Nina da Hora's 'Frankenstein' Paper Exposes Facial Recognition's Epistemicide
Facial recognition doesn't just misidentify—it destroys the face as a living surface.
Nina da Hora's paper, accepted to ACM FAccT 2026, reframes Mary Shelley's Frankenstein not as a cautionary tale about unintended consequences, but as a precise diagnostic for how facial recognition pipelines operate. She argues that the field enacts 'computational epistemicide'—a destruction of the face as a living, relational surface—by progressively narrowing identity through five stages: detection/cropping, landmarking, alignment/frontalization, embedding, and finally vectorization. The 'stitching' of these dissected parts produces a fixed-dimensional artifact (the embedding vector) that circulates across databases, while distance-based thresholding operationalizes a norm of 'close enough,' making recognition inseparable from standardization.
Da Hora goes further to critique reformist 'ethical AI' approaches, arguing they optimize within the same epistemic violence rather than challenging it. She concludes with a call for abolition—refusing vectorized identity as a legitimate basis for rights and access, and dismantling the institutional impulse to govern human life through dissectible data points. The paper draws on critical technology studies and Black feminist thought, offering a radical structural critique that moves beyond bias and fairness toward a rejection of the pipeline's very premises.
- Introduces 'computational epistemicide' as a new concept extending Sueli Carneiro's work to computational identity destruction.
- Traces the facial recognition pipeline through 5 stages: detection, landmarking, frontalization, embedding, and vectorization.
- Argues that 'ethical AI' fixes are insufficient and calls for abolition of vectorized identity as a basis for rights.
Why It Matters
This paper challenges the foundational assumptions of facial recognition, forcing a rethink of regulation and trust in AI identity systems.