AI v/s Human Perception

Auditory Priming Effects on Facial Identification (2023)

Background

The rise of artificial intelligence, specifically generative AI, has brought into question the ability for humans to properly recognize and differentiate between human and AI content. We wanted to understand the factors that influence an individual’s perception of what is and is not AI-generated. In particular, whether auditory priming can impact the way a person perceives media.

In the absence of sufficient stimuli and processing time, we want to know whether people will rely on auditory cues to fill in the source details of the people they see around them. Would a robotic, unfamiliar primer noise cause an individual to perceive a face as more likely to be AI-generated? Would a musical, familiar primer noise induce the perception that a face is more likely to be human?

Hypothesis

Participants will be more likely to label a face as AI-Generated if they are primed with a robotic sound (beep), and will be more likely to label it as Human if they are primed with a musical note (chime).

Methods

2x2 Within Subjects design; all participants were exposed to all 4 testing conditions:

  • Human Face x Chime

  • Human Face x Beep

  • AI-Generated Face x Chime

  • AI- Generated Face x Beep

The study was created and conducted on PsychoPy.

12 AI-Generated faces and 12 Human faces were used; the faces were displayed on the screen for 2 seconds each, prefaced by either a musical note (chime) or a robotic tone (beep) lasting 0.50 seconds before each face appearance.

Participants then had unlimited time to press one of the two response keys, denoting AI Generated Face (A key) and Human Face (H key)

Results

  • N = 34

  • Overall average of 102.5 correct face identifications when a Chime was used as an auditory primer, and an overall average 105 correct face identifications when a Beep was used.

  • These results were then analyzed using a repeated measures ANOVA test, yielding no statistically significant results (F(1, 29), p = 0.316).

  • Average correctness response of 49.4%, deemed statistically insignificant (p = 0.696).

  • However, subjects are significantly more likely (56%) to guess a face is human rather than AI-Generated (t(34) = 34, p = 0.006).

Findings

The study’s results did not support our hypothesis that an auditory primer will impact a participant’s perception of faces presented to them. While there was a significant difference in the number of “human” guesses rather than “AI”, there was no measured interaction between the tone of the auditory primer and the type of face classification produced as a result.

It was, however, found that participants were statistically more likely (56%) to default their guess of the origin of a face to “Human” rather than AI, which opens up the scope for further research into the biases that may influence participant behavior to see a face as “Human”, or familiar, rather than foreign.