It is easy to figure out who is hiding behind a user in the metaverse or virtual reality. And this will make it extremely difficult to ensure privacy. Although there have been efforts by the European Union to regulate facial recognition, the technology is far more advanced.
Researchers from the University of Berkeley, RWTH Aachen in Germany, and the business Unanimous AI found that it is feasible to identify someone using virtual reality with a high degree of precision in a study that was published in ArXiv. As? He moved his head and his hands.
Identifier on the Head
Personal identification like fingerprint or facial recognition is biomechanics. This sums up the study’s findings. With the same level of accuracy as other systems like facial recognition or fingerprint recognition, virtual reality uses each person’s distinctive movements to identify each of us.
This discovery has significant ramifications since it suggests that it is feasible to identify the individual who controls many avatars by observing their movements.
Up to 94% accuracy with 100 seconds of movement
55,541 users participated in the study, each entering the recognition model for 5 minutes of data. In other words, this accuracy is consistent with the analysis process used by the researchers, but it is acknowledged that a model that has been properly trained, like the ones used by major technology corporations, would produce even better results.
The end result is that a user may be identified with a 73.20% accuracy after just 10 seconds of movement analysis. a statistic that increases to 94.33% after a review of 100 seconds’ worth of movements. In other words, the person can be identified with a high degree of certainty in a little more than a minute.
‘Beat Saber’ and the Meta Quest 2 as a test bench
It was agreed to use a game and Meta’s glasses for these checks. Particularly “Beat Saber,” a virtual reality game likened to a cross between “Guitar Hero” and “Fruit Ninja,” along with the Meta Quest 2 (Oculus) glasses.
A total of 2.66 million recordings of individuals playing were collected from 713,000 game sessions involving 55,541 different users, totaling 3.96 GB of data. Although some players played just once and others as many as 4,500 times, the average for each of these users was 14. Although some recordings were barely 5 seconds long and others went on for more than an hour, the average recording time was close to 3 minutes. As we can see, a variety of studies have been conducted to gather various data.
What are they looking at?
According to the researchers, these virtual reality game sessions and the metaverse can be used to acquire up to 25 different personality traits. The position of your hands and head, how quickly you react to various stimuli like “blocks,” “walls,” or “bombs,” how quickly and precisely you react to certain obstacles, and last but not least, the metadata of the device from which you connected.
With the new VR glasses, you have access to “extra anthropometric measurements,” which are not available with even the most basic virtual reality systems. For instance, contemporary VR devices like the Pico 4 may display a user’s interpupillary distance because that data is necessary to create images that are suitable for each eye.
This fact was already anticipated (feared)
According to Vivek Nair, one of the study’s authors, “records from the 1970s already suggest that people can detect the movement of their pals.” “We’ve known for a while that movement reveals information about people, but this study recently shows that movement patterns are so particular to an individual that they could serve as biometric identification, just like face or hand recognition or fingerprints. This fundamentally alters the way we view the concept of “privacy” in the metaverse because simply moving about in virtual reality could result in your self-identification.
There is a vast field in front of us, from trap detection to recommendations depending on our movements. Many applications for biomechanics in virtual reality are made possible. These researchers propose that if developed, it could be used to detect cheaters based on their movements, try to predict scores, or even recommend specific maps for each user, depending on his initial moves. It is too new a field, however, and much more in-depth studies would be needed to uncover all the edges.
So, what happens to our privacy?
They can recognize us thanks to biomechanical marks, but that does not mean everything is over. Identification is not a difficult process. Let’s put words on paper. It is simple to identify who is behind it by the drawings or writing, but that is not the reason why it is forbidden. The same is what these researchers suggest for virtual reality.
The answer would involve using gadgets that conceal our movements. In the future, they hope to see work that “intelligently corrupts VR replays to hide identifying qualities, without hampering their original purpose.” Extensions and enhancements that alter the data obtained or the recorded videos fall under this category. The Berkeley team is attempting to construct something similar with MetaGuard, even though it currently appears to be difficult to implement.
Although the landing of the metaverse is not progressing quickly at the moment, this kind of research foresees some of the major difficulties that will arise.