The Atlas of AI by Kate Crawford

Author:Kate Crawford
Language: eng
Format: epub, pdf
Publisher: Yale University Press
Published: 2021-04-14T16:00:00+00:00

Columbia Gaze Dataset. From Brian A. Smith et al., “Gaze Locking: Passive Eye Contact Detection for Human-Object Interaction,” ACM Symposium on User Interface Software and Technology (UIST), October 2013, 271–80. Courtesy of Brian A. Smith

The Politics of Faces

Instead of trying to build more systems that can group expressions into machine-readable categories, we should question the origins of those categories themselves, as well as their social and political consequences. Already, affect recognition tools are being deployed in political attacks. For example, a conservative blog claimed to create a “virtual polygraph system” to assess videos of Congresswoman Ilhan Abdullahi Omar.104 By using face and speech analytics from Amazon’s Rekognition, XRVision Sentinel AI, and IBM Watson, the blogger claimed that Omar’s analytic lie score exceeded her “truth baseline” and that she was registering high on stress, contempt, and nervousness. Several conservative media outlets ran with the story, claiming that Omar is a “pathological liar” and a security threat to the nation.105

It’s known that these systems flag the speech affects of women differently from men, particularly Black women. As we saw in chapter 3, the construction of the “average” from unrepresentative training data is epistemologically suspect from the outset, with clear racial biases. A study conducted at the University of Maryland has shown that some facial recognition software interprets Black faces as having more negative emotions than white faces, particularly registering them as angrier and more contemptuous, even controlling for their degree of smiling.106

This is the danger of affect recognition tools. As we’ve seen, they take us back to the phrenological past, where spurious claims were made, allowed to stand, and deployed to support existing systems of power. The decades of scientific controversies around the idea of inferring distinct emotions from human faces underscores a central point: the one-size-fits-all recognition model is not the right metaphor for identifying emotional states. Emotions are complex, and they develop and change in relation to our families, friends, cultures, and histories, all the manifold contexts that live outside of the AI frame. In many cases, emotion detection systems do not do what they claim. Rather than directly measuring people’s interior mental states, they merely statistically optimize correlations of certain physical characteristics among facial images. The scientific foundations of automated emotion detection are in question, yet a new generation of affect tools is already making inferences across a growing range of high-stakes contexts from policing to hiring.

Even though evidence now points to the unreliability of affect detection, companies continue to seek out new sources to mine for facial imagery, vying for the leading market share of a sector that promises billions in profits. Barrett’s systemic review of the research behind inferring emotion from people’s faces concludes on a damning note: “More generally, tech companies may well be asking a question that is fundamentally wrong. Efforts to simply ‘read out’ people’s internal states from an analysis of their facial movements alone, without considering various aspects of context, are at best incomplete and at worst entirely lack validity, no matter how sophisticated the computational algorithms.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.