Anna Skarpelis — Face It! (ULiège)

Value of the Information & Strength of the Argument

The value of this talk lies in its critical, interdisciplinary perspective on AI, drawing from sociology, history, and media studies. Skarpelis effectively deconstructs the technical rationality behind facial recognition and data augmentation, revealing how racial categories are socially constructed and embedded in technical systems. Her argumentation is solid: she moves from a specific empirical example to broader theoretical reflections on personhood and datification, using historical analogies to challenge claims of unprecedented novelty. She successfully demonstrates the ethical stakes involved in seemingly benign technical fixes, making a strong case that the reduction of human identity to thin data representations is both morally problematic and historically continuous. The talk is well-structured, clearly presenting the motivations of computer vision engineers, the technical processes involved, and the social implications, while maintaining a critical stance that avoids both techno-optimism and outright dismissal.

Scientific Rigor, Source Quality, Title Accuracy

The talk shows rigorous scholarship, with references to specific research papers (e.g., the 2020 paper by ‘Uura’ and colleagues), datasets (Microsoft Celeb), and scholars from media studies and sociology (Estelle Blashke, Lev Manovich, Ian Hacking). Skarpelis is transparent about her own methodological approach, including her ongoing meta-analysis of scientific literature. However, as a talk, it does not provide full bibliographic details or direct URLs, so the audience would need to consult the description or further communication for exact references. The title ‘Face It!’ is catchy but vague; the actual talk content aligns with the description’s subtitle, which accurately reflects the focus on facial recognition and ethical consequences. Overall, the title-content fit is adequate, if slightly whimsical.

The title 'Face It!' is a catchy phrase, but the subtitle in the description clarifies the actual content. The video content matches the described talk about facial recognition and ethical consequences.

8/10

The talk is based on the speaker's ongoing research, cites relevant literature, and presents a critical analysis of AI practices. It demonstrates methodological rigor and contextualizes the topic within historical and sociological frameworks.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 00:00Introduction and welcome to the talk by Anna Skarpelis.
• 02:00Presentation of the racial translation matrix of four portraits.
• 05:00Explanation of generative adversarial networks and data augmentation techniques.
• 10:00Discussion of the meta-analysis of computer vision research on race.
• 15:00Historical context and continuity of racial classification practices.
• 20:00Implications for personhood and the thinning of data in AI.

• 2020 paper by Uura et al. on racial translation — Described as a group of Chinese researchers who developed the racial faces in the wild dataset and performed racial translation using GANs.
• Microsoft Celeb dataset — Source material for generating the racial faces in the wild dataset.
• Estelle Blashke on image annotation — Cited as showing that image annotation predates digital imagery.
• Lev Manovich on the database — Referenced regarding the database as a major cultural form of the 21st century.
• Ian Hacking on making up people — Mentioned in the context of how classification shapes personhood.