Recently, students at the University of Washington professional social computing group I noticed a trend on social media. People were creating short scientific videos using generative artificial intelligence. The problem is that these people are not scientists, and given that AI tends to be convincingly wrong, this could accelerate the spread of misinformation. So the lab wondered how scientists and other researchers could better adapt to platforms like TikTok.
“Another option is that the science is being told without scientists,” said the co-lead author. messiah ruby cristobala California State University doctoral student majoring in human-centered design and engineering.
These discussions led the team to papertokuis an AI tool that helps users convert research papers into 45-second videos. When a researcher uploads a paper to the tool, Google Gemini is used to create a short script that explains the paper. Researchers can then iteratively edit the transcripts and resulting video clips.
team published research April 17, at the Association for Computing Machinery Conference on Human Factors in Computing Systems, Barcelona.
“Most people don’t read research papers for several reasons,” says the senior author. gary sheaa professor at the University of California specializing in human-centered design and engineering. “Reading papers in unfamiliar fields is still a challenge, so we wanted to find a way to quickly convert papers into a format that the average person would want to work on. And we wanted to study how laypeople approach it.”
Currently, PaperTok is only accessible to users with a paid Google Gemini subscription. Those users are PaperTok site Upload your research paper. The system will then display four options to use as video hooks. For example, the PaperTok video about PaperTok itself begins, “Have you ever felt overwhelmed by reading a thick academic paper?”
“We started by interviewing eight science communicators and content producers about how to create engaging and believable videos,” said the co-lead authors. Shin Dong HoonPhD student at the University of California In the field of human-centered design and engineering. “We’ve found that hooks are essential for short-form videos. You’re competing with other videos online, and you only have a few seconds to grab someone’s attention.”
Once you select a hook, PaperTok generates a script that you can edit. During the storyboarding stage, the script is divided into scenes, similar to storyboarding a movie. Users can continue to improve their scripts and video clips. Once you are satisfied with the results, you can add a byline that will appear at the end along with the paper’s author.
The team asked 100 online participants and 18 academic participants to compare videos from PaperTok with videos from two other PDF-to-video generators. They found PaperTok easy to use and its videos were more engaging than videos from other systems. But some wanted to share it publicly because of signs of AI, such as meaningless text, because they thought it was “too AI-like” and could undermine the credibility of their scholarship.
The team plans to continue working on ways to customize AI-generated videos, such as allowing users to draw on specific parts of the scene and change elements according to their intent.
“The main motivation behind PaperTok was, ‘How can we enable researchers to create engaging short-form videos?'” Cristobal said. “With generative AI tools, anyone can generate a video from a PDF in minutes, which creates all kinds of problems, including misinformation and AI missteps. So we wanted to build a tool that keeps humans, ideally experts, involved. If anything, we want PaperTok to highlight how important humans are in science communication.”
Co-authors include: Hyun Jeong Byuna California State University doctoral student majoring in human-centered design and engineering. Tseyu Chen Dr. Boson AI. He contributed to this research as a master’s student at the University of Wisconsin. Ruoshi Xianga university doctoral candidate in human-centered design and engineering. Lui Kang Chonga California State University doctoral student majoring in human-centered design and engineering. and Tony Zhoua student at the University of California studying computer science. This research was supported by a Microsoft AI and New Future of Work Award, a Google PaliGemma Academic Program GCP Credit Award, and a National Science Foundation CISE Grade Fellowships.
For more information, please contact Hsieh. garyhs@uw.eduShinat dhoon@uw.edu And Cristobal meziah@uw.edu.
