Why Do People Trust Physiognomic AI?

Abstract

In this article, we examine the concerning trend of increasing physiognomic artificial intelligence (AI) applications, such as those assessing an individual’s employability, criminality, and even sexual orientation, and attempt to understand why trust may be given to these systems despite their biased and baseless conclusions. While the practice of linking facial features to human characteristics should be rightfully rejected at its core due to the harmful prejudices and antithesis it poses to responsible AI development, this still has not stopped the recent influx of literature claiming to provide unbiased insights in this field of study. To understand this trend, we examine these claims through the lens of a trust model to hypothesize why such applications are gaining acceptance. After reviewing recent literature for common trends, it appears that these applications are gaining acceptance and trust under the guise of big data through the use of exceptionally large datasets, cognitive bias toward believing the output of mathematical frameworks, and data dredging to find relationships justifying their physiognomic hypothesis. As such, we show that when these factors are combined, each contributing toward various dispositions to trust, it unfortunately leads to situations where acceptance of their faulty results becomes a plausible reality, harming individuals affected by their outputs.

Ähnliche Arbeiten