Associated Incidents
Fake recordings and visuals were used directly by political parties, but also by unknown actors in the background. They also abused the moratorium.
With the rise of so-called generative artificial intelligence tools, which can create images, videos or audio recordings almost on demand, justified concerns about their misuse in sensitive situations, such as elections emerged.
The early parliamentary elections were the first since the aforementioned tools came into the hands of the wider public. And as many expected, their misuse did indeed occur. However, not on a large scale -- their potential was definitely not fulfilled.
Nevertheless, the last days of the election campaign brought several interesting lessons, findings and warnings for the future. What awaits us perhaps soon? In the article you can read:
- What specific manipulated materials were spread at the end of the election campaign.
- In what format were they spread, what was their content and purpose.
- Which party helped itself with artificial intelligence and abused the voice of political opponents.
- What was the main weakness of the fake materials and, conversely, what was their strongest weapon.
- What are the main lessons, what can we expect in the upcoming elections next year and how the voters themselves can worsen the situation.
"The resemblance is purely coincidental"
The election campaign included a lot of disinformation, but purposefully edited recordings or audio created using artificial intelligence only appeared in its finale.
Just five days before the election, the far-right movement Republika helped itself in this way. It published a pair of promotional spots attacking the LGBTI community and using credible imitations of the voices of President Zuzana Čaputová and Progressive Slovakia leader Michal Šimečka.
"The voices in this spot are fictitious and their resemblance to real people is purely coincidental. But still, vote for Republika," the videos said, admitting that Čaputová and Šimečka never made the claims presented.
Large digital platforms have relatively strict rules for so-called deepfakes, i.e. artificially created or manipulated content. However, they usually turn a blind eye when the author admits that the content is not authentic.
This was also the case with Republika. The company Meta, the operator of Facebook, checked the videos of the far-right movement. It took into account the aforementioned warning that the voices of the people in the video were not real, and also perceived them as satire. This was confirmed to the Živé.sk editorial staff by Rasťo Kužel from the non-governmental organization Memo 98, which was in contact with Meta in connection with the manipulative campaign content.
They went to the limit
More on the edge was the video from the workshop of the extremist media outlet Kulturblog, which published its own manipulated audio recording of President Čaputová a week before the elections.
“After regularly and for a long time watching Kulturblog shows, I decided to express my support for Milan Mazurek in the upcoming elections. Circle number 150 on the Republika candidate list," says the president's fake voice on the recording.
The recording itself does not contain a warning that it is the work of artificial intelligence. Kulturblog only stated this in the text description of the recording, which it published on Telegram. It could potentially have been spread further without the information that it was a fake.
The fact that Kulturblog created the recording in favor of Republika is not at all surprising. Its personal and financial connection to Republika has been known for a long time. Milan Uhrík and Milan Mazurek they even have their own regular shows on Kulturblog.
Don't touch our beer!
However, truly subversive content only appeared on social media as the moratorium approached. On Tuesday, September 26, a direct attack came to Progresívne Slovensko -- a video presenting a fabricated intention of the PS to increase the price of beer by 70 to 100 percent.
A very simple lie with the potential to reach (and influence) a wider audience was seemingly presented by Michal Šimečka himself. "A drastic increase in the price of beer is therefore one of the key points of Progressive Slovakia and it will be one of the first things that the government led by Progressive Slovakia will implement," said the fake voice.
As in the case of the content of Republika, this was a fake audio distributed in the form of a video with illustrative visuals. So we did not see Šimečka speak directly, we only heard him speak against the background of his photo and an illustrative image of half-liter bottles.
The video went viral also thanks to the fact that it was shared by the former candidate of Kotleba's far-right ĽSNS, Rudolf Huliak, who finally on Saturday on the SNS candidate list (he is not a member of it).
The key was timing
On the subject of beer, it was a relatively easy claim to refute -- the PS has nothing similar in its program and Šimečka did not mention anything similar in any of his pre-election interviews.
However, it was not so easy in the case of another attack, which began to spread massively on Thursday, September 28, already during the moratorium.
Again, it was a video containing audio, in the background of which there was a static image in the form of photographs of the "protagonists": Michal Šimečka and Denník N journalist Monika Tódová. The video pretended to be a recording of a phone call between the aforementioned persons. The topic was the manipulation of ballot papers and the reward for the journalist.
Here the situation was much more complicated. At that time, Progressive Slovakia was already placing the latest pre-election polls in the position of a direct challenger to Smer-SD. The fabricated secret recording was intended to discredit the party and possibly its electoral success.
In addition, during the moratorium, not only the candidate parties themselves have limited options, but also the media. They cannot publish anything in connection with the elections -- on the contrary, they must consider every word so as not to violate the moratorium.
This may lead to a delay in the publication of news articles on topic. Moreover, even independent refutation of a hoax takes some time, and in this sensitive period for the media, immediately before election day, the true information may not reach all voters in time.
Such a strategically timed attack can thus achieve the intended purpose and influence a certain part of the population. At the same time, however, it is a lesson for the future for politicians, the media, or authorities and institutions overseeing the integrity of the electoral process.
Low quality? Sometimes it is enough
How many people could the manipulative videos have influenced? It is difficult to estimate. Certainly not as much as would be possible if the dubious actors had used the full potential of artificial intelligence.
Instead, we were treated to recordings of truly poor quality. In particular, the fictitious conversation between Šimeček and Tódová was downright amateurish. The expression sounded extremely unnatural and the intonation did not correspond to the "script" at all. Combined with the shocking revelations and the seemingly frivolous way in which the alleged actors revealed them, it was clear to many that they were not listening to an authentic recording.
However, quite a few people were caught. We are talking about tens of thousands of views and thousands of shares on Facebook alone, not to mention community dissemination via communication platforms such as WhatsApp or Telegram. It was literally a cheap way to convince a part of the voters that the PS is their enemy.
Let's prepare for the next elections
Artificial intelligence can already do wonders today. Including generating credible content based on a lot of publicly available input materials (photos and videos of public figures and samples of their voices). By the way, we wrote about it in a large article back in auguste.
What we saw just before the elections was only a weak sample of what we might encounter next year during the presidential or European Parliament election campaign.
Not only malicious actors, but also the general public will have more and more tools using generative artificial intelligence available, including free ones. Or those that do not have sufficient security limits implemented and thus make it even easier to create malicious content.
This time, disinformers have only tried out the possibilities of AI, were able to evaluate the resulting effect and people's willingness to easily believe even the less convincing fakes.
In just a few months, in addition to audio recordings, we may also encounter fake videos in which politicians appear in situations that never happened. Such deepfakes will be even more believable. After all, people are more likely to believe what they see with their own eyes, so to speak.
Moreover, with the increasing availability of these tools (especially in free form) to the general public, the risk that fake videos will be proactively produced by voters themselves increases. Whether in an attempt to help their candidate or with the intention of harming their opponent.
The possibilities are almost unlimited and the question is not whether this will actually happen, but whether it will happen in next year's elections, or in the one after that.
See the original report for copies of the deepfakes in question.