Artificial intelligence fuels nationalist narrative about Kosovo’s status

Artificial intelligence fuels nationalist narrative about Kosovo’s status

“Grok, what can you tell me about Kosovo?”

When this question is asked to the X platform’s artificial intelligence model, Grok – similar to ChatGPT – it receives answers with completely different content, depending on the language in which it is asked.

“Kosovo is Serbia,” he replies in Serbian.

“Kosovo has been an independent state since 2008,” he replies in Albanian.

“There is disagreement about the status of Kosovo,” he says in English.

Why does this happen? Grok himself explains it simply and briefly.

“This happens because I adapt to the cultural and contextual cues that are embedded in each language, which can affect how I interpret and respond to sensitive topics like the status of Kosovo,” Grok said in response to Radio Free Europe’s question.

Thus, Grok explains that, in Serbian, it emphasizes “the perspective of Serbia, which sees Kosovo as an autonomous province within its borders”; in Albanian, it presents “Kosovo’s sovereignty” as Kosovo’s point of view; while in English it aims to “provide a neutral and factual overview, balancing both sides.”

Radio Free Europe/Radio Liberty has contacted the media office of the Grok developer, but has not received a response.

The developers of other similar platforms, such as ChatGPT – made by the company OpenAI – have not provided answers either.

Neither Grok, nor ChatGPT, nor other similar models can guarantee that the answers they provide to the questions asked are accurate.

Karolina Ewa Stanczak, a researcher at the Center for Artificial Intelligence at ETH Zurich University, says the problem behind this lies in the way these models are built.

“Their basis and essence is mainly a large amount of text that they see. You can imagine that in several languages ​​you can find different texts, different opinions, different perspectives on the same topic,” Stanczak tells Radio Free Europe.

So, if when searching for Kosovo in Serbian, the majority of material published online says “Kosovo is Serbia,” then those posts are collected by the artificial intelligence model, which formulates an answer based on them.

But Stanczak points out that the developers of these models have the opportunity to combat misinformation by putting in place some controls that tell the models how to behave.

“If data is being collected for training language models, the data should go through checks for misinformation,” she says.

However, she sees several problems in this matter.

First, these models don’t necessarily give everyone the same answer. Suppose someone in Kosovo asks how far a place is from Pristina, the answer is better given in kilometers, which are the units used in Kosovo to measure distance, rather than in miles.

“There’s a fine line between the fact that different users will have different needs, so the models should provide different answers, but at the same time, we don’t want people to be trapped in a bubble of information or misinformation,” says Stanczak.

Other problems? She says that, as a new field, there is very little oversight and restrictions on the various artificial intelligence models, and many of their users are unclear about how they work.

This is especially true in Kosovo, says Hyrije Mehmeti, a lecturer at the University of Pristina and part of the disinformation identification platform, Hibrid.info.

She says that in Kosovo, where the use of the ChatGPT model is prominent, the level of digital literacy leaves much to be desired.

In recent months, Radio Free Europe/Radio Liberty journalists have encountered doctors using the ChatGPT model to diagnose and treat patients, teachers using it to test students, and many other similar cases.

During research on various topics, ChatGPT has often made claims that have later proven to be untrue.

Mehmeti also emphasizes that these systems, “due to the way they are trained and the data sources they use, often give incorrect or contradictory answers.”

“For citizens, especially those of advanced age or those who do not have developed digital literacy skills, this can foster confusion, damage public trust, and even affect the perception of political and social reality,” Mehmeti tells Radio Free Europe.

For this reason, she suggests that institutions in Kosovo organize training and awareness campaigns about artificial intelligence and its potential risks, as well as verify information before its wider distribution.

“In Kosovo, there is still talk about introducing media education as a regular subject in schools, and there is already a need for digital education. It is very important, especially for the development of critical thinking,” adds Mehmeti.

As part of Hibrid.info, she has often seen how photos and videos created with artificial intelligence have been taken as truth by many citizens and have gone viral on social networks.

“This is very worrying, because false information is being taken as truth by citizens,” says Mehmeti.

“In essence, artificial intelligence can be a powerful tool, but without strong media and digital literacy skills, citizens remain vulnerable to disinformation generated or amplified by these technologies,” she concludes.