Introduction
At the outset, and before
examining AI and healthcare from a legal perspective, this research briefly
tackles the concept of AI and how it is incorporated into healthcare. The EU
Commission AI Act regulation proposal defines AI as software that is made through
certain techniques, which provides outputs like ‘content, predictions,
recommendations, or decisions’ based on goals determined by humans to impact a
certain environment.[1]
The American computer scientist John McCarthy defines artificial intelligence as
the science that focuses on making computer programs intelligent. Noting that
such intelligence means the computational capacity to reach results.[2]
Artificial intelligence has a family (sub-sets) consisting of a son, a
grandson, and a great-grandson, each branching off from the other. The son is
machine learning; its most notable feature is relying on datasets labeled by a
human expert for learning. The grandson is a neural network that simulates the
human brain by passing data in layers that it consists of under certain
conditions. The great-grandson is deep learning, which can learn from both
labeled and raw data, thus it does not require expert intervention.[3]
AI started to exist in
the medical field at the beginning of the 1970s. As AI was used to solve
healthcare problems, research related to such matters started to increase. This
led to the establishment of the Journal of Artificial Intelligence in Medicine
in 1980, a worldwide journal, and the American Association for Artificial
Intelligence, which included a subdivision, concerned with medical AI
applications. In 2017, medical AI became the most important AI application in
terms of aggregate equity funding.[4]
There are various
applications of AI in the medical field; this research will indicate some of
them. The first is image-diagnosing apps, as it was found in published studies
from Google, Stanford University, and countries that participated in the Cancer
Metastases in Lymph Nodes Challenge 2016 competition that AI was better and
more precise than human physicians were in this regard. Noting that studies
involved different types of diseases, such as diabetic retinopathy, skin
cancer, and metastases in lymph nodes.[5]
The second is surgical
robots. In this field, AI can make complicated decisions related to surgery and
its consequences that are beyond the capabilities of most surgeons.[6]
Noting that there are five levels of surgeon robots, differing in the amount of
automation and doctor intervention.[7]
The third is intelligent personal records, such as wearable health devices,
like watches or a band that enable users to monitor their health by providing
signs. Another example is natural language processing, which is used for
translating doctors’ natural speech and analyzing data in electronic medical records.[8]
This leads the researcher to believe that AI could play a significant role in
resolving future health crises, such as COVID-19, where human existence is
threatened and all activities are postponed.[9]
What sparked the
curiosity of this research is the question: What will be the fate of doctors
and hospitals in terms of medical liability in the presence of AI? Noting that
AI may behave independently in producing outputs, as complete control of it may
be impossible. However, AI can provide results that are more accurate than some
doctors, which makes its application and contributions to the medical field a
necessity. Therefore, how could doctors use such technology and stay legally
secure?
The importance of this
research lies, on the one hand, in explaining the application of law in medical
cases that involve AI. For example, the research explored the relevant EU,
Kuwaiti, and American legislation and explained how general medical rules, such
as those in Kuwaiti and American legislation, and special rules that relate to
AI users, such as those in the EU legislation, would apply. On the other hand,
what may further demonstrate the importance of the research is providing what
could possibly be considered solutions, such as suggesting a new standard of
care, excluding some general rules from application, and altering deficiencies
in the explored legislation as per the scope of this research.
This research adopts
foresight, analytical, and comparative approaches. It looks for legislation,
whether it is in the form of acts or cases, in Kuwaiti and American
legislation. It also investigates the relevant proposals that are produced by
the EU Commission to analyze and compare all the above to foresee possible
solutions to such an international issue. Noting that this research is divided
into two main headings, which are:
1. The
development of the standard of care,
2. Hospitals’
liability with the incorporation of AI.
1.
The Development of the Standard
of Care
Under this topic, the
research will discuss two main headings. The first is medical malpractice under
the US and Kuwaiti legislation. While the second will address physicians’
liability under the EU legislation.
1.1. Medical Malpractice under the US and
Kuwaiti Legislation
Four essential elements
need to be proven in a successful medical malpractice case. They are: (1) the
existence of the doctor’s obligation of duty of care towards the patient, (2)
the breach of such a duty by the doctor through deviating from the standard of
care, (3) the patient’s damages, and (4) the causation between such damages and
the doctor’s breach to his duty.[10]
Therefore, the research will analyze those elements, as well as how they apply
to medical AI. Similarly, the analysis will entail the application of the
learned intermediary doctrine on AI under this requirement.
1.1.1. Duty of Care and Breach of Duty
In American legislation,
the duty of care is related to the establishment of the doctor-patient
relationship. Noting that such a link does not have to be formal where no
contract is demanded to establish it. Moreover, in one case, an appeals court
ruled that the duty of care does not have to be done through physical contact.
For instance, if a doctor prescribes a certain medication for a patient over the phone, the
relationship is considered established.[11] Similarly, the Kuwaiti
Medical Law considers whoever practices medical activities like diagnosis or
treatment, whether in person, through third parties or by any other means as a
doctor.[12]
This leads the researcher to believe that Kuwaiti law implicitly admits that a
doctor’s duty starts once the doctor-patient relationship is established
through the phone such in the previous example, email, or any other means,
whereby considering that a person who performs a medical activity by any means
as a doctor, the law is suggesting that his duty has started.
In many states, for a
doctor to breach their owed duty of care to the patient, courts must find that
the doctor failed to practice certain elements, such as providing care and
skill of a reasonably prudent doctor. Courts investigate this matter by
comparing the defendant doctor’s conduct with the conduct of another doctor in
the same state; other states apply a stricter standard of care, where the
defendant doctor’s conduct is compared with the conduct of a specialist and not
a general practitioner. Noting that experts play a major role in determining
whether the defendant’s conduct is irregular.[13] Similarly, the adopted
standard of care in Kuwaiti Medical Law tends to be strict, requiring the
comparison of similar specialist conduct.[14] It is noteworthy to
indicate that the discussed standards of care are not detailed in a certain
list, but they are the best practices embraced by physicians through applying
their training, education, and protocols.[15]
However, when technology
is integrated into the health field, changing the standard of care follows,
such as changing the standard of care of anesthesiologists through establishing
‘new methods monitoring oxygen levels during surgery’, which resulted from
misusing anesthesia that caused a massive amount of deaths and injuries in the
1980s.[16]
Noting that AI as a technology is providing outstanding solutions in the
medical field, this led to the belief that it would even heal cancer. However,
AI may be mistaken. Therefore, the question of whether and how AI can change
the standard of care becomes essential.[17]
According to the last
paragraph, scholars’ opinions were divided into four, respecting the matter of
the standard of care, as the first opinion focused on the reasonableness of
doctors’ judgment, which encompasses their decisions of whether or not to
utilize AI. For example, one scholar perceives that the standard of care may
change to recognize tumors as radiologists, regardless of the instruments
utilized. In the same context, the South California Supreme Court in McCourt ex
rel. McCourt v. Abernathy emphasized that the matter of adopting a certain
approach is not enough to impose liability, as treatment may be a result of
varied approaches, so long as a comparable reasonable prudent doctor can
exercise them in similar circumstances. Therefore, the application of the
principle that resulted from this trend would compare the diagnosis, treatment,
and decision to use AI with that of a reasonable, prudent doctor, as the mere
decision of whether or not to utilize AI would not hold physicians liable.[18]
The second opinion holds
that the influence of technology governs the standard of care change. This
opinion believes that not using AI would probably hold doctors liable, given
that AI provides more data to doctors for diagnosis and treatment, which makes
prescribing medicine through its use much better than the doctor’s sole
decision. Therefore, using AI would become an obligation, just as in ‘using X-rays
or automated external defibrillators’. Noting that by adopting the use of AI as
the standard of care, the required reasonableness will change, where it will be
considered that it is ‘unreasonable, per se, not to use medical AI’. However,
this opinion believes that the matter of whether to follow the recommendation
of AI should be left to the court to determine its reasonableness. Noting that
AI might provide an unforeseen recommendation due to its black-box issue, which
is a lack of understanding of how the AI reached its results,[19] or what
is known to be called ‘hallucinations’, which is a condition in which AI
produces outputs that are disproportionate to user inputs.[20]
The third opinion
believes that AI might become the standard of care over time, where the
physician will be liable in case he does not follow the AI recommendation,
regardless of the accuracy of the AI output, whether it was right or wrong.
Therefore, the physician should follow the AI recommendation to shield himself
against liability.[21]
Given this, the
researcher believes that the first opinion overlooked the capabilities of AI. For
example, it was mentioned in the introduction of this research that AI provided
results better than human physicians in image diagnosis and can make surgical
decisions better than human doctors,[22]
which leads the researcher to believe that the necessity of adopting AI as an
approach becomes a fact. Therefore, this research disagrees with the first
opinion.
On the contrary, this
research believes that the third opinion strongly adopts the idea of AI
superiority, relying on that AI can ‘perform better than even the best physicians’,[23]
passing the fact that AI is not risk-free, as mentioned earlier that AI may
provide incomprehensible recommendations due to the black-box problem.
Therefore, the researcher disagrees with this opinion and believes that doctors
cannot follow AI recommendations without monitoring, assessment, or
intervention.
Accordingly, the
researcher agrees with the second opinion. Doctors are obligated to use AI;
however, the reasonableness of following or not following its recommendation is
a matter that is left to the court examination on a case-by-case basis.
Notwithstanding, supporters of the third opinion who see the idea of following
AI as the proper standard of care may argue that the trend, such as in the EU Commission
AI Act regulation proposal, which regulates AI in general and states varied
provisions to reach a trustworthy AI, is to impose legal requirements on
high-risk AI systems, like ‘data and data governance, documentation and
recording keeping, transparency and provision of information to users, human
oversight, robustness, accuracy, and security’.[24] This, in
return, would overcome the black-box problem, along with other possible
problems, which makes the idea of changing the medical standard of care to
follow AI as the best one. In contrast, this research believes that such
requirements aim to mitigate risks and not end them as stated in the EU AI Act
regulation proposal.[25]
Noting such requirements would help doctors in interpreting and assessing the
AI output, which most probably matches the idea of changing the standard of
care to using AI with no obligations to follow.
Moreover, supporters of
the third opinion may argue that if the standard of care follows AI, doctors
could still avoid liability in case they deviate from the AI recommendation by
applying the respectable minority doctrine that allows doctors not to use AI.[26]
The respectable minority doctrine is where the defendant doctor demonstrates
that although his conduct differs from another practitioner’s conduct, it is
still ‘accepted by a respectable minority of practitioners’, and this may lead
the jury to decide that the defendant doctor’s conduct, which belongs to the
minority standard is the correct standard to follow.[27]
Similarly, this argument may be presented in Kuwaiti courts, noting that the
Medical Law allows doctors to deviate from similar specialist conduct, provided
that the deviated conduct complies with the medical principles.[28]
However, the researcher
believes that when changing the standard of care to adopting AI, the doctor
would have no motive to deviate from an inappropriate AI recommendation, as
they are protected anyway by only applying it. In this way, making the correct
decision would be left to the doctor’s conscience. In addition, doctors may
find themselves defendants in any case, and here, justifying and proving that a
respected minority accepts their conduct would be more complicated than proving
that they adopted AI. The researcher also believes that this view is hasty in
adopting AI as the only standard of care, leaving out the opinion of some
scholars such as Dreyfus, Searle, and Penrose, who argue that AI is unable to
think like humans, which elaborates the necessity of human intelligence.[29]
In contrast, altering the standard of care to use AI without the obligation to
follow its recommendations would push doctors to intervene and make the right
decision, as their role still affects the outcome of whether or not there is a
liability; unlike adopting AI, which excludes the role of the doctor. Noting
that law, in general, is about governing persons’ actions rather than
eliminating them.
The fourth opinion was
found in an empirical study, where potential jurors composed of 2,000 persons
showed that following the traditional standard of care or AI advice, regardless
of deviating from the traditional standard of care and causing injury to the
patient, would shield doctors from liability, noting that potential patients tended
to see that adopting AI is more reasonable and important than applying the
traditional standard of care.[30]
However, the researcher
disagrees with what might turn out to be with potential jurors. Such an opinion
may leave doctors with the option of either leaving the excellent results of AI
behind or adopting AI unthinkingly, and both cases would result in no
liability. This research is curious to know if the patient group targeted for
the study was introduced to the risks of AI or the impact of AI on healthcare,
such as the controversy of whether AI would improve or reduce the
patient-doctor relationship. Noting that according to the report ‘Crossing the
Quality Chasm’, which is published by the American Institute of Medicine, such
a relationship is crucial to enhance the healthcare system in general, [31]
or whether the targeted patient group was introduced to the issue of AI
hallucinations, or the issue of wrong AI outputs that rest on biased data that
relates to race, gender, and socioeconomic condition.[32]
It is noteworthy to
mention that an opinion believes that with the emergence of fully autonomous AI
medical devices, their legal status of deserving legal personality would be
reconsidered, as the same opinion discussed applying medical malpractice elements
to the fully autonomous AI medical devices. In fact, it was stated that it
would be easier to prove elements such as the existence of a duty of care,
causation, and damage through looking into the patient's electronic medical
records.[33]
However, the researcher believes that the hypothesis of vague, wrong, and
biased decisions still stands with fully autonomous AI medical devices, as it
is similar to the previous opinion of following AI without any human
intervention. This leads the researcher to believe that excluding human doctors
and leaving the mission to fully autonomous AI medical devices is not
necessarily a development, where the ultimate goal is to provide the best care
for patients as this matter is realized through the efforts of both AI and human
doctors as explained previously.
Nevertheless, meanwhile,
under the current standard of care in Kuwaiti and American legislation, a
physician who relies on AI recommendations could still face liability if
his/her conduct deviates from the current traditional standard of care and
results in an injury.[34]
For example, the first
scenario would be that the traditional correct dosage for a person who is
diagnosed with ovarian cancer is ‘15 milligrams per kilogram every 3 weeks of a
particular chemotherapeutic agent’. Noting that it is the traditional standard
of care dosage, and the AI recommended the same dosage. If the doctor follows
such a recommendation, there would not be an injury, and he would not be
liable. In contrast, if the doctor rejects the recommendation, he/she will
cause harm to the patient, as he/she will be liable for deviating from the
traditional standard of care. However, let us imagine that such a dosage is
inaccurate, which means that the traditional standard of care and AI are both
wrong. If the doctor had adopted the AI recommendation and harmed the patient,
he would not be liable, as he had followed the traditional standard of care.
However, if the doctor decides to reject such a recommendation, he/she may also
not be liable due to the nonexistence of the harm despite deviating from the
traditional standard of care.[35]
In the second scenario,
the AI recommends a different dosage that does not comply with the standard of
care dosage, such as ‘75 milligrams per kilogram every 3 weeks of a chemotherapeutic
agent’. Noting that the AI recommendation is the correct dosage, unlike what is
perceived in the traditional standard of care, especially since personalizing
care is one of the medical AI’s aims. If the doctor follows such a
recommendation, he/she will not harm the patient. Therefore, he will not be
liable despite deviating from the traditional standard of care. In addition, if
the doctor rejects such a recommendation, he will still not be liable, despite
the harm done to the patient, because he adopted the traditional standard of
care. Notwithstanding, suppose this non-standard dosage provided by AI was
incorrect as the doctor followed it. In that case, he will cause harm to the
patient and will be considered liable due to not following the traditional
standard of care and causing harm. In contrast, if the doctor rejects the AI
recommendation, he/she will not harm the patient and will not be considered
liable.[36]
1.1.2. Causation
According to American
law, the plaintiff must prove two elements of causation. First, cause in fact,
which means that if it was not for the doctor’s negligence, the harm would not
exist. Second is proximate causation, which means that the harm is an expected
result of the doctor’s negligence.[37]
Noting that causation is proved through experts ‘or a similar state law
standard’.[38]
For example, in the
Bramlett v. Charter case, a psychiatrist moved a patient with anxiety to a less
stringent observation, which resulted in the latter’s suicide. The patient’s widow
proved that if it were not for the psychiatrist’s decision, her husband would
have been alive. She also proved that suicide is an expected result of not
suitably observing a patient with such an illness.[39]
As for the Kuwaiti
Medical Law, proving medical malpractice is assigned to the body of medical
liability, which is a governmental institution that encompasses experts who
determine whether the doctor’s action constitutes medical malpractice from a
technical dimension.[40]
Some believe that proving
causation will be more complicated. As attributing the liability of the fault
to either the doctor or AI becomes a difficult matter for the jury, expert
testimony will be needed.[41]
However, the researcher disagrees with the alleged complexity, where it may be
inappropriate to compare the human doctor and the medical AI in terms of
causing harm when the only qualified defendant in a medical malpractice case is
the human doctor, and the factors of such a case apply only to him. Since then,
the research argues that the philosophy behind its adoption of using AI as a
standard of care was to allow doctors to avoid AI errors. Therefore, for
example, causation may be proved by demonstrating that not using AI or adopting
its wrong output by the doctor was the reason for the injury, after proving
that plaintiffs may sue AI makers under product liability grounds.
1.1.3. Damages
In American law, Damages
are divided into (3) categories: ‘compensatory damages for economic loss,
non-economic damages for pain and suffering, and punitive damages’.[42]
Damages, according to
Kuwaiti law, are divided into 2 types: physical, and moral.[43] The
Kuwaiti law differs from American law in that it does not impose punitive
damages. Noting that the amount of punitive damages exceeds the value of harm,
where the goal of punitive damages is deterrence and not remedy.[44]
However, AI may open the
door to unusual damages; for example, AI-enabled electronic health records can
predict patients who may die soon. This was proven in a study that involved
cancer patients. Therefore, overlooking the end-of-life discussions that should
take place between doctors and such patients may lead to damage.[45]
1.1.4. Learned Intermediary Doctrine
Under normal
circumstances, manufacturers must warn consumers about the product’s possible
dangers when in use. Notwithstanding, in medical devices, which may include
some AIs, the manufacturer’s duty is to warn the doctor about such dangers and
not the patient, where the latter becomes a learned intermediary. [46]
Therefore, the duty of warning patients about the device dangers shifts to the
doctors.[47]
The doctrine is accepted
by plenty of American courts. Nonetheless, the doctrine has two exceptions from
application. The first is when the manufacturer markets the product directly to
the consumer. The second is when the doctor’s role is considered as passive
between the product and the patient.[48]
An example of the second
exception manifests in the AI machine vision systems that are designed to read
the scan and then present the diagnostic results to the doctor. In this case,
misdiagnosing may not be attributed to the doctor, as he would be excluded from
liability due to his/her passive role in such a process, and manufacturers
cannot benefit from the defense of the learned intermediary doctrine.[49]
In contrast, the
researcher believes that there will be no passive roles for doctors when using
medical AI as the EU Commission AI Act regulation proposal imposed that AI
products should include ‘human-machine interface tools’ that help doctors
oversee, monitor, and assess AI operations and outputs as they allow doctors to
intervene and stop the operation of AI when needed.[50] However,
the researcher is against applying such a doctrine in cases that include
medical AI. As doctors should not bear the liability of the complex AI dangers,
the law should protect them in such a matter and provide them with an
environment where they can focus on interacting with AI suitably.
As for the Kuwaiti law,
the researcher believes that the learned intermediary doctrine is unknown.
1.2. Physician’s Liability under EU
Proposal
The EU Commission’s
non-contractual civil liability rules under the Artificial Intelligence
Directive proposal (AI Liability Directive), which regulates civil liability if
AI is involved in causing damage, consider any legal person who uses AI as a
user,[51]
where it refers to the definition of the user in the AI Act regulation proposal
that was mentioned earlier in this research.[52] Therefore, the research
will examine the doctor’s liability according to that.
Under this proposal, the
causal link between the user’s fault and the damage caused by an AI output or
the abstention of AI to provide the output is presumed by the court through
three major elements combined. First, the user’s deviation from a duty of
care that is stated at a union or national level, where this is presumed by
either that the defendant did not preserve or disclose evidence related to the
suspected AI according to the court’s order or through the claimant’s
clarification.[53]
However, suppose the defendant is a user of a high-risk AI who is governed by
the AI Act obligations in Chapters 2 and 3. In that case, the plaintiff should
prove the defendant’s deviation from duty of care through the following
elements:
a.
Not using or monitoring the AI system
upon the given directions, or did not interfere to suspend the AI when
necessary ‘under [Article 29 of the AI Act]; or’
b.
‘[E]xposed the AI system to input
data’ that is irrelevant to the AI system’s purpose ‘pursuant to
[Article 29 (3) of the Act]’.[54]
Second, the impact of the
user’s fault ‘reasonably likely based on the circumstances of the case’ on the
AI output or the abstention of AI to provide the output. Third, claimant’s
clarification that the damage was caused by the impacted AI output or the AI’s
failure to provide the output.[55]
According to the above,
the duty of care that the doctor is obligated to have is the one that is
assigned to the AI user and not the medical standard of care. Since the latter
is excluded according to the rule of lex specialis.[56] Noting
that such a standard of care does not exist in the Kuwaiti and American
legislations.
The researcher believes
it is better to impose rules when dealing with AI, such as applying the adopted
standard of care for the AI user in the non-contractual civil liability rules
to the artificial intelligence proposal. However, this research argues that the
proposal should have noticed that AI would probably produce inappropriate
outputs despite applying all the requirements. This previously prompted the
researcher to adopt the medical standard of using AI, as the reasonableness of
following its recommendations is left to the court. Noting that the AI
Liability Act proposal strongly relies on the AI Act proposal, which lays down
the requirements for high-risk AI systems to reduce risks and not end them.
Therefore, the researcher thinks it may be suitable to encourage doctors to
adopt two standards of care: the medical standard of care that this research
adopts and the AI user standard of care.
2.
Hospitals’ Liability with
the Incorporation of AI
Under this topic, the
research will address two main headings. The first is vicarious liability, and
the second is direct liability of medical facilities.
2.1 Vicarious Liability
In American law, the
theory of vicarious liability means that one person is liable for another
person’s act.[57]
Under the doctrine of respondeat superior, the employer is held liable for the
employee’s negligent act. However, the act of the latter should fall within the
scope of employment, which means that the act was committed during the
permitted time and zone, the employee was partially or fully motivated to
perform their work, and the committed conduct relates to the reason for hiring.[58]
Noting that the Minnesota Supreme Court stated that the hospital’s control of
the employed physicians’ conduct is a condition to apply vicarious liability under
the doctrine of respondeat superior.[59]
Similarly, the Kuwaiti law allows holding the employer liable on behalf of the employee who caused
damage, so long as the latter’s act occurred during or because of his work.[60]
An example of a
hospital’s vicarious liability, that involves AI, is when a negligent doctor
wrongly interprets the output of a sepsis prediction AI system.[61]
However, if the doctor
was not an employee but an independent contractor, alternatively, the doctrine
of apparent authority becomes applicable, as it applies when the principal
demonstrates that a certain person has the power to act on his behalf, which
leads the third party to believe in such power reasonably. Therefore, a
hospital may be liable in a case where an independent contractor causes harm to
a patient through using AI.[62]
In contrast, Kuwaiti law differs from American law in terms of this point,
where vicarious liability applies to the principal regardless of whether the
other person is a contractor or an employee.[63]
It is worth noting that
there is an opinion arguing that the damage caused by AI itself may shield
hospitals from being vicariously liable. This is achieved when courts consider
AI systems to be completely independent or subject to the control of their
maker. In this case, there will be no room to hold hospitals vicariously liable
due to the lack of the principal’s control. Notwithstanding, it is unknown
whether the apparent authority theory would apply to this case, noting that its
application does not require the control factor.[64]
Given this, the
researcher believes that the past hypothesis considered AI an uncontrolled
employee who committed a wrongful act. Noting that to be considered as an
employee or an independent contractor, AI must first possess a legal
personality. The EU parliament discussed such a matter and concluded that there
is no need to grant legal personality to AI.[65] Therefore, this hypothesis
is unlikely to happen. However, what could prevent hospitals’ vicarious liability
is altering the medical standard of care to adopting AI, where physicians will
never act wrongly by adopting AI; thus, hospitals won’t be liable for that.
2.2 Direct Liability
Hospitals’ direct
liability means that the hospital administration becomes directly liable due to
its negligence regarding doctors’ efficiency and patient care. The concept of
direct liability obligates hospitals to have a direct duty of care in terms of
medical judgment,[66]
noting that such a concept has various categories that increase and decrease
depending on the state.[67]
This leads us to the first opinion which believes that hospitals are governed
by two main direct liability theories regarding their decisions towards AI. The
first is the negligent selection and retention, and the second is the negligent
supervision.[68]
In the first theory,
negligent selection and retention, the hospital must ensure the doctor’s
efficiency and eligibility before accepting him into its workforce, noting that
this duty should be repeated every two years. In this context, the Supreme
Court of Wisconsin stated that a claimant should demonstrate that the defendant
hospital did not practice the level of care that a median hospital usually
practices ensuring the doctor’s eligibility. Therefore, a claimant may sue a
hospital on the basis that it hired rather than purchased AI without reviewing
its competency, contrary to what the theory requires. However, courts might not
accept the idea of ‘anthropomorphizing AI’, which may exclude this theory’s
application. If the court decided to apply the theory, the problem of
determining reasonable care would follow, noting that hospitals are beginning to
adopt AI.[69]
The second theory,
negligent supervision, presumes that hospitals must monitor doctors’ decisions
at the same time he/she make them daily. However, it is unlikely that courts
would apply this theory to each AI output or to doctors’ decisions that are
made accordingly. It is more suitable to apply such a theory to cases that
entail gross negligence but not normal negligence, where this would help impose
the appropriate obligations on hospitals and would not prevent them from
adopting AI.[70]
The researcher believes
that the application of both theories essentially revolves around hospital decisions
that relate to employees or medical staff. It is unlikely that they will be
applied to hospitals for decisions related to the adoption of AI, given that AI
does not possess legal personality, as mentioned previously in this research.
However, the researcher
agrees with the second, third, and fourth following opinions. The second
opinion states that once using AI becomes the standard of care, healthcare
facilities may expose themselves to liability if AI is unavailable. Even with
this, some AI scholars believe that the extremely high cost of AI would
constitute an obstacle for hospitals to provide AI, and in this case, courts
might not impose such a duty on hospitals. Others argue that the high-cost
obstacle can be solved when using medical AI ‘through the cloud as a
decentralized service’.[71]
According to the
hospital’s duty to provide suitable equipment and facilities,[72] the third
opinion holds that medical facilities may be held liable when they do not
provide care and maintenance of AI.[73]
For example, it was claimed in one case that the hospital did not exercise the
required maintenance and care for the Mako total knee robot.[74]
In addition, the
researcher believes that a fourth opinion was found in the case of Payas v.
Adventist Health System/Sunbelt Inc, where the court found that not only the
unsuccessful maintenance of the surgeon robot but also the failure to operate
it and not teaching the doctors about the appropriate use of it may hold the
hospital liable.[75]
In the Kuwaiti Medical
Law, the legislator stipulated a special provision in terms of providing
healthcare through AI, stating that it is permissible to use AI as per the
requirements and controls issued by the health ministry.[76] The
researcher believes that such a provision may not only constitute an obstacle
to applying hospitals’ direct liability if they refrain from providing AI, but
it may also prevent adopting a new standard of care, like using AI. Since the
text is permissible, hospitals may or may not use AI. Moreover, although the
researcher could not access the relevant requirements and controls issued by
the ministry, it is important to include other hospital obligations, like
training doctors on the use of AI or exercising maintenance on it.
Conclusion
This research has
addressed one of the important and relatively new topics. AI has become an
indispensable technology due to its results, which have contributed to the
advancement of the medical field. However, AI is not risk-free, as shown in
this research, and this emphasizes the importance of the human physicians’ role, which has always been the backbone
of the medical field. Therefore, AI and human physicians are complementary to
each other, and for the best results, the interaction between them is needed.
However, doctors may use AI incorrectly, and this would expose them to legal
action; hence, applying suitable rules becomes an essential requirement given the complexity
of artificial intelligence. This prompted scholars to request changing the
medical standard of care by providing some theories such as (adopting AI, using
AI, and following either AI or the traditional standard of care), noting that
one opinion preserved the idea of applying the traditional standard of care,
and this is what the current Kuwaiti and American legislation agrees with,
where doctors must apply the traditional standard of care, which does not
include AI, to shield themselves from liability. As for Vicarious liability in
Kuwaiti and American legislations, it would most probably apply to hospitals
due to the doctors’ wrongful use of AI. However, wrongful AI outputs themselves
may not cause hospitals to be vicariously liable, noting that theories in
American legislation, such as respondeat superior or apparent authority, require
human physicians’ mistakes as a condition to apply, and that does not include
the wrongful results of AI because it is not an employee and does not possess
the legal personality. Similarly, theories in American legislation, such as
negligent selection and retention and negligent supervision, relate to hospital
decisions on employees; therefore, they may not apply to hospitals’ decisions on
whether or not to adopt AI. However, hospitals may be found directly liable
when not providing and maintaining AI and when not preparing doctors for the
use of AI. On the contrary, the EU has chosen a different path in determining
the liability of using AI, where they focused on the characteristics of AI,
which led them to adopt a special standard for its users. Despite all that has
been said, the researcher believes that to reach suitable rules considering
medical liability for the use of AI, it is important that physicians have two
standards of care in the EU, Kuwaiti, and American legislation. The first is
the medical standard that obligates the use of AI, and the second is the AI
user standard. It is also important for Kuwaiti and American legislators to
issue legislation similar to the model of the EU AI Act regulation proposal
that generally regulates AI, which helps provide safe AI systems, including
medical AI. In addition, they need to issue legislation similar to the EU AI
Liability directive proposal, which helps extract AI user liability, noting that
both legislations strongly relate. In addition, the EU AI liability directive
proposal needs to indicate the special nature of standards of care when
stipulating the AI user standard of care, such as the medical standard of care
using AI. Moreover, the researcher should not forget to mention that American
law needs to exclude the learned intermediary doctrine in AI products, which
may backfire on doctors and make them bear complex dangers that they did not
cause, as the Kuwaiti Medical Law needs to exclude the permissible text
regarding using AI and replace it with an obligatory one since keeping the
permissible text contradicts the necessity of AI existence in hospitals and the
idea of adopting its use as the standard of care, as it may shield hospitals
from liability in case of not providing AI.
Bibliography
I. Books:
Abdul Reda AR and Alnakkas J, The
concise in the General Theory of Obligations: The First Book Sources of
Obligation and Proof (4th ed., House of Books Institution, Kuwait,
2014-2015).
Abu Al-layl IA, Civil Liability and
Unjust Enrichment: A Study of the Non-Voluntary Sources of Obligation According
to the Kuwaiti Civil Law (2nd ed., House of Books, Kuwait, 1998).
Bohr A and Memarzadeh K (eds), ‘The Rise
of Artificial Intelligence in Health Care’ Artificial Intelligence in
Healthcare (Academic Press, 2020).
II. Articles:
Benneh GB, ‘AI and Medical Negligence’(2024) 1 Africa
Journal of Regulatory Affairs.
Dhaouadi M, ‘An Exploration into the Nature of the
Making of Human and Artificial Intelligence and the Qur’anic Persepective’
(1992) 9 American Journal of Islamic Social Sciences.
Dickinson J, ‘Direct Hospital Liability As A Legal
Path to Improve Safety?’ (2024) 27 Journal of Health Care and Policy 279.
Elloumi A, ‘Justice in a Time of Health Crisis – A
Comparative Study’ (2023) 12 International Review of Law.
Eldakak A and others, ‘Civil Liability for the Actions
of Autonomus AI in Healthcare: An Invitation to Further Contemplation’ (2024)
11(1) Humanities and Social Sciences Communications.
Ghanam MG, ‘Civil Financial Penalties towards the
Penetration of the Idea of Punishment in Civil Law’ (2022) 12(80) The Journal
of Economic and Legal Research 3.
Giuffrida I and Treece T, ‘Keeping AI Under
Observation: Anticipated Impacts on Physicians’ Standard of Care’ (2020) 22
Tulane Journal of Technology & Intellectual Property 111.
Griffin F, ‘Artificial Intelligence and Liability in
Health Care’ (2021) 31 Health Matrix: The Journal of Law-Medicine 65.
Harned Z, Lungren M P and Rajpurkar P, ‘Machine
Vision, Medical AI, and Malpractice’ [2019] Harvard Journal of Law &
Technology Digest 2.
Husgen J, ‘Product Liability Suits Involving Drug or
Device Manufactures and Physicians: The Learned Intermediary Doctrine and the
Physician’s Duty to Warn’ (2014) 111 The Journal of Missouri State Medical
Association 478.
Hodge SD Jr, ‘The Medical and Legal Implications of
Artificial Intelligence in Healthcare - An Area of Unsettled Law’ (2022) XXVIII
Richmond journal of law & technology 405.
Maliha G and others, ‘Artificial Intelligence and
Liability in Medicine: Balancing Safety and Innovation’ (2021) 99 The Milbank
Quarterly 629.
Mirnezami R and Ahmed A, ‘Surgery 3.0, Artificial
Intelligence and the Next-Generation Surgeon’ (2018) 105 British Journal of
Surgery 463.
Nagy M and Sisk B, ‘How Will Artificial Intelligence
Affect Patient-Clinician Relationship?’ (2020) 22 American Medical Association
journal of ethics 395.
Neyers J W, ‘A Theory of Vicarious Liability’ (2005)
43 Alberta Law Review 287.
Price WN II, Gerke S, and Cohen IG, ‘Potential
Liability for Physicians Using Artificial Intelligence’ (2019) 322 Journal of
the American Medical Association 1765.
–––. ‘How Much Can Potential Jurors Tell Us About
Liability for Medical Artificial Intelligence?’ (2021) 62 Journal of Nuclear
Medicine 15.
–––. ‘Liability for Use of Artificial Intelligence in
Medicine’ [2022] Law & Economics Working Papers 1.
Quest D and others. ‘Demystifying AI in Healthcare:
Historical Perspectives and Current Considerations’ (2021) 8 Physician
Leadership Journal 59.
See B, ‘Paging Doctor Robot: Medical Artificial
Intelligence, Tort Liability, and Why Personhood May Be the Answer’ (2021) 87
Brooklyn Law Review 417.
Shumway, Hayes HJ, ‘Medical Malpractice Liability in
Large Language Model Artificial Intelligence: Legal Review and Policy
Recommendation’ (2024) 124 Journal of Osteopathic Medicine.
Schweikart SJ, ‘Who Will Be Liable for Medical
Malpractice in the Future? How the Use of Artificial Intelligence in Medicine
Will Shape Medical Tort Law’ (2020) 22(2) The Minnesota Journal of Law, Science
& Technology 1.
III. Legislation
Decree-Law no 67 of 1980 Promulgating
Civil Law < https://2h.ae/vGrB>
accessed
21 September 21, 2024
Law no 70/2020 on the Practice of the Medical and
Paramedical Professions, the Rights of the Patient and Health Facilities < https://www.amr.gov.kw/ar/law70-2020.php>
accessed 21 September 21, 2024.
IV. Caselaw
Kuwait court of cassation judgment, session 6/11/2022,
challenge no 498/2016 civil <https://laalaws.com/KTash/kwtAhkamTameezBySub>
accessed 21 September 2024
V. Proposals:
VI. Websites:
Jamjoom AAB, Jamjoom AMA and Marcus HJ, ‘Exploring
Public Opinion About Liability and Responsibility in Surgical Robotics with the
Irobotsurgeon Survey’ (2020) 2 Nature Machine Intelligence <https://discovery.ucl.ac.uk/id/eprint/10096389/3/Marcus_1517_0_attach_6_12933_%20AJ%20edits.pdf>
accessed 5 July 2024.
McCarthy J, ‘What is Artificial Intelligence?’ (Stanford,
12 November 2007) <http://jmc.stanford.edu/articles/whatisai/whatisai.pdf>
accessed 8 July 2024.
Papakonstantinou V and Hert P, ‘Refusing to Award
Legal Personality to AI: Why the European Parliament got it Wrong’ (AI-Regulation,
26 November 2020) <https://ai-regulation.com/wp-content/uploads/2020/11/Refusing-to-award-legal-personality-to-AI-Why-the-European-Parliament-got-it-wrong.pdf>
accessed 9 July 2024.
‘Standard of Care’ (The Climate Change and Public
Health Law Site) <https://biotech.law.lsu.edu/books/lbb/x131.htm>
accessed 7 July 2024.
‘Tort Law: The Rules of Medical Malpractice’
(Lawshelf, 21 October 2020) <www.youtube.com/watch?v=4hvGIETLfkM&t=311s>
accessed 6 July 2024.
Villaronga F and others, ‘A Human in the Loop in
Surgery Automation’ (2021) 3 Nature Machine Intelligence <www.researchgate.net/profile/Eduard-Fosch-Villaronga/publication/351462735_A_human_in_the_loop_in_surgery_automation/links/609b65b8299bf1ad8d955086/A-human-in-the-loop-in-surgery-automation.pdf?_sg%5B0%5D=started_experiment_milestone&_sg%5B1%5D=started_experiment_milestone&origin=journalDetail>
accessed 5 July 2024.
‘What is Machine Learning’ (IBM) <www.ibm.com/topics/machine-learning>
accessed 4 July 2024.
International 
المجلة
الدولية
للقانون