Did you know that Cwmbran has the most roundabouts per square kilometre? 

It even got recognised by the Guinness World Records. 

At least that is what Google’s A.I. tool is trying to make you believe and all of that is just based on one article written by Ben Black, as an April Fools’ joke, as BBC reports.  

Even though the article is marked as an April Fools’ joke, A.I. is taking it literally, it being known for having a notoriously rocky relationship with jokes and sarcasm. 

Humans design the construction of A.I. and a training algorithm by which the A.I. is supposed to learn through analyzing massive datasets and mathematical patterns how to generate answers. But not even A.I.-labs really understand how it is connecting all this data and come to conclusions, showing a huge flaw: the Black Box problem. 

This becomes especially problematic when solving mistakes in A.I. like A.I. hallucinations.

A.I. hallucinations are factually incorrect or completely fabricated outputs. These errors occur because these systems generate text by predicting statistical patterns in their training data without actively fact-checking or being able to comprehend the real world. 

Since they often sound plausible, they can lead to a wide spread of misinformation easily and fast. 

When you are looking something up on google you first get a nice A.I. summarization, let’s be honest: how often do we actually fact check those? That is where media literacy comes in handy, the ability to critically access, analyse and evaluate any form of information we receive. Especially when considering that A.I. just takes what it can find in its datasets no matter if they are facts or stereotypes. This is worrying organisations like Reporting Diversity Network, a group of civil society and media organisations, established with the goal of promoting diverse media representation across the Western Balkans. Executive Director, Ivana Jelaca stated about how A.I. generated content is distributing harmful narratives that:

“AI relies on what humans have already produced, and there is sexism, misogyny, GBV [gender based violence], homophobia or transphobia out there. It is easy to check how likely AI is to reproduce prejudices and stereotypes, unless one is really precise in giving prompts that will avoid using those, which implies that we are well MIL literate and sensitised to the issue of stereotypes and hate speech. AI is so young, but it develops faster than we can learn how to use it, so we are likely to see this still being perpetuated.“

She mentions that it is also on us how we consume media. A skill despite getting more and more important is left behind, media literacy, the ability to critically access, analyse and evaluate any form of information we receive. Actually we can even see a decline in critical thinking skills related to A.I., specifically with the use of Large Language Models (LLM) like ChatGPT. 

A study from researchers at MIT’s Media Lab searched for the cognitive cost of using LLMs. They divided 54 participants, consisting of 18 to 39 year-olds, into three groups. They were requested to write an essay where each participant depending on the group used a Large Language Model, a Search Engine or just their brain. During this process the brain activity of the participants got measured with an EEG. This revealed that the group using LLM exhibited the lowest brain activity and they performed worse in neural, linguistic, and behavioral levels. After several months LLM users got lazier per essay and often ended up in just copying and pasting the given answers. But the sample size of the study is quite small and yet to be peer checked. 

Like everything, media literacy can be learned but rarely gets teached at schools. Becoming obvious in the Media literacy index from 2026 published by the Open Society Institute – Sofia, where Macedonia is ranked on the penultimate place before Kosovo.

Nevertheless, there are simple things we can do to improve our way of analyzing news, for example by questioning our media experience more often. 

Platforms like NAMLE, a leading organisation in the US for advancing media literacy education, lists key questions on their website to help review the media consumed on their website. Questions like “Who made this?”,  “Why was this made?” and “Who paid for this?” are simple but help understanding the intentions behind the media published.

“However, it is not all about education. We have to recognise that some tools are problematic in their making. Grok was an example of a tool that one can use to generate a close to naked image of whoever one wants. Such tools help GBV thrive online, and so little was done to prevent this.“  Ivana also differentiates and brings up the topic of A.I. generated images. Seeming harmless at first especially deepfakes can turn dangerous easily. Those are A.I. generated or manipulated images, videos or audios based on existing people.
That can be you editing yourself in your favorite video game for fun but also pornographic videos generated of people non consensually. It is estimated that in 2023 of 100.000 deepfake videos that were published on the internet, 96% were deepfake pornography.

A huge scandal now erupted in Germany as well after TV Star Collien Fernandes accused her ex husband of spreading non consensual deepfake pornography online, showing loopholes in German laws that lead to most deepfake publishers never facing any consequences. But this is nothing new, in 2017 a reddit user called “deepfake” was publishing pornos with the faces of actresses and a Deep Learning Algorithm on how to create such videos yourself easily as well. Before 2015, Noelle Martin found pictures of herself edited into nudes with the help of photoshop on accident through reverse image search. After suing did not lead to anything, she became a leading activist for criminalising image based sexual abuse. Specifically her case shows those videos are not just of public figures, one picture is already enough to create a deepfake video of someone with the advanced technology we have today.

Despite this the main fear in social debates about deepfakes is political manipulation and disinformation being spread easily and undetected by most people specifically with elections. This leads to a general distrust of the media which is becoming obvious in the annual Edelmann Trust Barometer as well. It shows that in the 28 participating countries the fear of leaders lying to us is at an all time high in its 25 years of the inception of this survey.
In particular for journalists and reporters the percentage of people who worry about disinformation rose from 59% to 70% from the year 2021 to 2025.
Specifically A.I. has a huge impact in the spread of misinformation since it lowers the time and skills necessary to create and spread convincing lies drastically.
The malicious misuse of generative tools got obvious through the example of deepfakes but also flooding the internet with convincing fake articles can erode the trust in verifiable facts and be used for manipulation. Which is especially dangerous in politically uncertain times where people tend to fall for polarisation easier. That would mean that proper journalism should be sustained and encouraged, instead we see that the situation for journalists worldwide has deteriorated again year after year. As the World Press Freedom Index from Reporters Without Borders (RSF) shows, it is the first time in its 25-year history that more than half of all countries fall in the two worst categories, an alarming development.

Despite all of that fearing and demonizing A.I. now will not help either. More so we need to be aware of its unreliability and question its output.
With that improving our way of handling news critically becomes increasingly important especially with the fast improvement of making it harder to divide A.I. generated content from fact checked ones. So next time we read the news, let’s ask ourselves “Who made this?”,  “Why was this made?” and “Who paid for this?”

Nelli Winzer