CZECHS WANT CERTAINTY ABOVE ALL ELSE FROM AI. SIX OUT OF TEN USERS HAVE DOUBTS ABOUT THE ACCURACY OF THE OUTPUTS

Around two-thirds of adult Czechs currently use artificial intelligence. Although most of them have a positive attitude towards it, the question remains as to how much they can trust its answers. According to a survey by the STEM/MARK agency, 60 per cent of AI users are unsure whether the output they receive from the tool is actually correct. According to the survey results, users would therefore appreciate it if AI could distinguish more clearly between verified information, estimates and situations where it is simply unsure.

According to the research, 85 per cent of AI users have a positive attitude towards it. They also use it more frequently and for a wider range of tasks, such as searching for information, obtaining practical advice, translations, generating images or working with technical information. Conversely, people with a negative attitude towards AI use it less frequently and for a narrower range of activities.

Trust is not just a matter of satisfaction

Doubts about the accuracy of answers are not, however, a problem exclusive to those who have a negative attitude towards artificial intelligence. 61 per cent of users with a positive attitude towards AI lack confidence that they can trust the AI’s output. Among users with a negative attitude, the figure is 52 per cent.

The link between trust and satisfaction is also interesting. The need for certainty regarding accuracy is cited more frequently by people who are satisfied with AI outputs. The answer may therefore be useful and well-formulated for the user, yet they may still be unsure whether the information it contains is correct. According to STEM/MARK, satisfaction decreases more significantly only when doubts about credibility are combined with the use of out-of-date data.

Users want to know where the answer comes from

The survey results also point to a desire for greater transparency. According to STEM/MARK, it would help users, for example, to be able to distinguish whether the AI has taken the information from a source, calculated it, derived it from available data, or whether it is general knowledge or merely an estimate. It is also important to know the time period to which the answer relates and how up-to-date the data used is.

It is precisely the ability to acknowledge uncertainty that may be one way of strengthening trust in AI. However, estimating uncertainty is not the same as verifying truthfulness. Language models are primarily trained to generate responses that match the prompt and are useful and plausible to the user. This does not, in itself, mean that they are factually correct.

Critical assessment of the outputs therefore remains, for the time being, the responsibility of the person using the AI. Verifying information and the ability to recognise potential errors therefore remain an important part of working with artificial intelligence tools.

The STEM/MARK survey was conducted in July 2026 via an online questionnaire on the National Sample panel. It involved 506 participants representing the internet population aged 18 and over, stratified by gender, age, education, region and size of place of residence.

Source: mam.cz