Brain rot, clickbait, fake users, political PR bots, online panic, and Reddit as an “investment advisor” – these are just some of the topics tackled by students of the Artificial Intelligence programme. They presented their posters during this year’s Open D.AI event.

AI students presented the results of their work prepared as part of the Social Media Analysis course, based on analysing millions of data points from YouTube, Reddit, Google Maps, platform X, and OpenReview. Their projects showed that digital media influence our brains, emotions, and decisions.

One of the analysed phenomena was the so-called brainrot. EEG studies showed that highly stimulating content – chaotic, loud, and fast – caused the brain to enter a state of “shallow processing”. The reward system was stimulated, while the visual cortex was suppressed. Content was not memorised, but only “scanned”. The project also produced a prototype plugin that warns users about brainrot in real time.

Another study focused on political communication on the internet. An analysis of 46,000 posts from the X platform revealed that party accounts were characterised by a “robotic” style. Politicians from one party turned out to be stylistically almost identical to the party account, while a more chaotic style of expression was associated with greater authenticity. The study suggested that many politicians do not write their own posts.

Some projects focused on the academic environment. An analysis of reviews for the ICLR and NeurIPS conferences showed an increase in linguistic complexity, unusual gaps in scores – such as the absence of sevens – and a style increasingly resembling texts generated by large language models. This could indicate that some scientific reviews may have been written by AI.

Contrary to the “dead internet” theory, only about 4% of Reddit content was generated by AI. At the same time, the share was 20.6% for posts and only 3.5% for comments.

Analysis of Google Maps data showed that the social “centre” of Warsaw did not correspond with the administrative boundaries of the city. Instead, it was defined primarily by gastronomy, entertainment, and tourism. Differences in ratings were also observed: Polish users tended to be more critical, while foreigners were more enthusiastic.

Research on clickbait demonstrated that the most effective detection method involved analysing both images and text simultaneously. The multimodal BridgeTower model proved significantly more effective than analysing titles or thumbnails separately. Some videos appeared “innocent” on their own but became clickbait only when both elements were combined.

The final project presented was the Fear Index – a measure of social panic. It correlated with real-world events such as wars or crises, while also showing that memes can “disarm” fear.

The project presentations took place on 2 February at the Student Culture Zone as part of this year’s Open D.AI event. The poster session was part of an open meeting for candidates applying to the Artificial Intelligence second-cycle programme. During the event, the programme and its teaching staff were presented. Students talked about other projects developed during their studies, while graduates shared their experiences and career paths after graduation.

The students’ work will be available for viewing from 10 January in building C-5.