SCIENTISTS have called for better privacy measures after they found that artificial intelligence (AI) can predict the views of social media users even if they never post online – and could leave people open to being targeted by fake news.

Researchers said that people’s opinions on politics, religion and other topics likely to provoke debate can be predicted, even if they never post anything about them. Their study suggested that the accounts Twitter users follow and what they like are more accurate predictors of their views.

Computer scientists from the University of Edinburgh said their findings highlighted the need for improved privacy measures to prevent what is publicly available data being used to deduce people’s personal views. Having access to this data could enable malicious users to target people with fake information about contentious subjects.

The Edinburgh team examined more than 2000 public Twitter accounts to show how social media data can reveal a person’s views on issues including atheism, feminism and climate change.

They found that people’s networks and the way they engage with content provided a better gauge of their views than existing methods, which assess the text of users’ own posts.

The most accurate predictions were made by combining the two approaches which, said the team, achieved a success rate of almost 75%.

Their paper – Your Stance is Exposed! Analysing Possible Factors For Stance Detection On Social Media – will be presented in November at the Association For Computing Machinery (ACM) Conference On Computer-Supported Co-operative Work And Social Computing in Austin, Texas.

The team worked with three network-based models using interactions, preferences and connections, which all “outperformed the state-of-the-art methods that depend on textual features only”.

They wrote: “We also presented an analysis of the top features to identify the correlation between stance and topic with respect to the features groups. We explored with the key important features that have a positive effect on detecting stance for each target. The results denote accurate learning of the stances at the user-level representation that improves the content-related features model.”

The team said their new approach meant that for the first time the views of people who rarely – or never – post on social media, the so-called “silent users”, can be accurately predicted.

Dr Walid Magdy, of the university’s School of Informatics, who led the study, said: “Social media users are highly vulnerable to having their personal views predicted, without them even discussing the topics online. This shows the power of artificial intelligence when it is applied to big data.”

Co-author Abeer Aldayel, from the same school, added: “Our study highlights a need to develop regulations and counter algorithms to preserve the privacy of social media users.”