AI Experiment with a TikTok Influencer

“Okay, hi everyone. I’m Janie,” begins a young woman with curly red hair and glasses, her hands clasped shyly in front of her as she speaks to the camera from her dorm room. “It’s day one of rush at the University of Alabama,” she goes on, casually sharing where she got her outfit: a pleated skirt from Abercrombie, a knit top from J. Crew, socks from Target, and sneakers from Nike.

This is Janie B’s debut video on her TikTok account, @jani.b086, and it’s been quite the success, garnering over 120,000 views since it was posted on August 8. The comment section is filled with praise for the college freshman, with several commenters using the word “adorable” to describe her.

Janie’s “outfit of the day” clip, often referred to as an “OOTD,” fits in with a popular trend among potential new members participating in the sorority recruitment process, known as “RushTok” on TikTok.

Recruitment Fees Skyrocket for Sorority Rush Coaches

Interestingly, Janie’s approach differs from many other RushTok participants this year. She doesn’t actually attend the University of Alabama, she didn’t go through a sorority rush, and her clothes—well, they weren’t sourced from J. Crew or Nike.

This is because Janie isn’t a real person. Every aspect of her online presence, from her dorm room to her style, is the product of artificial intelligence.

Created by Olivia Moore, an investment partner at Andreessen Horowitz, Janie serves as an experiment to gauge consumer reactions to AI-generated content. For a mere $100 in credits, Moore crafted a week’s worth of videos featuring her fabricated RushTok influencer and observed how the audience reacted, often questioning whether Janie was a real person.

Understanding Consumer Engagement with AI

Moore, who is part of the consumer investing team at Andreessen Horowitz, has a keen interest in how audiences relate to AI content. “The image, video, and audio models are advancing rapidly,” she noted, predicting we will soon surpass the “uncanny valley” stage in AI development.

“As we face an influx of this content, understanding how we humans distinguish AI-generated content will become essential,” she remarked. Moore’s experiment raises philosophical questions about consumer feelings toward AI content, especially when viewers are unaware they are engaging with something artificial.

Setting Up the Janie B. Experiment

Using tools like ChatGPT and MiniMax, Moore brought Janie to life. She tracked down a genuine University of Alabama dorm video online and uploaded it to ChatGPT, prompting it to create an image of Janie, complete with frizzy red hair and glasses. The “RushTok” influencer was thus digitally created.

Moore animated Janie’s videos by processing the static image through MiniMax H3 and inputting prompts from ChatGPT to dictate Janie’s movements and dialogue.

She chose to introduce Janie during RushTok after observing the trend over the last five years, believing it was an ideal platform to test her concept without any commercial motive that could harm anyone.

Audience Reactions to AI Content

When Janie’s first OOTD video went live, comments reflected a divided reaction. “I think anyone under about 25 recognized her as AI immediately,” Moore commented, adding that older viewers were more likely to engage with her as if she were genuine.

By the second day of the experiment, viewers had started to catch on. Discussions erupted on platforms like Reddit, and many TikTok users began correcting those who questioned Janie’s authenticity in the comments. “Some still left lovely comments, saying things like, ‘She seems so sweet. I hope she succeeds,’ even after realizing she’s AI,” Moore observed.

As awareness grew, Moore adapted her strategy, recognizing that people would still watch AI content as long as it entertained them. She began dropping subtle hints in her videos, allowing digital detectives in the comments to identify the clues.

The Detection of AI-Generated Content

Moore reported that while people are becoming adept at spotting AI, TikTok itself often failed to immediately flag her AI-generated videos. Out of the 20 videos uploaded, only 8 were initially recognized by the platform as AI-based content. As of recently, TikTok has since issued warnings on 16 of these videos.

“Platforms might struggle to detect such content at scale, especially with constantly updating AI models and strategies,” Moore warned.

Implications for AI in Media

Wrapping up her experiment, Moore presented a final video revealing her identity and the purpose behind Janie’s creation. She was pleasantly surprised by the positivity in the comments.

“Consumer attitudes towards AI remain predominantly negative, but when viewers feel they are not being deceived and are entertained, opinions can shift dramatically,” she said, anticipating a surge in consumers using AI to create engaging content that can rival human production.

“I understand the caution surrounding AI. Often, what people see is content designed to outsmart them rather than provide value,” Moore reflected. “If we approach AI honestly, showcasing how it can enhance human creativity, we could move toward a future where AI is just another tool to amplify human expression.”

The Alabama Panhellenic Association has chosen not to comment on Moore’s project.