CATEGORY AUSTRALASIAN DENTIST 99 FEATURE: ARTIFICIAL INTELLIGENCE stewardship when implemented in the world of AI.5 A systematic literature review of AI in dentistry identified 1,533 articles, among which 178 studies were highlighted. These studies identified 53 different applications of AI in dentistry and defined 6 ethical principles for their use: prudence, justice, data protection, responsibility, democratic participation, and solidarity.6 AI is not routinely used in dental practices as yet, possibly due to data availability, a lack of structures, or questions on the benefits for the clinician, ethics, and responsibilities. Data must be used ethically throughout its retention period, yet this requires harmonized data quality assessment, management, and control to strengthen trust in AI. Thus, factors such as data quality, algorithmic bias, lack of transparency, security, and assignment of responsibility can all influence the trustworthiness of medical AI. For a comprehensive ethical classification, these factors must be evaluated from the perspectives of technology, law, and health care. Medical data is currently unstructured due to the lack of standardized annotation, and data quality can directly impact the quality of medical AI algorithm models, thereby impacting AI clinical predictions. This can also affect patient and doctor confidence in AI and pose significant risks and potential harm to the patient. In this context, the term democratization is often used and mainly focuses on the patient as the consumer and relies on limited market-based solutions. Thus, defining and imagining democratization requires a set of social goals, as well as processes and forms of participation to ensure that those affected by AI in healthcare have a say in its development and use.7 While the Hippocratic Oath and Belmont Report describe basic principles for the doctor-patient relationship, the increasing use of big data and AI techniques calls for a re-examination of the principles of privacy, confidentiality, data ownership, informed consent, epistemology, and injustice, as physicians have a traditional, fiduciary responsibility to protect the interests and privacy of their patients.8 The autonomy of doctors and the dignity of patients can be further threatened by the inclusion of AI. In addition, in the case of complications or treatment errors, the allocation of responsibility is currently unclear. Thus, the legal implications of using AI in the medical field need to be clarified here. Due to the rapid increase of AI, concerns are warranted and require a legal response to manage current and future uses. However, there is comprehensive summary in literature that examines and identifies the legal concerns of health-related AI.9 When it comes to legal questions about AI in medicine, it is also important to consider a wide range of interest groups and involve political decision-makers, developers, healthcare providers, and patients. To ensure the ethical implementation of medical AI, the following points can be defined as basic assumptions: 11 u promotion of human health is the ultimate goal u current medical AI has no moral status, so the human remains the duty bearer u strengthened data quality management u improved transparency and traceability of algorithms u reduction of distortion by algorithms u ongoing regulation and review of entire process of AI industry to control risks The AI algorithms used for decisionmaking or targeted actions are based on data and models that contain relevant information on the question to be analyzed. Since AI application in the medical field relates directly to the health and life of patients, the data and corresponding algorithms must all be collected, cleaned, and organized with extreme precision for unequivocal interpretation. Data dependency must be avoided as far as possible through comprehensive and reliable data collection to avoid bias, false assumptions, or other types of error that can affect the users and patients. Since AI applications are modelled or programmed by engineers, algorithmic procedures are needed to ensure implementation safety and avoid unforeseen consequences. Thus, supervisory bodies need to be established to monitor technological developments and ensure that preventive safeguards are in place to protect stakeholders from direct or indirect harm. Thus, it is the responsibility of AI researchers to ensure that the future impacts are positive, while ethicists and philosophers need to be deeply involved in the development of such technologies from the outset. Explainability is also a touchstone for AI decisions, meaning that it should be possible to reconstruct why an AI system produced certain predictions.12 Explainability is not just a technological issue, but also raises questions of a medical, legal, ethical, and social nature. Thus, test simulations of each question should be conducted to evaluate the accuracy and precision of the AI system, including intentional mislabeling of training images according to different values, called a “mislabeling balance” or “corruption parameter”. Even a slight corruption can affect the accuracy.13 An ethical evaluation framework for algorithms has already been proposed by Beauchamp and Childress in their Principles of Biomedical Ethics, including the criteria of autonomy, beneficence, non-harm, and justice to assess explainability. The omission of explainability in medical decision support systems represents a threat to basic ethical values in medicine and can have adverse consequences for individual health. Thus, the algorithms must be clearly communicated to the treating physicians on the basis of their explainability in order to make facilitate appropriate conclusions and highlight unclear questions. Thus, the main task lies in the collection of basic data in an ethically justifiable framework, along with open-minded thinking to discuss the problems and solutions with expert teams with different perspectives. Fig. 2: The use of AR (Augmented Reality) during surgery Fig. 3: External use of AR during surgery for education purpose
RkJQdWJsaXNoZXIy MTc3NDk3Mw==