 1973.¹⁷ Casey later observed that advances in neonatal care had moved viability somewhat earlier. Dobbs has since overruled both decisions, so they figure here only as historical instances of courts describing viability by reference to the medicine of their day.¹⁸ Public debates today over whether the decisive line rests at twenty-four weeks, twenty-two weeks, or somewhere earlier reflect not a change in metaphysics but a change in capability. Viability is not a fixed boundary. It is a measurement of human skill at a particular moment in history. An unborn child is therefore not medically unsavable in any permanent sense. The child is unsavable only relative to what current knowledge and tools permit. As neonatal medicine progresses, so does the horizon of what life can be sustained, and with it, the responsibilities tied to that possibility.

John Paul II wrote that those who care for the sick are called to be “guardians and servants of human life.”¹⁹ Neonatal technology illustrates this truth. Expanding capability can change what reasonable care requires, a possibility that artificial intelligence’s expansion of medical foresight brings into focus. Does the same technological trajectory that reshapes viability reshape the expectations placed upon those who practice?

The history of negligence law suggests that advances in prevention can change what reasonable care requires. Nowhere is that more evident than in medicine, where the standard of care is inseparable from the tools that help clinicians diagnose, foresee, and respond to danger. Artificial intelligence intensifies this connection. It can widen the range of what can reasonably be known and, where validated tools are accessible and appropriate to the clinical setting, may change what a reasonable provider must do.

Modern malpractice law already frames this expectation. Statutes such as section 766.102 of the Florida Statutes define the medical standard of care as the “level of care, skill, and treatment” “recognized as acceptable and appropriate by reasonably prudent” similar providers.²⁰ The phrase “reasonably prudent” has always carried an implicit dependence on the tools available. A cardiologist without access to an echocardiogram will be judged differently from one who ignores the results of a high-quality echocardiogram placed directly into her hands.²¹ Technology informs what prudence requires.

Artificial intelligence magnifies this relationship because it transforms the raw material of medical knowledge. In a proof-of-concept study, a deep neural network matched board-certified dermatologists on two skin-lesion classification tasks. A retrospective study used a neural network to identify a signal of paroxysmal atrial fibrillation in electrocardiograms recorded during normal sinus rhythm. AI tools can also structure clinical documentation, though a 2024 systematic review found error rates still too high for broad use of comprehensive documentation assistants. These are defined research capabilities, not general clinical superiority. But a sufficiently validated tool changes what a provider could have known and, therefore, what a court may later say the provider should have known.²² Courts have already begun to respond to this shift in more traditional technology cases.

In Sampson v. HeartWise Health Systems, the arrangement before the court assigned HeartWise the task of generating reports from screening data, while interpretation, diagnosis, and referral decisions remained with the treating physicians. The court affirmed summary judgment for HeartWise on the negligence claims. It reinstated and remanded those claims against Isaac Health and its physicians because they had not been properly presented for summary judgment, without reaching their merits. The decision illustrates how responsibility depends on the role each participant undertakes.²³ The tool may assist, but the judgment remains human.

Byrne v. Avery Center for Obstetrics & Gynecology is not a technology-reliance case. It recognized a common-law duty of confidentiality against unauthorized disclosure when a provider responded to a subpoena and held that HIPAA could inform the applicable standard. It supports only the broader analogy that a new regulatory setting does not necessarily displace independently existing provider duties; it does not establish a rule concerning software or artificial intelligence.²⁴

As AI becomes more reliable, courts may begin to treat it the way they once treated new diagnostic tests. Helling v. Carey provides one historical illustration. Although tonometry testing for glaucoma was not customary in patients under forty years of age, the test was simple, inexpensive, and could detect glaucoma early enough for treatment to arrest its progression. On those facts, the court held that custom did not excuse the failure to give this patient the test. The reasoning was unapologetically functional: the profession’s habits could not override the demands of reasonable care.²⁵

It is not difficult to imagine a parallel in AI-supported diagnostics. If a validated artificial-intelligence system consistently outperforms human interpretation in a specific setting, and if that system is accessible, affordable, and integrated into routine clinical practice, courts may view its omission much like the omission in Helling.²⁶ Custom would remain relevant, but it would not necessarily be decisive. A tool that reliably illuminates danger may in time become a tool that reasonable care requires.

Helling’s aftermath was contested. Washington enacted new medical-negligence standards after the decision, but its supreme court held that those statutes did not make professional custom dispositive; reasonable prudence remained the governing standard. Florida law contains a separate limit on liability for omitted supplemental diagnostic tests. Section 766.102(4) provides that failure to order, perform, or administer supplemental diagnostic tests is not actionable when a provider acts in good faith and with due regard for the prevailing professional standard of care. The provision therefore creates a narrow safe harbor without displacing the general professional-standard inquiry.²⁷

This shift will not happen overnight. Courts will continue to insist on human oversight and individualized judgment. If particular AI tools become validated, accessible, and accepted in defined clinical settings, the standard of care may rise accordingly. Providers may then be expected to use those tools competently, verify their outputs, and integrate them into clinical reasoning. In time, failure to consult an appropriate system may be evaluated much like failure to order an indicated test or attend to a clear warning.

Artificial intelligence need not displace provider duty; in defined circumstances, it may expand what reasonable care requires. It can reveal more, earlier, and with fewer barriers. Because duty can follow usable knowledge, the law may increasingly expect clinicians to employ validated tools while retaining human judgment.

The expansion of medical responsibility has never belonged to providers alone. Courts have long recognized that a patient’s conduct may affect recovery when, after receiving medical advice or follow-up instructions, the patient acts unreasonably and that conduct causally contributes to the injury. The doctrine does not impose a general obligation to adopt every available technology. Artificial intelligence may, however, change the information available to a patient and therefore the reasonableness of a response in a particular case.²⁸

Comparative negligence allows courts to allocate fault between provider and patient when both contribute to an injury.²⁹ In medical cases, comparative-fault principles may permit a factfinder to consider whether a patient unreasonably failed to follow medical advice, return for recommended care, or disclose relevant symptoms, where that conduct bears on causation. The law does not expect perfection. It expects sincerity and attention.³⁰

Dunnington v. Virginia Mason illustrates the distinction. Dunnington first received a choice between surgical excision and cryotherapy, not between biopsy and cryotherapy. At follow-up, after the lesion proved recalcitrant, the physician discussed excision and biopsy and favored excision; Dunnington again chose conservative care and did not return as instructed. After a December MRI, he deferred excisional biopsy, sought a second opinion, and again chose conservative treatment. A January punch biopsy revealed melanoma. He underwent excision, but the cancer recurred despite chemotherapy and radiation, and part of his leg ultimately was amputated. The court did not find him negligent; it held that disputed facts required the contributory-negligence defense to go to the factfinder.³¹

Witherell v. Weimer presented a similar pattern. The defense alleged that the patient refilled prescriptions without approval, ignored dosage instructions, failed to reduce medications when advised, undertook long automobile trips contrary to medical warnings, and delayed reporting leg pain for years. A jury attributed forty percent of her injury to her own conduct, and the Illinois Supreme Court found evidence supporting at least some of the asserted grounds and no reversible error in that allocation. The case stands for a straightforward proposition: a patient’s own unreasonable conduct can reduce recovery when it contributes to the harm.³²

These precedents support a narrower rule: when a patient receives clear medical advice or follow-up instructions, unreasonable noncompliance that causally worsens the injury may reduce recovery. They do not establish a general duty to adopt any inexpensive tool within easy reach. Extending that logic to AI monitoring is this paper’s prediction, not a rule those cases announced.

AI-enabled wearables create a possible extension of this doctrine. A deep neural network has passively identified atrial fibrillation from commercial-smartwatch data, and physiological deviations detectable by wearables have sometimes preceded reported symptoms of COVID-19. These studies demonstrate emerging detection capacity, not legally actionable notice. Neither establishes that patients currently receive clear, clinically validated warnings requiring action. A prescribed and validated device that delivers a specific, understandable warning could affect what a patient reasonably should do, but that is a conditional legal argument rather than a result established by these studies.³³

The legal implications follow naturally from the historical pattern. If a patient receives a clear alert that signals risk, ignoring that alert may resemble ignoring a doctor’s direct instruction. If a patient refuses to use a simple monitoring device that a provider recommends, and if that decision contributes to a preventable harm, courts may treat the refusal as contributory negligence. The analogy to the seatbelt defense is instructive. A seatbelt does not prevent every injury, but it prevents many. It is inexpensive, accessible, and effortless to use. In Spier v. Barker, New York allowed nonuse of an available seatbelt to bear on damages when the defendant proved that it increased the injuries. This injury-mitigation rule is distinct from allocating comparative fault for causing the accident.³⁴ AI-enabled monitoring may in time stand in a similar place, though any such analogy would require proof of availability, a clear recommendation or warning, unreasonable nonuse, and causal enhancement of the injury. It cannot prevent every illness, but it can reveal some risks sooner and with little effort.

Artificial intelligence does not automatically create a new patient duty. If a validated, affordable tool is prescribed and gives a clear warning, unreasonable nonuse or inaction that causally worsens harm may become relevant to comparative fault. Whether the law will recognize that consequence depends on evidence of availability, clarity, reasonableness, and causation. Responsibility may grow with usable information, but the doctrine remains grounded in the circumstances of the individual case.

Through December 2025, federal AI policy had undergone significant changes while states pursued their own approaches.³⁵ Florida hosts substantial AI research and medical-data infrastructure. OneFlorida+ provides a large, multi-institution clinical-research network.³⁶ These resources may influence future professional practice, but they do not by themselves establish statewide clinical adoption, medical supervision, or regulation.

At the University of Florida, the 2020 initiative combined a $50 million gift—$25 million from alumnus Chris Malachowsky and $25 million in NVIDIA hardware, software, training, and services—with an additional $20 million UF investment. It was anchored by HiPerGator and included curriculum and UF Health research. UF researchers also developed DeepSOFA, a retrospectively developed and externally validated model using longitudinal electronic-health-record data to predict in-hospital mortality. These facts demonstrate research and educational capacity, not statewide clinical deployment.³⁷

UF’s integration of AI into education and its research into patient-care applications may eventually influence professional expectations. The cited sources, however, do not establish that those tools have become prevailing practice across Florida. Whether a particular tool bears on reasonable care would require evidence about its validation, availability, acceptance within the relevant specialty, and use under comparable circumstances.

Florida’s telehealth statute requires a provider to practice consistently with the prevailing professional standard applicable to in-person care. The statute does not regulate AI or itself establish an AI standard of care. It does, however, support this paper’s inference that delivering care through technology does not lower the governing professional standard. Whether a particular AI tool becomes part of competent care would still depend on evidence of prevailing professional practice and the surrounding circumstances.³⁸

Public and private support for research and clinical infrastructure may help normalize technology-assisted care. That development could influence—but does not itself determine—future standards of care.

Florida’s UF-centered research infrastructure gives the state the potential to influence future norms, but the cited sources do not establish that surgical robots or AI-guided care are already common statewide or that AI is legally obligatory. If particular AI tools become broadly adopted and accepted within a specialty, Florida courts may consider that practice when defining reasonable care. The state’s present research and educational infrastructure is therefore relevant background, not proof that technology has already become an obligation.

Aquinas taught that reason carries responsibility, and the history of negligence law reflects the same idea. This paper has traced early expressions of reasonable care and later decisions about diagnostic precautions and professional responsibility. Artificial intelligence may enter that same lineage. For providers, validated diagnostic and predictive tools may alter what can reasonably be known about a patient’s condition and may eventually affect what reasonable care requires. For patients, validated monitoring devices and clear warnings may strengthen established expectations to follow medical advice and participate in their own safety. Florida’s expanding AI research and educational infrastructure offers relevant background for how such norms might develop. AI does not overturn traditional principles; it may extend them by enlarging human capability. As the means to foresee and prevent harm grow, so may the duty to use them, but any legal obligation will remain dependent on validation, accessibility, professional practice, reasonableness, and causation.

Alan Turing, whose work helped define early debates about machine intelligence, closed his 1950 essay with the observation that “We can only see a short distance ahead, but we can see plenty there that needs to be done.” Pope Francis likewise cautions that immense technological development must be matched by growth in human responsibility, values, and conscience. The future of medical duty will be shaped in that shared space, where technology, foresight, and responsibility meet.³⁹