Nearly three-fifths of respondents in the most recent wave of the CEDMO Trends survey could not correctly identify a “deepfake” as a seemingly real, artificial intelligence-created image, sound or video that is not based on truth. Public awareness of the principles of generative AI language models is also low. These facts emerged from the eleventh wave of measurements of a representative longitudinal panel survey commissioned by CEDMO.
For the question asking for knowledge of “deepfake”, only 41% of respondents selected the correct answer out of four offered, with more than half indicating they did not know and the remaining tenth indicating a wrong answer. At the same time, more than a third of people (35%) said they had encountered deepfake content in recent months.
“The correct option was statistically significantly more frequent for men than for women,” comments Lukáš Kutil, data analyst at CEDMO Hub. “In the age categories, this was true for all generations from the youngest to 44 years old. On average, they answered correctly 59% of the time. In contrast, the older year groups over the age of 45 answered correctly, on average, only 29% of the time. In terms of education, the deepfake can be defined correctly by university-educated respondents. In terms of region, it is people from Prague and Central Bohemia. From the economic point of view, apart from students, they are also self-employed, and in terms of political preferences, they are most often voters of the PirSTAN coalition but also of the Trikolora + Svobodné + Soukromníky coalition, which is interesting because these are voters of two coalitions whose voter base does not typically overlap. As a common denominator, we can look for the inclusion of oneself on the liberal side of the political spectrum, as the correct answers were significantly more common among liberal voters. Voters of the SPOLU coalition are also close behind,” Lukáš Kutil continues.
The fake video of Minister of Interior Vít Rakušan, which went viral on social media in the second half of January in connection with his discussion with citizens in Karviná, was seen by about a third of the population. It was considered authentic by 17% of respondents. This was one of the disinformation that most attracted the attention of respondents in the studied period.
63% of respondents expressed concern that politicians and political parties will start using false AI-generated videos to smear their political competitors on a large scale. More than three-quarters (79%) also believe that people will stop trusting real videos, and around the same percentage (78%) agree that social media and internet operators should delete AI-generated false videos.
“The share of disinformation generated by artificial intelligence in the total number of false news circulating in the public space is still relatively low,” says Petr Gongala, coordinator of the fact-checking platform Demagog.cz, which verifies daily the statements of politicians in the media and the veracity of widely shared content on social media in the Czech Republic. At the same time, Demagog.cz cooperates with dozens of partner fact-checking organisations in the pan-European EDMO network, of which CEDMO is a part. “In the same period when the fake video of the Minister of Interior Rakušan was circulating on social media, the deepfake technique was used for only 4% of the false reports out of nearly fifteen hundred that we and 34 other partner factchecking organisations of the EDMO network verified during January,” Gongala adds.
In a pan-European context, deepfake videos which circulated during January mostly featured alleged footage of protesting farmers, generated images of US politicians (e.g. Nikki Haley in jail or Donald Trump in the company of Epstein) or modified videos showing the Eiffel Tower on fire.
The research agency Median, which conducts the CEDMO Trends survey for the CEDMO hub, also investigated the extent to which the Czech population is aware of the principles of language models such as ChatGPT and to what extent these artificial intelligence tools are used.
Only less than a third of respondents demonstrated knowledge of the principles behind how large language models provide users with answers to their questions. Almost half admitted to not knowing, and the remaining respondents chose incorrect answers.
Generative AI tools are not used by the vast majority of Czech society. Only 1% of respondents use them at least once or more a day, and 5% use them more than once a week. Up to 74% of respondents declare that they do not use generative AI tools at all. Age emerges as a significant factor in relation to the use of generative AI. Specifically, there was an increasing trend in respondents’ answers, where the frequency of the answer “I do not use generative AI at all” increases along with increasing age. While up to 72% of respondents in the younger age group up to 24 years old use generative AI tools, this is only 42% in the 25-34 population, 29% in the 35-44 age group, 22% in the 45-54 age group, 12% in the 55-64 age group, and only 7% in the 65+ senior population.
CEDMO Trends ČR is a unique longitudinal panel research conducted over a period of thirty months. It offers exceptional insight into the evolution of the population’s behavior in the consumption of different types of media content, focusing on particular types of information disorders such as misinformation and disinformation. These not only undermine public trust in the institutions necessary for the functioning of a pluralist democracy but can also amplify individual infodemias. It is being carried out for CEDMO (Central European Digital Media Observatory) by the research agency Median on a representative sample of more than 3,000 respondents over the age of 16.
The CEDMO TRENDS research is funded in the Czech Republic within the framework of the project 1.4 CEDMO 1 – Z220312000000 Support for increasing the impact, innovation and sustainability of CEDMO in the Czech Republic. This project is funded by the EU’s Recovery and Resilience Facility (RRF).
