sharing intention (Left panels). ▪Introduction ▪Methods ▪ Political content Content with negative images were more likely to being shared (Left panel). ▪ Entertainment content There were no consistent patterns in the effect of negative valence (Right panel). ▪ Political content ▪ Entertainment content Kei Ichikawa, Jiayu Chen, and Kazutoshi Sasahara School of Environment and Society, Tokyo Institute of Technology • RQ1: Does negative valence in image affect content sharing? • RQ2: Does altering negative valence in image mitigate contents sharing? RQA1: Yes, but it depends on the topic. RQA2: Yes and no; it depends on the sharing intention, especially in entertainment case. Most results were similar to the politics case, but replacing negative images to neutral images encouraged sharing among those highly willing to share them (Right panels). Political content Entertainment content Political content Entertainment content This work was supported by JST, CREST (JPMJCR20D3), JSPS KAKENHI (JP23H00504) ▪Discussion Prolific • US citizen • =>18 years old • N=250（each condition） Qualtrics • Demographic • partisanship • Cognitive reflection • SNS usage Etc.. Qualtrics • Knowledge and Experience with Deepfakes • Reasons for Sharing (Emotions, etc.) Qualtrics Intention to Share AI-Generated Content (Politics, Entertainment) Experiment 1: Impact of Emotional Valence Experiment 2: Image replacement. ★ Negative images were selected rather than non-negative ones. Blue for negative choice and orange for neutral choise. • Participants were recruited to the survey experiment (see the flowchart below). • GPT-3.5 and Stable Diffusion were used to create pseudo-news and images, respectively. • The pseudo-news with two images (one negative and one neutral) were presented to participants in an A/B test fashion (Study 1) and sequentially (Study 2), and then participants indicated which they would prefer to share. Political content Entertainment content Neu->Nega Nega->Neu Neu->Nega (interaction with 1st sharing intension) Nega->Neu (interaction with 1st sharing intension) Neu->Nega Nega->Neu Neu->Nega (interaction with 1st sharing intension) Nega->Neu (interaction with 1st sharing intension) Extended abstract (PDF) • Negative valence affects the intention to share AI-generated political content, which decreases when negative images are replaced by non-negative (neutral). • Generative AI has transformed the creation of emotionally impactful material, which poses significant challenges to content verifications. • We examine how emotional valence in AI-generated content (deepfakes) influences their sharing intentions, exploring the impact of altering emotional valence on users' willingness to share. • In entertainment content, higher initial sharing intention weakens this effect. Sharing intentions are more complex than in political content.
ichika035
jnunemaker
bermonpainter
shlominoach
samanthasiow
lauravandoore
tmiket
seathinner
kristinabergwall1
szymonslowik
akihiro_kokubo
shpigford
michielstock
