10/12/2026 12:00

Sala Descubre/Online Teams. Edificio 8E. Acceso J - 4ª Planta | Universitat Politècnica de València. 46022

Abstract: The development of artificial intelligence depends not only on algorithm design and training, but also on the infrastructures required to construct ground truth, reference datasets that specify the correct or accepted outputs used to train and evaluate algorithms. Research on invisible work in ground-truthing has focused on contexts of imposition, where precarious annotators perform low-valued and unrecognized labour, rather than on settings in which developers and domain experts negotiate the conditions under which ground truths are produced. Drawing on a case study of a project developing medical-imaging algorithms for Covid-19 detection and quantification, this article examines how invisible work emerges in these negotiations. We identify what we term algorithmic infrastructure work: the design and training of algorithms not intended for clinical deployment, but instead supporting expert radiologists’ participation in annotation. We focus on a pre-segmentation model developed at radiologist co-developers’ request to reduce annotation time and sustain radiologists’ participation in segmentation. Although this model had the same architecture as the final segmentation model developed by the project and was crucial to constructing the ground truth, it was largely absent from two key representations of the project’s work: the scientific article reporting its results and an internal project progress report. We show that its emergence and invisibility were shaped by the epistemic regime of radiological AI. Radiologists’ authority generated the need for this infrastructure and may also have contributed to limiting its recognition, while valuation schemes backgrounded the technical work supporting ground-truth construction. The study demonstrates that even algorithm design and model training—the culturally most prestigious activities in AI development—can become invisible when an epistemic regime treats them as infrastructure supporting the work of others.

Bio: David Barberá es Científico Titular en el Consejo Superior de Investigaciones Científicas (CSIC). Antes de iniciar su carrera académica, trabajó durante siete años como gestor de proyectos de I+D en una empresa de tecnología médica, donde lideró numerosos proyectos de desarrollo de nuevos productos. Parte de los resultados de esta actividad se materializaron en dos patentes europeas en las que figura como inventor. Su investigación se centra en distintos ámbitos de la innovación, incluyendo la innovación médica, las políticas de innovación y la innovación social y el emprendimiento social. Sus trabajos han sido publicados en revistas internacionales de alto impacto en los campos de los Estudios de Innovación y los Estudios Organizacionales, como Research Policy, Technological Forecasting and Social Change, Academy of Management Journal o Journal of Business Venturing. Ha participado en contratos y proyectos de investigación financiados por instituciones locales, nacionales e internacionales. Asimismo, ha sido investigador visitante en universidades como UC Berkeley, Stanford University, Copenhagen Business School, VU University Amsterdam, Aalto Business School (Helsinki) y la Universidad Carlos III de Madrid, donde ha presentado su investigación en distintos seminarios académicos.

________________________________________________________________________________

Reunión de Microsoft Teams

Unirse: https://teams.microsoft.com/meet/342044936838456?p=IQrbN1Qc5c6EQFqmEL

Id. de reunión: 342 044 936 838 456

Código de acceso: jw3MQ2qS