My clinical work shows me every day where AI can create real added value – but also where the hurdles lie. AI that delivers impressive results on paper but cannot be replicated in everyday clinical practice or integrated into existing processes will not be used. My dual role allows me to derive research questions directly from clinical needs while at the same time applying the high quality standards that apply in clinical routine. AI must be measured by the same standards as any other diagnostic method: reproducibility, transparency and clinical validation.

Imagine an AI model being trained at a single location. It learns the local conditions - specific colourings, specific scanners, specific tissue preparations. At another location, the same model may then deliver completely different results. That is why cross-location collaboration is not a nice-to-have, but a basic requirement for reliable medical AI. If we want AI results in pathology to become relevant for therapy – for example, in the field of companion diagnostics - then we must develop, test and evaluate transparently together from the outset.

I am convinced that in ten years' time we will see a fundamentally different pathology. Intelligent assistance systems will accompany entire diagnostic workflows in the future – from the preliminary analysis of digitised tissue sections to recommendations for further examinations. You can imagine it as a very well-informed colleague who thinks along with you and prepares things, but deliberately leaves the final decision to the human being. Pathologists will not be replaced, but will grow into a more supervisory role thanks to AI. However, the path to get there is challenging: we need models that can explain their recommendations, solid validation standards and clear regulatory rules. If we succeed, AI will not only make pathology more efficient, precise and accessible, but will also give it a completely new role at the heart of personalised medicine.