←Go back
Project timeline

November 2024 – January 2025

Creators

Hsin-Lun Chiu Olle Grönstedt Söderberg Oona Karppinen Oskar Rebermark Polina Sydorenko 

Project goal

Investigate whether the perception of human faces differs when it comes to real, or AI generated synthetic human faces in user experiences.

Abstract

There is an increasing amount of AI-generated media in design, however, it still remains unclear how people experience the use of such media. The purpose of this project is to investigate whether the perception of human faces differs when it comes to real, or AI-generated (AIG) synthetic human faces in user experiences (UX). This is done by generating and embedding authentic and AIG visual imagery of human faces to a chosen user interface (UI), more specifically a gym website. The UX of both media types is then evaluated using quantitative methods. The quantitative evaluation showed no statistically significant difference in the effectiveness of AIG faces versus real faces in conveying wellness, inclusivity, and socialization. However, according to the findings, real faces are more effective in fostering social connections while AIG faces performed better in conveying cultural inclusivity. These findings underscore the nuanced role of generative AI in visual communication design.

Method

  • A between-group study design was used to evaluate differences in user perception.
  • 30 participants were divided into 2 groups, each exposed to a gym website screenshots featuring either real or AIG faces.
  • Responses to questions assessing perceptions of wellness, inclusivity, and socialization were collected via questionnaire.
  • Closed-ended questions were in the form of a Likert scale of 1-5 where 1 represented “Strongly disagree” and 5 “Strongly agree”.
  • Open-ended questions asked the participants to describe their reasoning for their answers, while the responses provided additional qualitative context.
  • To evaluate statistical significance, the Mann-Whitney U test was used due to its suitability for non-normally distributed data,
  • Descriptive analysis is used to describe the central tendency of the data (visualized with box plots).

Results

While most categories showed no significant differences (p > .05), descriptive data highlighted nuanced perceptions favoring AIG faces for inclusivity and real faces for social connection.

  • Wellness: Both real and AIG faces effectively conveyed a healthy lifestyle (p = .329) but failed to inspire behavioral change (p = .475), suggesting no significant difference in how these faces influence perceptions of wellness.

  • Inclusivity: AIG faces scored significantly higher in cultural inclusivity (p = .032), likely due to AI-generated adjustments. However, no significant differences were found for inclusivity in age (p = .98), body types (p = .61), or relatability (p = .59).

  • Socialization: Real faces were perceived as better at fostering a sense of community, though not significantly so (p = .122). A significant difference was observed for opportunities to connect with others, with real faces rated higher (p = .04).

Conclusion

In conclusion, while AIG faces demonstrate potential in achieving comparable perceptions to real faces across various dimensions, their ability to convey emotional nuances and keep a human touch remains an area for improvement. These findings underscore the importance of thoughtful content selection and acknowledging the subjective nature of user perceptions when designing interfaces with AI-generated visuals.