TITLE:
ABSTRACT:
KEYWORDS:
1. Introduction Imagine Alex. A month ago or so, Alex met Cal via an online dating app. They like each other. After some weeks of chatting they begin to "sext" (that is, to exchange intimate pictures). Alex likes Cal, and is aroused by the pictures. Alas, after some weeks, Alex realizes that the pictures are not photos depicting Cal. Instead, they are sophisticated AI-generated images created by Cal with the help of some state-of-art deeplearning algorithm. Now, imagine Sam. Sam owns a small start-up which sells eco-sustainable swimswears on the web. To promote them, Sam is looking for male and female models posing with them. Of course, the sexier the better. The obvious option would be to hire actual models and have them posing with the products. However, another option is available: hiring a skilled graphic designer that creates virtual models. They would be indistinguishable from real ones, if not for the watermark that is mandatory when using virtual models in Sam's country.
Until a few years ago, such scenarios could have been but the plot of some sci-fi movie. Nowadays, they describe realistic real-life situations. Hence, it becomes pressing to be able to answer questions such as: Would Alex still feel equally aroused by the images after having learnt that Cal's alleged 'nudes' are not real photos? Is it possible that Alex' arousal impeded the recognition that Cal's images are AI-generated? And should Sam hire actual models or getting virtual ones in order to get the most efficacious marketing campaign? More generally, does the belief that an image represents a real rather than a 'fake' person influences how enticing, sexually arousing and pleasant it is found? And conversely, does this arousal influence the belief that an image is a genuine photo rather than AI-generated?
This investigation has obvious practical implications for social network and dating apps policies (shall they enforce norm against presenting some else's semblances as their own? will it be easier/harder to be fooled by someone showing a sexy avatar?), models and sex workers' jobs (will virtual avatars steal their job?). However, it is also worth noting that affective reaction to sexual stimuli, far from being of interest only for the 'niche' of sexual research, represent a paradigmatic category of positively valenced and highly arousing stimuli in many popular databases of stimuli for psychological research, be them visual (Lang et al., 1997; Kurdi et al., 2017; Wierzba et al., 2015) or audio (Yang et al., 2018; Holz et al., 2022). Hence, investigating the question above may shed light on the broader theoretical issue of the relationship between beliefs of reality/fictionality and affective reactions at large. Extant studies comparing the affective reaction of a same stimulus presented as referring to either real facts or to fictions suggest that the fictional condition might dampen some dimension of the affective reaction (Mocaiber et al., 2010, 2011; Sperduti et al., 2016, 2017; Makowski et al., 2019). Interestingly, Sperduti and colleagues (2016, 2017; Makowski et al., 2019) suggest that this affective dampening is mediated by an emotional (down-)regulation strategy called 'fictional reappraisal' ("it's not blood, it's ketchup").
Among their many possible applications, state-of-art photorealistic AI-generated 'persons' have been proposed to play a role in scientific research due to the high possibility of experimental control they afford (Becker & Laycock, 2023). On the one hand, unlike virtual faces generated up to some years ago, whose processing differ in several repsect from actual human faces (Balas & Pacella, 2017), virtual faces generated by contemporary algorithms are indistinguishable by real ones (Nightingale & Farid, 2022; Miller et al., 2023). In fact, in some circumstances AI-generated faces are perceived as more trustworthy than real ones on average (Tucciarelli et al., 2022). On the other hand, believing that some face is AI-generated (irrespective of its being real or fake) modifies the observer's affective reaction. By presenting subjects several real faces presented as either real or AI-generated, researchers found that in the "allegedly artificial" condition the same face is reported as less trustworthy (Liefooghe et al., 2023) and that allegedly fake smiles evoke less corticoelectrical activity (Eiserbeck et al., 2023). A similar pattern can be found in recent works comparing the affective and aesthetic reponse of allegedly human-made with that of allegedly AI-generated artworks, the former being usually preferred than the latter (Chamberlain et al., 2018; Chiarella et al., 2022; Di Dio et al., 2023; Kirk et al., 2009; Ragot et al., 2020; Shank et al., 2023; Wu et al. 2020), suggesting a common positive bias toward the alleged human and human-made - or a negative bias against the artificially-made, if you prefer.
Based on their findings, it seems reasonable to expect that Alex' enthusiasm for Cal's alleged "photos" -- actually, AI-generated creations -- would be dampened. And that, all else being equal, Sam should rather hire actual models rather than having graphic designer creating virtual ones. Yet, the literature examined above refers to different stimulis, such as artistic creations, virtual faces, and negatively valenced pictures or clips. What ensures that their results generalize to the sexual arousal of whole-body pictures? A partial positive answer is provided by a couple of studies by Marini et al. (2024). They presented subjects with images depicting people in underwear and found out that images judged to be (by the experimental subject, in Study 1) or presented as (by the experimenters, in Study 2) real photos rather than AI-generated images also correlated with higher sexual arousal. However, several features of Marini and colleagues' studies undermine robust generalization. First, they used a relatively small sample size (N=58 and N=108) from a single country (Italy). Second, they use only mildly sexually arousing stimuli (people in underwear) collected ad hoc for the experiment. Third, they only investigated a single dependent variable, namely self-reported sexual arousal from a first-person perspective.
The present studies are aimed at replicating Marini and colleagues as well as to expand their scope. To do so, we set up a many-lab collaboration to collect data from a larger and more diverse sample, in 4 languages (English, French, Italian, Spanish). Moreover, in order to ensure a better control over image properties and enable comparisons with normative data, we employed stimuli from extant databases (EroNAPS, ), selected in a bottom-up fashion via easily reproducible procedures. Finally, as it is well documented that third-person, 'semantic' judgments and first-person, 'affective' feelings can sometimes come apart (Itkes & Kron 2019), we asked subjects to provide different ratings of first-person subjective sexual arousal, third-person evaluation of how enticing they found the image, and of affective valence elicited by the image (pleased-displeased). [Moreover, in Study 3 we collected physiological data] Lastly, we measured whether these dependent variables predicted the 'feeling of reality' of the images.
2. Material & Methods ...
3. Results ...
4. Discussion
5. References
STILL TO BE INSERTED PROPERLY: Mocaiber et al., 2010, 2011; Sperduti et al., 2016, 2017; Makowski et al., 2019 Chamberlain et al., 2018; Chiarella et al., 2022; Di Dio et al., 2023; Kirk et al., 2009; Ragot et al., 2020; Shank et al., 2023; Wu et al. 2020 Itkes & Kron 2019 Marini et al. 2024
Becker, C., & Laycock, R. (2023). Embracing deepfakes and AI‐generated images in neuroscience research. European journal of neuroscience, 58(3), 2657-2661. Eiserbeck, A., Maier, M., Baum, J., & Abdel Rahman, R. (2023). Deepfake smiles matter less—the psychological and neural impact of presumed AI-generated faces. Scientific Reports, 13(1), 16111. Holz, N., Larrouy-Maestri, P., & Poeppel, D. (2022). The variably intense vocalizations of affect and emotion (VIVAE) corpus prompts new perspective on nonspeech perception. Emotion, 22(1), 213. Kurdi, B., Lozano, S., & Banaji, M. R. (2017). Introducing the open affective standardized image set (OASIS). Behavior research methods, 49, 457-470. Lang, P. J., Bradley, M. M., & Cuthbert, B. N. (1997). International affective picture system (IAPS): Technical manual and affective ratings. NIMH Center for the Study of Emotion and Attention, 1(39-58), 3. Liefooghe, B., Oliveira, M., Leisten, L. M., Hoogers, E., Aarts, H., & Hortensius, R. (2023). Are Natural Faces Merely Labelled as Artificial Trusted Less?. Collabra: Psychology, 9(1), 73066. Miller, E. J., Steward, B. A., Witkower, Z., Sutherland, C. A., Krumhuber, E. G., & Dawel, A. (2023). AI Hyperrealism: Why AI Faces Are Perceived as More Real Than Human Ones. Psychological Science, 34(12), 1390-1403. Nightingale, S. J., & Farid, H. (2022). AI-synthesized faces are indistinguishable from real faces and more trustworthy. Proceedings of the National Academy of Sciences, 119(8), e2120481119. https://doi.org/10.1073/pnas.2120481119 Tucciarelli, R., Vehar, N., Chandaria, S., & Tsakiris, M. (2022). On the realness of people who do not exist: The social processing of artificial faces. iScience, 25(12), 105441. https://doi.org/10.1016/j.isci.2022.105441 Wierzba, M., Riegel, M., Pucz, A., Leśniewska, Z., Dragan, W. Ł., Gola, M., ... & Marchewka, A. (2015). Erotic subset for the Nencki Affective Picture System (NAPS ERO): cross-sexual comparison study. Frontiers in psychology, 6, 1336. Yang, W., Makita, K., Nakao, T., Kanayama, N., Machizawa, M. G., Sasaoka, T., ... & Miyatani, M. (2018). Affective auditory stimulus database: An expanded version of the International Affective Digitized Sounds (IADS-E). Behavior Research Methods, 50, 1415-1429.