TikTok users are more productive — and negative — when they use AI voice: study

When TikTok introduced the ability to turn text into AI speech, it was adopted by millions of users around the globe. Suddenly creators who didn’t have time, recording skills or a command of the spoken language — or who simply wanted to remain anonymous — could easily add a clear, confident voice to their videos.

But research from the UBC Sauder School of Business shows that AI voice is much more than a handy new tool: It’s fundamentally changing the kinds of videos users create.

Researchers used machine-learning algorithms to analyze a dataset of 554,252 videos, looking at the length of the videos, how novel they were, whether the creators disclosed their identities, and whether the overall sentiment of the videos was negative or positive. They also examined whether users of the AI voice became more or less productive. 

What they found was that creators who used AI voice produced 24-per-cent more videos per week, and the total duration of the videos they create per week were 63-per-cent longer. Creators with less experience or popularity had a greater increase in productivity.

“Video creation is a very complicated process, and there are many components: the topic, the storyline, the video, the audio,” said study co-author and UBC Sauder associate professor Gene Moo Lee. “If you can offload one task to AI, you can use your cognitive resources, your time and your energy for the other elements. So your productivity goes up.”

The study also shows the AI voice leads to more novel videos — that is, ones that are different in terms of text, visuals, and audio content. This effect is more pronounced in experienced creators, likely because they have the skills to take full advantage of the new tool’s potential.

AI voice also allows for anonymity, which may lead to more negativity.

“After people adopt this tool, the sentiment they use in the videos tends to be more negative — and creators tend to show up in the videos less frequently, so they are trying to hide their identity more,” said UBC Sauder assistant professor Mi Zhou, who co-authored the study with Lee and UBC Sauder PhD student Xiaoke Zhang. 

“We think it’s because of the disinhibition effect: Because your identity isn’t tied to the content, you can be more open,” said Lee. The researchers found the effect was also lasting. Even when users went back to using their own voice, the sentiments remained more negative.

Interestingly, Zhou said it’s becoming more challenging to study AI speech because the computer-generated voices have evolved to the point where they are almost indistinguishable from human ones. Next, the researchers plan to examine the effect of AI voice on consumers and determine whether the use of AI-generated speech changes their response to video content. 

“A simple tool can have a huge impact. Sometimes it’s an intended impact, like improving productivity, but sometimes there are things we can’t foresee,” said Lee, who along with his co-authors, is currently working on an AI-based design project with South Korean carmaker Kia. 

Zhou agrees, and adds that it’s essential that social media platforms, policymakers and regulators step in to ensure transparency. “It’s important for the platforms and policymakers to come up with effective ways to help people distinguish which content is AI and which is human. People have the right to know,” she said.

Recently, TikTok has been taking steps to label AI-generated content, and just this past week introduced a partnership with the Coalition for Content Provenance and Authenticity (C2PA) to become the first video-sharing platform to implement their Content Credentials technology.

“People worry about “deep fakes” that can create trust issues in online platforms,” said Lee. “I think these are important steps that TikTok is taking to build trust and transparency in the creator economy. However, with the rapid development of AI technologies, there will be technical challenges in automatically identifying AI-generated content.”

The paper “How Does AI-Generated Voice Affect Online Video Creation? Evidence from TikTok” was presented at the Conference on Information Systems and Technology 2023.