Walter Quattrociocchi

Two research teams, with some authors in common, have shown in two separate studies that interaction with chatbots using artificial intelligence (AI) can significantly change a voter's opinion about a presidential candidate or a policy proposal. One of the studies, published in Nature, was conducted in three countries (the US, Canada, and Poland), while the other, developed in the UK, is published in Science. Both studies reach the same conclusion: the persuasive power of these tools stems less from psychological manipulation than from the accumulation of fact-based claims that support their position. However, this information is not always accurate, and the greater the persuasive power, the greater the inaccuracy and fabrication.

A US research team has devised a method that uses a browser extension to alter the algorithm of X (formerly Twitter) to study its impact on user behaviour. In a 10-day experiment with 1,256 volunteers during the 2024 US presidential campaign, they used the method to vary the content expressing anti-democratic attitudes and partisan hostility. According to the authors, the results—published in Science—provide causal evidence that greater or lesser exposure to this type of content alters polarisation in the same way. Their conclusions contradict previous research published in the same journal, which found no such relationship on Facebook or Instagram. In that case, the study was conducted in collaboration with and funded by Meta.

A team of researchers from the University of California (USA) has analyzed the presence of hate speech on the social network X (formerly Twitter) since its purchase by Elon Musk in October 2022 until June 2023. Their findings are that this type of racist, homophobic and transphobic speech increased by approximately 50 % throughout this period. In addition, the presence of bots and fake accounts did not decrease, contrary to Musk's own promises. The results are published in the journal Plos One. 

A study led by researchers from Northeastern University and Stanford University in Boston (USA) has analysed the source of exposure to partisan or unreliable news when searching on Google. After tracking the information consumption of approximately 1,000 people in the 2018 and 2020 US election periods, their conclusions are that such exposure is determined more by users' own active search than by the content displayed by the search engine's algorithm. The results are published in the journal Nature.