What is beautiful is still good, especially if you are man: The Attractiveness Halo Effect in the era of AI-based Beauty Filters

Authors: Gulati, A. , Martinez-Garcia, M. , Lepri, B. , Oliver, N.

Publication: International Conference on Computational Social Science, 2024

Human perception, memory and decision-making are impacted by tens of cognitive biases and heuristics that influence our actions and decisions. Despite the pervasiveness of such biases in human decision making, they are generally neither considered nor constructively leveraged by today’s Artificial Intelligence (AI) systems that model human behavior and interact with humans. A widespread yet understudied cognitive bias in this context is the attractiveness halo effect. According to this bias, perceived attractiveness impacts our judgments of unrelated attributes, such as intelligence, trustworthiness or sociability. Leveraging AI-based beauty filters, we study the attractiveness halo effect on a diverse set of 462 different individuals in two conditions: ‘unfiltered’ (original pictures) and ‘filtered or beautified’ (images with the beauty filter applied). We conduct a large-scale user study with 2,748 participants who rated faces from this dataset regarding their attractiveness and six other attributes (intelligence, trustworthiness, sociability, happiness, femininity and unusualness). In this paper we report results corresponding to ratings of attractiveness, intelligence and trustworthiness, given their relevance in human-AI collaboration and human-to-human interaction. We find that beautified images receive statistically significantly higher ratings not only of attractiveness but also of the other attributes. While we report strong evidence of the existence of the attractiveness halo effect both before and after applying the beauty filters, we observe a significant reduction in its strength in the beautified condition. Thus, we hypothesize that beauty filters could be used as a potential tool to mitigate the attractiveness halo effect. We also study the impact on this bias of the age, gender and race of the stimuli and the raters. We identify that gender —both of the stimulus and the rater— plays a significant role on perceptions of attractiveness, intelligence and trustworthiness. Our findings raise ethical concerns and call for the need of transparency and ethical guidelines regarding the use of beauty filters.