Psychologists from Lancaster University analyzed smartphone data of a group of seven hundred people. They calculated 4.5 thousand days of app data into usage models. Each of these days, people were paired with each other (among the 780 users). In such a way, the AI model learns these people's daily app usage patterns.
The researchers then find whether the AI model could identify a person or not. It was given only a single day of smartphone activity by anonymous people and never paired with any user.
Researchers from the University of Bath said their AI models that got trained in only six working days track the app usage data of each person. It could correctly identify the accurate person from just one day of anonymous data. The anonymous data was only one-third of the total data.
It might not sound good, but when these AI models predicted the ownership of a given data, they also provided a list of the most likely candidates. It was possible to get a top ten list of most likely candidates to that specific data belonged. Around seventy-five percent time, the chosen user is inside the top-ten list of most likely candidates.
Therefore, it is always compulsory to know that one's app usage data, collected by a phone (automatically), can reveal a person's real identity.
While having new opportunities for law enforcement in this sector, it can also risk one's privacy if such data gets misused.