Since its introduction in 1956, artificial intelligence (AI) has experienced cycles of excitement and disappointment. We are now in a renewed phase of interest, driven by increased user engagement and the 2022 launch of OpenAI’s ChatGPT. Indeed, AI has moved out of our laboratories and conferences to become part of our everyday lives. In healthcare, a crossroads has been reached where it is essential to recognize patients as stakeholders rather than bystanders. Their perspectives on AI integration into radiological workflows, trust in AI-assisted diagnoses, and views on accountability for potential misdiagnoses are crucial for effectively incorporating AI into the process and for clarifying the perceived role of radiologists within this workflow. According to a systematic review of multi-stakeholder preferences by Vo et al, patients showed positivity toward AI implementation in healthcare [
1]. Still, this support came with a key condition that AI should assist rather than replace human doctors, with a clear preference for human interaction over AI-driven communication [
2,
3]. It can be questioned whether these conditions and preferences are also perceived by patients undergoing radiological examinations, where, as noted, the notorious “invisibility” of radiologists perpetuates the misconception that many imaging services are mere commodities [
4]. In this issue of
European Radiology, Fransen et al conducted a multicenter survey examining the acceptance of AI among 212 patients undergoing multiparametric MRI for prostate cancer (PCa) diagnosis [
5]. The authors found that the majority of respondents were highly receptive to AI involvement, whether as a secondary or primary evaluator (79–91%). Encouragingly, most respondents showed a preference for AI working alongside a radiologist rather than as a replacement. Specifically, 91% of respondents prefer a radiologist to review AI-based diagnoses, while only 15% would accept AI as the sole decision-maker. However, other findings may be interpreted as red flags. Indeed, the proportion of patients willing to accept fully autonomous AI increases to 52% if it is demonstrated to surpass radiologists in terms of diagnostic accuracy, especially among highly educated persons. It can be hypothesized that individuals with higher education levels may also possess greater digital literacy and have encountered recent large language models, recognizing their potential. Nevertheless, targeted studies are needed to confirm this hypothesis. …
