AI in Healthcare

Gone are the days when the Turing test used to be the benchmark for assessing the
capabilities of an AI (Artificial Intelligence) algorithm as the technological development in
the field of AI seems to have ventured into the second half of the chess board in the last
decade. To wit, consider the curious case of Blake Lemoine, a Google engineer who recently
claimed that Google’s AI language model LaMDA is sentient as the AI expressed fears over
being “turned off” and equated it to death.1

While the veracity of Lemoine’s Kubrick-esque encounter remains shrouded in uncertainty,

there is no denying that AI has come a long way and finds its application in myriad fields.
One such field, which shall form the focus of this article, is the healthcare industry. The
process of integration of AI into the healthcare industry is still in its nascent stages but in
light of the concerns that AI algorithms might pose, such as the abovementioned, it is
essential to take stock of whether the regulatory frameworks overseeing the healthcare
industry are well-equipped to tackle this new-found challenge. This article seeks to explore
the potential applications of AI into the healthcare sector and the potential challenges that
might arise on account of such application. Further, the paper will assess whether the present
regulatory frameworks are equipped to deal with AI and the necessary changes that would
allow them to stand up to the challenge.

APPLICATIONS

AI has the potential to become a transformational force in the healthcare sector owing to the
varied and successful applications AI has found in the core areas of the industry such as
diagnostics and drug development. This section briefly looks into the five major and game
changing applications of AI in the healthcare sector.

ACCESSIBILITY: DIGITAL CONSULTATION

Patients may now consult doctors seamlessly thanks to the introduction of AI in healthcare.
Although the idea of digital consultation is not new, the earlier apps for consultation had

1
Chris Vallance, ‘Google Engineer says Lambda AI system may have its own feelings’ ( BBC, 14 June 2022)
<https://www.bbc.com/news/technology-61784011> accessed 14 June 2022.
many drawbacks. Many of the problems were resolved by recent and sophisticated
breakthroughs in AI.

First, deep learning breakthroughs allowed consumers to make more educated choices. These
AI-powered systems learned from numerous genuine use cases to ask questions that are
pertinent to the patient's health data as opposed to asking queries on the spur of the moment.2

Second, the development of natural learning has made it simpler to respond to the patient's
question. It has dramatically changed how machines used to answer queries by effectively
learning and comprehending complex language.3 These two innovations significantly altered
the healthcare industry.

DIAGNOSIS

The days when determining a disease's diagnosis required painstaking work are long past.
Healthcare has become much simpler since the introduction of AI. Doctors are able to
diagnose patients more quickly and accurately because of AI algorithms. AI makes it easier
to identify medical conditions while reducing errors.

Additionally, the employment of AI in radiology and cardiology departments has enabled the
early detection of serious problems. For example, AI-based technologies have made it
feasible to identify cancer cells in their earliest stages and have also offered strategies for
halting its spread.4 In essence, AI has transformed the way doctors used to detect illnesses
and has also helped to lower diagnostic errors.

PRECISION MEDICINES

The use of precision medicines represents one of the most significant breakthroughs in
healthcare on account of AI algorithms. Utilizing sizable multidimensional biological
datasets that contain individual heterogeneity in genes, function, and environment, precision
medicine aims to build and optimise the pathway for diagnosis, therapeutic intervention, and
prognosis.5 This gives doctors the chance to individually adapt early interventions.
2
Angela Spatharou, Solveigh Hieronimus, Jonathan Jenkins, ‘Transforming Healthcare with AI: The Impact on
the Workforce and organisations’ (Mckinsey, 10 March 2020)
<https://www.mckinsey.com/industries/healthcare-systems-and-services/our-insights/transforming-healthcare-
with-ai> accessed 10 June 2022.
3
Ibid.
4
A. Farooq, S. Anwar, M. Awais and M. Alnowami, ‘Artificial intelligence based smart diagnosis of
alzheimer's disease and mild cognitive impairment,’ (2017) International Smart Cities Conference (ISC2).
5
Mohammed Uddin, Yujiang Wang, Marc Woodbury-Smith, ‘Artificial Intelligence for Precision Medicine in
Neurodevelopmental Disorders’ (2019) 2(112) npj Digital Medicine.
Precision medicine techniques isolate patient phenotypes with less frequent responses to
therapy or particular medical requirements. 6 Through the use of complex computing to draw
inferences, AI helps to develop insights, allows the system to reason and learn, and enhances
clinical decision-making.7

DRUG DEVELOPMENT

AI systems are addressing the needs of medicine development in addition to providing

accurate diagnosis. Leading pharmaceutical companies have turned to AI based systems to
helm new drug discoveries. In order to find immuno-oncology treatments, Pfizer is using
IBM Watson, a machine learning system. While Roche subsidiary Genentech is using an AI
based system from GNS Healthcare based out of Cambridge, Massachusetts, to helm the
multinational company's search for cancer treatments, Sanofi has agreed to use the artificial
intelligence (AI) platform of UK start-up Exscientia to look for therapies for metabolic
diseases.8 The majority of sizable biopharma businesses have comparable internal initiatives
or alliances.

To find new potential treatments, they are making use of enormous databases and important
data. These medical advances can be highly helpful in the treatment of serious illnesses like
Ebola, cancer, and many others.

In addition, AI could aid in the discovery of medications that could be useful in more
successfully treating fatal diseases. It has the ability to examine subtle medical trends, which
can aid scientists or researchers in precisely tracking the development of new drugs. 9 AI can
identify hit and lead compounds, as well as expedite therapeutic target validation and
structural design optimization. Essentially, the use of AI in medicine could speed up the
process of finding new in the healthcare sector.

SURGICAL ASSISTANCE

6
K.B. Johnson, W.Q. Wei, D. Weeraratne, ‘Precision Medicine, AI, and the future of personalised healthcare’
(2021) 14(1) Clinical and Translational Science 86.
7
Ibid.
8
Nic Fleming, ‘How Artificial Intelligence is Changing Drug Discovery’ (Nature, 30 May 2018)
<https://www.nature.com/articles/d41586-018-05267-x> accessed 10 June 2022.
9
Debleena Paul, Gaurav Sanap, Snehal Shenoy, Dnyaneshwar Kalyane, Kiran Kalia, Rakesh K. Tekade,
‘Artificial Intelligence in Drug Discovery and Development’ (2021) 26(1) Drug Discovery Today 80.
The widespread adoption of AI within the healthcare industry first began in the form of
“computer-augmentation of human performance”.10 AI finds its usage in surgeries primarily
in the domains of navigation, imaging and computer-assisted intervention for intra-operative
guidance.11 Minimally invasive surgery (MIS), which is increasingly paired with robotic aid,
lowers surgical trauma. An essential component of MIS has always been computer-aided
intra-operative guidance. In order to improve localization and visualisation during surgery,
learning methodologies have been deeply integrated into the creation of intra-operative
guiding.12

Modern surgery is being transformed by the development of AI toward more autonomous and
precise intervention for the treatment of both acute and chronic ailments. These methods have
enabled significant advancements in surgical robots, intraoperative guidance, and
preoperative planning.

REGULATORY CHALLENGES

While the previous section paints a rosy picture and makes a strong case for a smooth
transition towards AI based healthcare, it is imperative to take note of the significant
regulatory challenges that inculcation of AI in healthcare poses to glean a clearer picture of
the costs and benefits involved. Although the application of AI in healthcare could lead to
better patient care, it also raises concerns about data privacy and health equity. Primarily,
adoption of AI poses three significant challenges at the outset:

 Biased AI

It has been found that AI’s replicate the biases of their creators. The expression of such biases
is rather subtle as was observed in a study published in Science in 2019, wherein it was
demonstrated that a healthcare prediction algorithm used by insurance providers and hospitals
across the US to identify patients in need of "high-risk care management" was much less

10
Daniel Hashimoto, Guy Rosman, Daniela Rus, Ozanan R, Meireles, ‘Artificial Intelligence in Surgery:
Promises and Perils’ (2018) 268(1) 70.
11
Erh-Ya (Asa) Tsui, ‘Application of Artificial Intelligence in Surgery’ (Imperial College London, 27 July
2020) <https://www.imperial.ac.uk/news/200673/application-artificial-intelligence-ai-surgery/> accessed 12
June 2022.
12
Ibid.
likely to nominate black patients.13 The bias was subtle because it was detected only when the
researched peeked inside the hood to analyse the data used to train the AI wherein it was
found that spending on healthcare was used as a proxy to determine an individual’s
healthcare needs.14 This resulted in a bias in the nominations as even though their
requirements were greater, black patients with worse health earned lower risk scores since
their healthcare costs were comparable to those of white patients with better health. The issue
of subtle bias on account of the input data brings us to the second significant problem of the;

 AI black box.

The complexity of AI prevents it from being used more widely since it makes certain of its
principles opaque to users and unreliable (the so-called "AI black box"). The decision paths
of machine learning algorithms are frequently too complex to be traced back and made
understandable to their users without human intervention because they aggregate thousands
of data points. Therefore, an AI could be effectuating racial biases and no one would even
know as the root of such bias could only be deduced from deep dive into its data processing.

 Approval Mechanisms

Current regulatory frameworks prevent medical devices from changing without first going
through a time-consuming re-authorization procedure, which poses a danger to innovation
and uptake.15 But machine learning is designed to gain knowledge from data and gradually
enhance its performance. The "locked versus adaptable" AI dilemma is what is being
presented here.16 Although the authorities' efforts are necessary, the regulation at their
disposal was not created for a rapidly developing technology like AI. Therefore, the present
regulatory approval methods are contrary to the USP of a machine learning AI which is
“dynamism” and its ability to learn and modify itself in accordance with new data.

13
Ziad Obermeyer, Brian Powers, Christine Vogeli, Sendhil Mullainathan, ‘Dissecting Racial Bias in an
Algorithm used to manage health of populations’ (2019) 366 Science 447.
14
Ibid.
15
Sara Gerke, Boris Babic, Theodoros Evgeniou, Glenn Cohen, ‘The Need for a System View to Regulate
Artificial Intelligence/Machine Learning- based software as medical device’ (2020) 3(53) npj Digital Medicine.
16
Ibid.
 TAKING STOCK

Therefore, a structural shift is required in the present regulatory practices to ensure a smooth
transition towards AI based healthcare products. At the core of such reform should be a focus
on;

1. Promoting the growth of ethical machine learning standards to assess and enhance machine
learning algorithms.

2. Promoting a patient-centered strategy that includes user disclosure of device information.

3. Creating techniques to assess and enhance machine learning algorithms.

4. Improving pilots for real-world performance monitoring.17

Presently however, no laws specifically address the use of AI in healthcare settings, despite
the fact that the industry is extensively regulated. The use of AI in healthcare has been
addressed by rules proposed by a number of nations and organisations, including the US,
although no regulations have yet been approved.18

Even outside of the realm of healthcare, governments have only just started to create
guidelines for the application of AI. While some current data privacy laws and sector-specific
restrictions do apply to the usage of AI, no nation has passed regulations specifically
governing AI. The European Union published its proposal for the initial legal framework
governing the use of AI in January 2021.19

Examples of "high-risk" AI applications in the healthcare industry are given in the EU's
proposed framework, including the use of AI to prioritise emergency aid. 20 Despite the fact
that the EU's plan does not specifically target the health care sector, analysts anticipate that
the EU legislation will act as a model for future, more focused standards. 21 The EU's plan
aims to achieve a balance between maintaining the market's safety and security while also
17
‘How FDA regulates Artificial Intelligence in Medical Products’ (Pew, 5 August 2021)
<https://www.pewtrusts.org/en/research-and-analysis/issue-briefs/2021/08/how-fda-regulates-artificial-
intelligence-in-medical-products> accessed 10 June 2022.
18
Jamison Chung, ‘How will healthcare Regulators Address Artificial Intelligence?’ (The Regulatory Review,
18 October 2021) <https://www.theregreview.org/2021/10/18/chung-how-will-health-care-regulators-address-
artificial-intelligence/> accessed 12 June 2022.
19
Ibid.
20
ibid
21
Ibid.
fostering innovation and financial investment in AI. The U.S. initiatives to deal with AI in
healthcare likewise display these opposing principles.

Guidelines for the use of AI in the healthcare sector are now being developed by both the
U.S. Food and Drug Administration (FDA) and the U.S. Department of Health and Human
Services (HHS).

A proposed regulatory framework for changes to AI-based software as a medical device was
laid out in a discussion paper that the FDA published in 2019. (SaMD).22 Software "designed
to treat, diagnose, cure, mitigate, or prevent disease" is how the FDA defines AI-based
SaMD.23 The FDA makes a promise to ensuring that AI-based SaMD "will deliver safe and
effective software functionality that improves the quality of care that patients get" in the
agency's discussion paper.24 According to the FDA, the regulatory approval process for AI-
based SaMD calls for a thorough assessment of both the product and its creator. Therefore,
legislators across jurisdictions (primarily western jurisdictions, wherein the adoption of AI is
already somewhat widespread) are alert to the rising need to reforming regulatory structures
to clamp down on the potential downsides of this newfound technology.

Conclusion

While the benefits are more obvious and the costs more obscure and nuanced, one cannot
underestimate the harms that an unregulated AI healthcare market poses. Therefore, the
existence of a robust regulatory framework should predate its widespread adoption in
healthcare services. It is essential that regulators exercise a strict oversight over the data used
to train AIs and endeavour to design an approval framework that can accommodate the
dynamism of machine learning based AIs. The European Union and the US have come up
with model regulatory proposals as a response to the rapid technological development in the
healthcare industry which is transitioning towards a more widespread use of AI based
products but proposals, even though are a step in the right direction, are a far-cry from the
necessary reform that is necessary to tame the beast that is artificial intelligence.

22
Ibid.
23
Ibid.
24
Ibid.