Ethical Use of Artificial Intelligence in Dentistry

Abstract

Artificial intelligence (AI) is a technology that aims to create a machine (algorithm-software) that can mimic intelligent human behavior. In order to respect human-technology interaction in a clinical environment, AI in medicine and dentistry should have a complementary role in the work of clinical practitioners. In dentistry, various software-type algorithms are used as the basic application of AI, which is expected to improve the accuracy of dental diagnosis, provide visualization of anatomical guidelines during treatment, and, due to the possibility of analyzing large amounts of data, predict the occurrence and prognosis of oral diseases. Conscientious and ethical AI use in dentistry has to consider: (1) when to apply AI and (2) how to use AI appropriately and responsibly. Patients should be notified about how their data is used, also about the involvement of AI-based decision-making, especially if there is a lack of regulatory policy if AI is utilized to diminish costs rather than improve the health of patients or if the dentist has a conflict of interest. As many dentists are speeding in the direction of integrating AI systems into diagnostics, prognostics, and dental treatment, the legal and ethical questions are becoming even more pertinent.

1. Introduction

Artificial intelligence (AI) is a technology that aims to create a machine (algorithm-software) that can mimic intelligent human behavior. In order to respect the human-technology interaction in clinical practice, AI in medicine and dentistry should have a complementary role in the work of clinical practitioners [1]. In dentistry, various software-type algorithms are used as the basic application of artificial intelligence, which is expected to improve the accuracy of dental diagnosis, provide visualization of anatomical guidelines during treatment, and due to the possibility of analyzing large amounts of data, predict the occurrence and prognosis of oral diseases [2].

This chapter aims to provide insight into current considerations on the ethical issues that may arise due to the use of artificial intelligence in dental practice, as well as to encourage debate on ethical missteps.

A systematic literature search was conducted using three relevant international databases, namely Scopus, PubMed/MEDLINE, and Web of Science, and the following keywords were employed: ethics of artificial intelligence, artificial intelligence in dentistry, artificial intelligence in healthcare, regulations and recommendations for use of artificial intelligence, machine learning, and healthcare. The inclusion criteria were the following: (1) peer-reviewed articles in scientific journals written in English on ethical issues and AI (2) papers on regulations of using AI in healthcare; (3) ethical issues related to the use of AI systems, development, and monitoring; (4) research designs where results describe patient’s or healthcare worker’s experience. In total, 28 studies were included in the review.

2. Ethical use of AI in dentistry

Conscientious and ethical AI use in dentistry has to consider:

• when to apply AI

• how to use AI in a responsible manner

• how to avoid unethical behavior.

2.1 When to use AI

In dentistry, AI should be used if it promotes the quality of oral and systemic health, in addition to being cost-effective [3]. Both the dentist and the patient (or the designated surrogate) will need to reach a resolution about whether the best line of action implies the use of AI-based tools. Given that providing high-quality care involves both an exceptional understanding of the finest forms of care and an understanding of how to do so, the prudent dental practitioner, who aims to use AI, must resolve the following:

• Does AI-software support the principle of the best outcome for patient with the least intervention?

• Are there published studies to support these claims?

• Is there a legal regulation surrounding its use?

• Is informed consent necessary?

2.1.1 Benefit that AI may deliver to dental practice

The use of AI-based tools could accelerate diagnostic processes and enable dentists to easily access medical and dental history information, necessary for personalized approaches to patients [4], and especially important for managing patients with complex medical histories.

AI-based technology could reduce disparities [5] by improving access to oral and dental health services in resource-limited settings through applications such as for oral cancer, periodontitis, or caries detection.

From a sustainability perspective, AI, if used in preventative measures or identification of the earliest signs of oral/dental disease, may significantly reduce costs and resources engaged in therapeutic treatments.

AI has the ability to change the labor market, as well as business partnerships and patient participation. AI could be utilized to augment knowledge sharing via the collection of massive amounts of data to create a database of methods and practices that can be compared against results [6].

2.1.2 Risks associated with AI implementation in dental practice

AI use should be beneficial overall, and reliable in particular groups of patients where it is used. There is overall concern that in medical AI, inadequate performance for different races, ages, or genders could be caused by the data sets used to train the system not being sufficiently representative [7]. Thus, dental clinicians should be included in AI-based software development and monitor it, and if aware that particular models are poor predictors for particular groups, dentists should test the software first on cases involving their patients. Lack of diversity in datasets in dentistry could be a limiting factor for AI use due to several reasons:

1. There is morphometric data suggesting that there is a difference between particular male and female teeth [8] as well as aging and sexual differences of the face bone structure [9], and anatomic parameters included in dental restorative treatment planning.

2. Oral diseases are complex diseases with multiple genetic and environmental risk factors which lead to a variety of clinical manifestations and progression of the disease as well as its susceptibility to therapy or prevention. Huge environmental influences, such as smoking or diet, underlie cultural and ethnic origins, while it may vary among individuals of the same ethnic or cultural group as well.

3. The oral microbiome is the second largest microbiome in the human body with over 700 different species of microbes, showing large diversity among adults and children [10].

For example, there is an AI-based recommendation system for teeth extraction due to orthodontic treatment, which relies on cephalometric analysis. Since it has been shown that there are variations in cephalometric landmarks between Caucasian men and African-American Men [11], and if the training data was underinclusive of data of African-American men, the AI system made for cephalometric analysis could likely give treatment recommendations that are not appropriate for the African-American population. Therefore, AI system should undergo pilot testing in the dental clinic setting supervised by clinicians aiming to optimize the usage and establish trust in the system.

And how does this bias occur? When it comes to prediction, AI makes decisions based on attributes that it has “learned” from the input data through training that matched labels to the data features. Nonetheless, the model might pick up on these labels, human prejudices, or preferences. There is a potential area for human bias to appear [12]. Although AI performs well with classification tasks, it is important to note that ethical standards such as fairness and equity depend on humans. Constant monitoring should be done in order to detect errors in the system. On the other side, one needs to understand why and how AI made the recommendation that it did [12]. For an AI system to be transparent, it means to be both perceptible and understandable to outside viewers. Lack of transparency diminishes trust in AI but also, make AI system more susceptible to cyber-criminal. Wide transparency, however, can compromise privacy by making personal information concealed in underlying data sets visible [12].

2.1.3 Education and skills needed to decide whether and how to apply AI

Dentists need to acquire specific and AI use-related skills in order to apply AI safely and effectively to dental patients. The results of our recent survey conducted among experienced dentists and final-year undergraduates at the School of Dental Medicine, University in Belgrade show that working experience and having a specialization/PhD degree are associated with greater motivation and knowledge to use AI. Moreover, undergraduates are even skeptical about whether they should use AI in the practice at all, which may suggest slow adoption of AI in dental practice since both patients and health systems rely on dentists’ reactions (under review). The currently observed unwillingness of Belgrade University dental community to adopt AI underlies the evident lack of basic and continuing education regarding this subject, as well as by fear of risk that AI will replace dentist, both factors shown to be important previously also [13]. However, the present survey revealed that anxiety due to the lack of a regulatory policy represents an important factor also, and it is understandable since it may impose legal uncertainty for both patients and dentists when using AI software. Noteworthy, wide AI use may enable the general dentist to attain the level of diagnosis/treatment of the specialist by using AI software as support in decisions. Under such circumstances, the lack of experience/qualification of a general dentist may represent a matter of responsibility and liability [14]: the general dentist is not qualified to act at the level of a dental specialist. In line with this, educational programs in dental studies, as well as AI use training are necessary for achieving responsible AI use in dental practice.

2.2 How to use AI

The prudent dental practitioner who aims to use AI must resolve the following:

• Is there a legal regulation surrounding its use?

• Who is accountable if unwanted effects occur while using AI?

• How to manage data obtained while using AI.

2.2.1 Regulation

Before being used in dentistry and medicine, AI-based software must receive the approval of duly selected regulatory boards to safeguard the security of patients and their data. This implies also that both dentists and software developers need to validate AI products and continuously monitor their safety and efficacy. Regulations, standards, and guidelines must be agreed upon enabling transparency, protection of patients, and vigorous data management control.

The regulative policy includes information on whether the AI was assessed by the FDA or an alternative regulator, but also in what manner the dentist should use AI: whether the dentist should follow the AI recommendation always or ignore it if not agrees with it? [15, 16]. And what to do, if he/she has the opposite recommendation? Does this require notification of a colleague or a board of experts and a decision made afterward? Currently, AI-based software in dentistry is designed support the clinical decision and not as a major clinical decision tool, suggesting that the dentist has to supervise the AI system.

By the end of 2022, FDA has authorized over 500 AI-enabled medical devices, [17] which are mainly used in radiology. As for dentistry, FDA approved the following AI applications, engaged in the improvement of radiographs interpretation: VideaHealth’s artificial intelligence algorithm, which has been demonstrated in clinical trials to be superior to dentists in the detection of caries, while also reducing their incorrect caries diagnoses by about 15%; Overjet’s Dental Assist software, which automatically measures bone loss in radiographs, thereby speeding the time necessary to begin periodontal disease treatment, and Pearl’s Second Opinion solution, which helps dentist to spot conditions like cavities, tartar and inflammation. Noteworthy, a recent article argued about FDA regulatory practice, where authors emphasized that there were no established best practices for evaluating AI algorithms in order to ensure their reliability and safety. Namely, after reviewing publicly available information on FDA-approved devices, the authors concluded that almost all of the AI devices were approved after conduction of only retrospective studies while the majority of approved devices have been evaluated only at a small number of sites, suggesting lack of geographic diversity [18].

2.2.2 Accountability

Safe use of AI tools in dentistry needs the dentist’s supervision and the role of dental practitioners is crucial in preventing dental complications as well as in reviewing and monitoring AI systems. On the other side, AI use could be associated with an automation bias, due to the human tendency to favor machine-generated decisions, ignoring contrary data or conflicting human decisions [19]. Automation bias causes mistakes when people miss or ignore an AI system’s advice, or when a clinician chooses to follow a machine’s opinion despite opposing data [19].

The question of who takes the blame when unwanted events occur during AI use must be addressed. If it is not addressed and clarified, patients’ trust in AI could be severely damaged. Is it a developing company/software engineer, engaged in AI software development, a dentist as a major AI user, or government agency, which selects, validates, and deploys AI-based software in the health facility? Indeed, WHO finds it difficult, both legally and morally, to assign responsibility for the use of AI for health care, since it is diffused among all the contributors in this AI usage chain: developers and providers as well as the government agency or health institution [5]. However, under circumstances of the lack of legal policy on dentist use of AI software, it is suggested that the dentist is accountable for harm resulting from AI use: the dentist is accountable if fails to critically evaluate AI-based recommendations, if misuses AI decision-support tool as a major diagnostic resource, or if uses an inadequate AI system in dental practice [20]. On the other side, governmental agencies as well as health facilities, which approved specific AI system to be used, could be accountable and liable if there is no adequate training or technical support for AI software as well as for harm resulting from inaccurate AI system because it is their recommendation which AI system should be used in clinical practice. On the other side, algorithm developers could be accountable for injuries that result from poor and biased AI design, or manufacturing defects [20].

It should be emphasized, however, that even if liability is difficult to assign, compensation for the harm that a patient suffered due to AI use, could be provided without the assignment of fault or liability. Such circumstances could be seen, for example, in the case of adverse effects associated with vaccine administration. There is a suggestion that in such cases, compensation could be paid from funds founded by companies that develop AI software [5].

2.2.3 Data management

Data management is one of the fundamental ethical issues dealing with acquiring, assessing, and storage of data. Key areas include informed consent, privacy, and data protection [21]. Dentists have a moral duty to use the data they assemble to improve the patient’s health as well as to improve the dental practice. However, they are not permitted to utilize data in ways that could endanger patients, have a negative impact on them, or be discriminatory. Since AI system development requires a huge number of high-quality data, these data have huge value and there is social pressure to make them available commercially [22].

Patients have the right to see what personal data the dentist has about them and how she/he using it, have the right to restrict or ask her/him to stop processing or delete their data [23].

According to General Data Protection Regulation [24], in order to ensure data security, the dentist should: limit the amount of personal data collected or delete data of no longer need; pseudonymize or anonymize personal data whenever possible; foster email security and device encryption and notify the patients if a data breach occurs.

2.2.4 Informed consent

In the case of an AI system that shapes dental recommendation, the question of informed consent is of key importance. However, the complexity surrounding AI software development and its mode of action, makes communication with the patient difficult—how to explain accurately such complex information to the patient in an understandable way? There is also a question of what amount of information is sufficient and optimal to achieve patient comprehension necessary for decision-making. Furthermore, despite well-intentioned and well-designed attempts, many patients are said to not understand the information given to them, and even if they did, many are said to not apply the knowledge to help them make decisions [25]. Another important issue arises from concerns about biased training data for AI/ML. Whether and in what way it should be delivered to the patients [26]?

Having this in mind, there are a few situations where patients need to be extensively informed:

• when patients inquire about the involvement of AI dental systems, many dentists are racing towards the adoption of AI in the dental practice and advertising its use as an advantage to the competitors. Thus, patients could inquire whether and how AI system will be used in their particular case.

• when the AI system is more opaque- Algorithms can be opaque because the rules they are relying on are too complex for dentists and patients to explicitly understand, or because of the machine-learning techniques employed, which often could not be explainable not even to algorithm developers. So, it is probable that the rules governing the algorithms used in medical AI maybe not just not explained but also not possible to explain.

• when an AI system is given a crucial role in the final decision-making, patients need to know if AI system is monitored by a dentist, and used as support in clinical decisions, or if it is used as the main decision-making tool

• if there is doubt that the AI system is used for reasons other than to improve patient health, such as when dentist or health facility has a conflict of interest. Namely, a dentist who has his own research or financial interests may be tempted to order a scientifically useful procedure or test that offers marginal, or no, benefits to the patient. When choosing whether to consent to a suggested intervention, a reasonable patient would want to know whether any interests other than the patient’s health may have played a role in the dentist’s choice. This is crucial to informed consent since it is relevant to the patient’s choice.

3. Unethical concerns and how to avoid them

3.1 Conflict of interest

The growing number of cases where dentists are involved in the development of AI applications for dental use, as founders or Board members in companies engaged in AI- dental applications development, impose the need to discuss liability matters under the circumstances of conflict of interest.

Dentist must disclose personal interests which are not related to the patient’s health, whether research or economic, since it may affect his/her professional judgment. The liability issue exists due to the fact that the financial motive (to commercialize developed AI application) may compromise the reasoning of the dentist when it comes to decision-making [27]. By favoring his/her own interest over patients, the dentist violates fiduciary duties. Namely, law recognizes dentist’s duties to patients as fiduciary duties, which oblige dentist and other medical professionals to act in a manner only to protect the patient’s overall interests from doctor’s self-interest [27]. Resolving a conflict of interest could be achieved by avoidance of the dentist being directly involved or, when avoidance is not possible, by disclosing it to patients before treatment or research, and then letting the patients decide whether or not such interests are acceptable to them [28].

3.2 Misuse of AI software

Currently, AI-based systems for dental practice are just supporting systems in clinical decision-making and must be supervised and continuously monitored by a dentist. Misuse of AI decision-support tool as a primary diagnostic tool represents unethical behavior and could be a matter of legal liability [14]. At this moment, there is little experience regarding the operational characteristics of AI-based systems for dental care in diverse clinical settings [29]. So far, extensive research is done to clarify technical difficulties in AI development; however, there is a lack of those which consider the community’s ethics. Dentists have a duty to understand the risks associated with AI software use, to alert patients to it, and to monitor AI-based systems to prevent harm. Dentists should be aware that despite many positive impacts, the use of AI systems may induce social and economic changes, which often disproportionately negatively affect the most vulnerable communities. Thus, continuous postimplementation monitoring for unwanted effects should be accompanied by a strict risk management protocol for determining causes and implementing corrective action [30]. Dentists will stay responsible for patient care and will need to obtain education and training to obtain new skills in order to use AI systems ethically and responsibly. Likewise, besides educational and ethical, dental community needs legal frameworks as well, in order to successfully implement AI system as safe, reliable, and sustainable medical devices in dental practice.

3.3 Breach of data privacy

It is essential to fully inform patients about how their data gathered by an AI system is processed, aiming to promote mutual trust, as well as trust in AI system. If patients and dentists do not trust AI, the adoption of AI into clinical dental practice will be slowed down and ultimately fail. The value of health data can reach up to billions of dollars, and it is unethical to sell patient data for profit. According to the General Data Protection Regulation (GDPR) [24], a European law that established the privacy of personal data protection, patients have a right to withdraw their data and request the deletion of their data at any time. Some AI health applications may jeopardize patient’s data privacy by sharing patient data not only with the doctor but also with friends and family members. This could be tricky since, In contrast to the doctor who is subject to duties of confidentiality set out by governing law, family members or friends do not have such legal obligations [31, 32].

The introduction of AI-based software may reduce disparities and increase the operative efficiency and dissemination of dental knowledge and best practices, sharing information between medical and dental providers, which is especially important for medically compromised dental patients. However, caution is needed since the large diversity in orofacial bone- and teeth anatomy, as well as huge environmental impact on oral disease development and progression, may influence AI bias in the decision-making process in dentistry, in a great manner. In short, ethical practice accompanying AI use in dentistry should follow: (1) AI system needs to be approved by legitimately chosen regulatory boards before they are put into use, (2) Dentists should be educated and trained in AI use, and should supervise and continuously monitor the performance of AI-based system in their group of patients, (3) To ensure the safety of patients and their data, AI use should enable transparency, protection of patients, and vigorous data management control; (4) Patients need to be extensively informed if there is doubt that the AI system is used for reasons other than to improve patient health, such as when dentist or health facility has a conflict of interest.

4. Conclusions

Introduction of AI-based software may reduce disparities and increase the operative efficiency and dissemination of dental knowledge and best practices, sharing information between medical and dental providers, which is especially important for medically compromised dental patients. However, caution is needed since the large diversity in orofacial bone- and teeth anatomy, as well as huge environmental impact on oral disease development and progression, may influence AI bias in the decision-making process in dentistry, in a great manner. In short, ethical practice accompanying AI use in dentistry should follow: (1) AI system needs to be approved by legitimately chosen regulatory boards before they are put into use, (2) Dentists should be educated and trained in AI-use, and should supervise and continuously monitor the performance of AI-based system in their group of patients. (3) To ensure the safety of patients and their data, AI use should enable transparency, protection of patients, and vigorous data management control; (4) Patients need to be extensively informed if there is doubt that the AI system is used for reasons other than to improve patient health, such as when dentist or health facility has a conflict of interest.

Acknowledgments

This research was funded by the Ministry of Education, Science and Technological Development of the Republic of Serbia, Grants No. 451-03-9/2021-2114/200129 and No. 200110.

Conflict of interest

The authors declare no conflict of interest.