Digital Transformation Needs Trustworthy Artificial Intelligence

摘要

As the Editor-in-Chief, Francisco Lopez-Jimenez,1Lopez-Jimenez F. Digital health in the 21st Century: the best is yet to come.Mayo Clin Proc Digital Health. 2023; 1: 52-53https://doi.org/10.1016/j.mcpdig.2023.03.001Abstract Full Text Full Text PDF Google Scholar very impressively pointed out in his first editorial that there is no way to stop the digital transformation, whether we like it or not, much like the steam engine or the electric current. And the comparison is apt—data is our oil today, and artificial intelligence (AI) is the new electricity because AI is now nearly everywhere. When AI successes, such as the current chat generative pretrained transformer-4 (ChatGPT4), are discussed in daily newspapers, it is safe to say that we are not only living in a new AI spring but already in an AI summer. In fact, ChatGPT is a good example. It shows what modern machine learning methods are capable of. However, it also very clearly shows the limitations.2Thorp H.H. ChatGPT is fun, but not an author.Science. 2023; 379: 313https://doi.org/10.1126/science.adg7879Crossref PubMed Scopus (116) Google Scholar After the initial enthusiasm, disillusionment sets in, and then the question arises as to how these machines can be used trustworthily in medicine. And that brings us to our topic. Perhaps the most important topic of AI in medicine, but also in many other areas of application, is trust. The new work by Farah et al3Farah L. Murris J.M. Borget I. Guilloux A. Martelli N.M. Katsahian S.I.M. Assessment of performance, interpretability, and explainability in AI-based health technologies: what healthcare stakeholders need to know.Mayo Clin Proc Digital Health. 2023; 1: 120-138Abstract Full Text Full Text PDF Google Scholar shows very impressively and convincingly that trust in AI-based medical devices depends on transparency (interpretability and explainability of the results) and ethics (in the sense of trustworthiness and regulation).4Müller H. Holzinger A. Plass M. Brcic L. Stumptner C. Zatloukal K. Explainability and causability for artificial intelligence-supported medical image analysis in the context of the European in vitro diagnostic regulation.New Biotechnol. 2022; 70: 67-72https://doi.org/10.1016/j.nbt.2022.05.002Crossref PubMed Scopus (9) Google Scholar The authors, after identifying the 3 main evaluation criteria for AI-based medical devices according to the Health Technology Assessment guidelines, provided a set of tools and methods to help understand how and why Machine Learning algorithms work and what predictions they make. This is of particular interest now and in the future because digital transformation (with AI as a vehicle to get there) is expected to change medicine permanently and help doctors diagnose and treat diseases, but also facilitate workflows in daily life, such as the time-consuming but mandatory medical documentation. Ideally, by reducing the routine tasks, the time freed up should be used for the following things that only human experts and not machines can do now: Although AI is able to generate art, music, or even writing, it is to date not able to create something similar to what humans can do; AI simply lacks the intuition and creativity of a human mind.5Alfaro-LeFevre R. Critical Thinking, Clinical Reasoning, and Clinical Judgment: A Practical Approach.7th ed. Elsevier Saunders, 2013Google Scholar AI can make decisions on the basis of factual data quickly and in parallel, however, AI lacks the ability to consider the wider context, social norms, ethical considerations, and genuinely human personal values, which are often essential in complex decision making.6Schoonderwoerd T.A.J. Jorritsma W. Neerincx M.A. Van Den Bosch K. Human-centered XAI: developing design patterns for explanations of clinical decision support systems.Int J Hum-Comput Stud. 2021; 154: 102684https://doi.org/10.1016/j.ijhcs.2021.102684Crossref Scopus (38) Google Scholar Although AI is technically able to recognize emotions using