In recent years, intriguing developments have been made possible by the fusion of artificial intelligence (AI) and extended reality (XR) technologies, such as augmented reality (AR) and virtual reality (VR).
The current study, however, indicates that there might be privacy issues with this integration. In AR and VR environments, AI systems may analyze user motion data, revealing personal information with startling accuracy.
Discussions regarding how to find a balance between technological advancement and personal privacy have been sparked by this new potential.
Researchers from U.C. Berkeley claim that users of augmented and virtual worlds reveal more information than was previously thought possible from motion data.
In two experiments done by the university that were published earlier this year, authors found that users could be recognized using just a few seconds of their head and hand motions. Numerous linked characteristics, including age and the presence of a disability, can be inferred from such obtained data.
Studies reveal that AI in AR and VR can extract personal information from user motion data.
Users disclose much more data than they anticipate. They make it available to all users, not just the software or hardware. The study’s lead author, Ph.D. student Vivek Nair, from Berkeley’s Department of Electrical Engineering and Computer Sciences, said as much in a press statement. It presents a clear threat as a result.
The Centre for Responsible, Decentralized Intelligence’s Metaverse security and privacy research effort resulted in the completion of both articles.
One study was based on a dataset that was more than 100 times larger than those used in other studies and incorporated open-source data from more than 50,000 Beat Sabre virtual reality players.
The study showed, according to the press release, that body movements are “as singular and reliable an identifier as fingerprints.” On each player’s motion data, researchers constructed a machine learning categorization model for five minutes.
They found that the model could correctly identify that user with 73% and 94% accuracy in 10 and 100 seconds, respectively.
In the second study, scientists created a virtual reality game that pits players against one another to collect as much information as possible from 50 people in a lab setting in 10 to 20 minutes. Location, age, and height are just a few of the more than 25 characteristics that the authors accurately recognized or inferred.
Due to ethical considerations, the study did not attempt to identify other personal traits like sexual or political orientation, but the researchers think these findings are equally possible.
In order to protect users’ privacy, Nair will then look into preventative measures. She is concerned that criminals may utilize virtual worlds to steal identities or divulge information.
“We’ve done a thorough job of demonstrating that there is a privacy risk here and that it is a different type of privacy risk than what we’ve seen on the web,” he claimed. Our main focus going ahead will be on these methods for either modifying the data or limiting who has access to it. The risks to security and privacy posed here are currently especially important for gamers.
Around 10 million virtual reality headsets were bought last year, according to U.C. Berkeley. About half of the participants in both studies wore headsets like the Valve Index, Quest 2, and other models from Meta Platforms Inc.