In modern life, the application of artificial intelligence (AI) has promoted the implementation of data-driven algorithms in high-stakes domains, such as healthcare. However, it is becoming increasingly challenging for humans to understand the working and reasoning of these complex and opaque algorithms. For AI to support essential decisions in these domains, specific ethical issues need to be addressed to prevent the misinterpretation of AI, which may have severe consequences for humans. However, little research has been published on guidelines that systematically addresses ethical issues when AI techniques are applied in healthcare. In this systematic literature review, we aimed to provide an overview of ethical concerns and related strategies that are currently identified when applying AI in healthcare. The review, which followed the PRISMA guidelines, revealed 12 main ethical issues: justice and fairness, freedom and autonomy, privacy, transparency, patient safety and cyber security, trust, beneficence, responsibility, solidarity, sustainability, dignity, and conflicts. In addition to these 12 main ethical issues, we derived 19 ethical sub-issues and associated strategies from the literature.

Despite the benefits of AI-enabled solutions in different industrial sectors, their technology acceptance remains challenging. The acceptance of AI technologies depends on both the Human-AI (HAI) interaction and the ethics of AI. HAI interaction significantly affects the acceptance of AI-enabled solutions. Many guidelines have been developed to support HAI interaction design, including Microsoft's Guidelines for HAI interaction. On the other hand, many ethics by design guidelines were developed, such as the Ethics Guidelines for Trustworthy AI (EGTAI) developed by European Commission. However, there is less discussion about the possible relations between these two sets of guidelines for developing AI-enabled solutions. This study aims to analyze how current AI-enabled solutions comply with these two guidelines using a case study approach. To realize this aim, we conducted a co-evaluation workshop investigating how two existing AI-enabled apps, Strava and CoronaMelder, comply with these two guidelines. In this workshop, four participants with prior knowledge of designing with AI were asked to analyze the two cases by identifying whether these guidelines were met. The workshop results implied that when HAI interactions are designed according to the HAI interaction guidelines, they do not necessarily align with the EGTAI guidelines and vice versa.