Artificial intelligence creations are becoming more and more sophisticated, and the time will soon come when we will no longer be able to distinguish authentic records of real events (e.g. photographs and videos) from AI-generated products. AI will thus dramatically change the way we access information – photos and videos will no longer be a guarantee of authenticity, as they may (or may not) be artificially created by AI.
Consequently, increasing importance will be laid on the media literacy of online media users and their ability to critically evaluate, question, and verify information – however in the real world we cannot verify everything (we would do nothing else). There will also be an increase in the importance of specific sources of information, such as specific individuals or organisations, whose information can be assumed to be of high quality and unbiased. Either way, we have to gradually get used to the fact that what we see and hear on the internet may not reflect reality at all.
The problem is that before the majority of the population gets used to this paradigm shift (which may take several years), AI will cause many problems. One of them will be the targeted manipulation of elections as one of the basic democratic processes. Using AI, it will be possible to create and disseminate on a large scale over the internet videos and audio recordings of politicians saying absolutely anything. Political marketing will use AI in the mass media to damage political opponents and improve the image of their own candidates, and the supporters or opponents of individual political opponents will also be active in this area.
This is already happening, as can be seen in the case of the fraudulent recordings disseminated during the election campaign in Slovakia (which we wrote about here), and also current cases from the US presidential election campaign (2024). In the USA, faked photographs of Donald Trump surrounded by African-Americans to encourage black voters to vote for him have been widely shared. However, these are fakes created probably by the politician’s own supporters. At first glance, the materials seem very plausible, but closer examination reveals too shiny skin, missing fingers, and various redundant fragments.
Alas, it is extremely easy to create such materials, both in the form of photos and videos. Josef Šlerka and his team have recently published an experiment with a deepfake video of former Czech President Václav Klaus saying things about global climate change that he would probably never say in reality. The post attracted a lot of attention, with many users actually being tricked, commenting on the video and not recognising that it was a hoax. Sadly, a large number of internet users do not realise that similar videos can now be produced by anyone who learns to work with artificial intelligence tools. And they can be produced in a matter of minutes. The range of potential ways how these creations could be misused is extremely wide.
Deepfake videos have so far been used mainly in scams, such as the classic cases of fraudulent investments, in which deepfake videos of famous politicians making promises of miraculous gains emerged. Andrej Babiš, Michael Žantovský, Daniel Beneš, even President Petr Pavel and other important Czech figures have been targeted by fraudsters; in other cases, the logos of ČEZ Group, the Office of the Government of the Czech Republic, and CNN Prima News have been misused. You can read more about this type of fraud here.
However, it can be safely assumed that in the near future, AI products will be used to deceive voters in elections on a massive scale – as a tool of actively influencing their moods and preferences. Imagine that a well-known politician (such as the President or the Prime Minister) would call you and try to persuade you to vote for them. Or that, on the contrary, they would convince you not to vote at all. Does that sound like science fiction? As a matter of fact, this is exactly what has been happening in the USA, where Joe Biden’s artificial voice was used to discourage Democratic voters from taking part in the primaries (more on CNN).
Similar scenarios can be expected in other countries as well, and the European Union has warned of the risks of interference in elections, urging major tech platforms X, TikTok, and Facebook to identify and label AI-generated content. The UK and other European countries are also concerned about the manipulation of elections through deepfake videos.
What do we need to be prepared for before the elections in the Czech Republic? For the flood of deepfake videos of Czech politicians and celebrities that will have an impact on voters.
It is highly probable that a wave of deepfake videos will also appear in the Czech Republic before the elections. It has been documented in experiments that a large part of social media users cannot identify deepfake videos and debunk them as fake; they perceive them as authentic and capturing reality.
Deepfake video with Czech politician Tomio Okamura (more videos here)
Deepfake video with Czech politician Alena Schillerová (received via WhatsApp)
The above videos of politicians are very simple, obviously created from source photos found on the internet. It only takes a few minutes to create a deepfake video – all that is needed for its production is a voice sample (a few minutes of speech) and an image of the person (either an original photo or a video). In the next step, you just open certain online applications (no need to mention their names, they are publicly available) and create a simple or more advanced video. Deepfake videos circulating on the internet are often amateurish, but sometimes more sophisticated and realistic creations using more complex algorithms can be found.
A good example is a recent video of Andrej Babiš “apologising” for his e-mail efforts to find compromising materials on Minister Lipavský. The use of AI in this video is much more advanced, and it can be very difficult for a layman to recognise that it is a deepfake.
Deepfake video with Andrej Babiš circulating on Facebook
Several recommendations concerning deepfake videos
A. Check the source: Always find out where the video comes from. Trustworthy sources can be assumed to share verified information. If the video comes from an unknown or questionable source, use caution.
B. Focus on quality: Deepfake videos often contain visual or audio flaws. Look for flaws that could indicate manipulation such as unusual lip movements, suspicious shadows, poor intonation, overly perfect language, etc.
C. Verify the information: If the video appears suspicious, try to find other sources to confirm or refute the information. If you cannot find any additional information on the issue, be cautious. You can also use various apps that can help identify deepfake videos.
D. Educate yourself: The more you learn about deepfake technologies and their methods/functioning, the better you will be able to identify them. Keep up to date with the latest developments in artificial intelligence and media literacy. Tools such as E-Safety’s AI course could be of help.
E. Watch out for videos triggering emotions: If the video tries to arouse a strong emotional reaction or seems to be too shocking or controversial, this could be a warning sign that it may be a hoax. Disinformation and hoaxes often appeal to emotions, aiming to cloud rational judgement and cause users to react hastily.
F. Learn to respond: If you come across a deepfake video, do not spread it further. Instead, inform the relevant platforms (social networks) and warn others. Strengthening the community against the spread of misinformation is key.
G. Actively flag sources spreading false information and support credible institutions.
For E-Safety
Prof Kamil Kopecký
Palacký University Olomouc