UNIVERSITY PARK, Pa. — The rabbit hole contains madness, according to author Lewis Carroll. Online, that madness manifests in the form of increasingly extreme content, often without users realizing it. A new study by Penn State researchers suggests that giving users control over the interface feature of autoplay can help them realize that they are going down a rabbit hole. 

The work — which the researchers said has implications for responsibly designing online content viewing platforms and algorithms, as well as helping users better recognize extreme content  — is available online and will be published in the October issue of the International Journal of Human-Computer Studies.

“Anyone who has used YouTube or similar websites will know that these platforms automatically play the next video, without waiting for us to initiate it,” said senior investigator S. Shyam Sundar, Evan Pugh University Professor and the James P. Jimirro Professor of Media Effects at the Penn State Bellisario College of Communications. “We often hear about people going down rabbit holes of extreme content online, because these platforms automatically transition from mainstream to extreme content, in order to maintain audience interest.”

For example, a search for jogging may recommend increasingly extreme content, moving from jogging to running, then marathons to ultramarathons of 50 to 100 miles. The same can be said for any topic, the researchers said, including those that already lend themselves to polarization, such as politics.

“People tend to blame the autoplay feature — when one video ends, another plays automatically — for the rabbit hole perception, but we’ve yet to unpack the psychological effect of autoplay in the context of online viewing,” said lead author Cheng "Chris" Chen, assistant professor of communication design at Elon University who earned her doctorate in mass communications from Penn State. “Prior studies have pointed out that being stuck in a rabbit hole is a complex experience, which could also be influenced by one’s prior media consumption experience.”

To understand how autoplay and prior media consumption may work together to influence a user’s perception of falling down the rabbit hole of extreme content, the researchers designed an experimental video platform dubbed VIDNATION. The platform had 12 versions, each with video combinations of consistently non-extreme content or increasingly extreme content under three different autoplay modes: the ability to toggle autoplay on or off, autoplay without the option to turn it off and manually clicking the next video to play.

The researchers recruited 394 online participants and randomly assigned them to different VIDNATION versions. After receiving a tour of the interface, the users watched four one-minute-long videos of either non-extreme or increasingly extreme content with varying levels of control.

Participants completed a questionnaire before using VIDNATION, documenting how much and what type of content they typically watched. They also completed a questionnaire after exiting the platform, indicating the control they felt, the extremity they perceived in the content and whether they believed they had gone down a rabbit hole.

“We found that altering aspects of online media technology can have an effect on people’s perceptions of what they are consuming,” said Sundar, who is co-director the Media Effects Research Laboratory and director of the Center for Socially Responsible Artificial Intelligence at Penn State. “It is important to provide a reasonable level of control to users, so that they can decide for themselves whether or not they want to continue watching mainstream content that veers toward extreme material.”

The researchers pointed to the concept of “interpassivity,” or the idea that a user can allow technology — such as enabling the autoplay feature — to make decisions on their behalf, as key to triggering a sense of control in them.

“Autoplay is not just a passive experience; it offers both automation and interactivity,” Chen said. “Rabbit hole perception is not only influenced by algorithms and browsing history but also by how users engage with autoplay. Our study shows that when users have control over the action of autoplay, their engagement with autoplay — toggling it on or off — can increase or decrease their perception of falling into a rabbit hole.”

Once a user toggled autoplay on, the researchers found that they felt more conscious of their media experience and were more likely to see the rabbit hole. However, the feeling of control also made them less conscious of the fact that the content violated their expectations, making them less likely to perceive a rabbit hole.

For those who reported lower amounts of online viewing prior to the experiment, the passive viewing experience of autoplayed videos without any manual control led to an increased perception of falling down the rabbit hole even when the videos consisted of non-extreme content.

The researchers said these results can be used to promote more mindful use of online platforms and inform more responsible platform designs.

“Often, people go down rabbit holes without realizing that they are exposed to fringe, extreme content,” Sundar said. “The more novel the next video, the more likely that a person would continue watching, even if the material is sensational. They may mistake it to be mainstream opinion. We want to identify and promote interface designs that enable people, especially heavy users, to be thoughtful consumers who realize that what they are watching — beyond a point — is not mainstream content, so they can adjust their views and behaviors based on that understanding.”

Other co-authors include Jingshi Kang, a doctoral student at Fudan University who was a visiting scholar at Penn State in 2023, and Pejman Sajjadi, user experience researcher at Meta in California, who was a postdoctoral scholar at Penn State.