Send message to Author

Paper short abstract

This paper examines how neurosurgeons navigate AI-assisted planning tools. Grounding my argument in a collaborative ethnographic study at the Charité in Berlin, I suggest the framework of "sketchy logics" to describe how ML-based brain modelling meets the contingencies of the operating room.

Paper long abstract

Machine learning tools for white matter tractography are increasingly embedded in neurosurgical planning. Their outputs look stunning: clean streamlines, vivid colours, tracts where they are expected. Artefacts become indistinguishable from findings. yet when applied to pathological ones — deformed by tumours, oedema, shifted anatomy — these models do not announce their limits.

Susan Leigh Star (1989) documented the neurosciences' historical quest for certainty through localisation. Drawing on six years of embedded ethnography at the Charité Image Guidance Lab in Berlin and collaborative work with the Speculative Realities Lab (SpecLab), this paper picks up that struggle from the other side: surgeons who work to keep their maps unsettled. The gap between AI workflows and clinical reality becomes a site of friction, adjustment, and improvisation. Some consider that the automated outputs are just as good, although they might be "too safe"; others insist on performing tractography manually, treating parameter adjustment as a form of thinking rather than a task to expedite. Trust here is calibrated not through confidence in the tool, but through a disciplined wariness of its visual rhetoric of certainty.

I propose the empirically grounded framework of sketchy logics — graphic, provisional, and transparent about its uncertainties — to describe these modes of practice. "Sketchy" carries a productive double meaning: iterative and explorative, but also dubious and unsettled. Through collaborative prototyping at the SpecLab, I explore what it might mean to design decision-support tools that invite contestation rather than foreclose it.