Uhlík vs Křemík · Czech podcast about artificial intelligence

Why we trust artificial intelligence more than we should

Episode 5 · 17 March 2026 · 1:07:47 · Guest: Kateřina Lesch

Kateřina Lesch is responsible for AI transformation and adoption across six countries at the Dr.Max pharmacy chain. On the Uhlík vs Křemík podcast, she talked with hosts David Salomon and Adam Hanka about how quickly people get used to artificial intelligence and stop checking it. The conversation also covered rules for companies, the emotions of artificial intelligence and raising children.

From caution to blind trust

When David Salomon asked how the public is handling AI, Kateřina Lesch answered bluntly: "I am afraid it is losing its self-reflection, that it has started to believe it." She described an earlier project in which AI generated discharge reports for doctors at a spa. At first the doctors insisted that no computer would write a report better than them, then they stopped reading the reports. A button had to be added to the system so that the doctor confirms having read the generated text.

She recently made a mistake herself. In the autumn she had AI plan a hike across the Krkonoše mountains, including the buses home, but the connection from Malá Úpa belonged to the summer timetable and was no longer running. She had not checked it and had to call a taxi. Worse, in her view, is when executives of large companies let an agent loose on their profit and loss statement and never find out that an SQL query skipped a column somewhere.

Adam Hanka added an explanation: models hallucinate less today, so people have dropped their defence mechanism. The mistake that eventually comes is all the more costly. Kateřina Lesch added that without clean and reliable data the output looks very convincing and the error is much harder to find.

Rules first, agents second

The individual countries at Dr.Max are fairly independent and headquarters helps them. According to Kateřina Lesch, how far along they are with AI depends on what the local IT directors currently have on their plate, and also on regulation. The Serbs, who are not in the EU, do not keep GDPR so much in mind, while the other countries have to deal with it and also address the European AI Act.

That is why they work on AI governance, a set of binding principles. They state what data employees may upload into AI tools and which tools they may use, so that data does not leave the European Union and the company keeps control over it. In her view it makes no sense right now to impose a single AI platform on all countries, because within a month another tool could be more efficient. It is easier to have shared rules and an operating model above the tools.

Whether a company is ready for AI depends, in her view, on data and mindset. It helps to show people quick wins, but above all to teach them to define a business case that brings real value: "building an agent for two days that then saves one person half a minute a week does not always pay off." For employees she relies on Copilot, because the data stays in a secure company environment. She recognises a good consultant by the fact that they ask about the specific use case, do an assessment in advance and train people from finance differently from people in marketing.

It can trigger emotions, not feel them

Kateřina Lesch studied emotions in her dissertation, where she compared cultural differences using film reviews: Americans wrote that a film was the best in the world and gave it two stars, Scandinavians wrote that it was quite all right and gave it five. When she worked on recognising emotions in call centres, not even the operators could agree on when a caller was angry. Emotions, she says, are individual and are also physiological reactions that AI will never experience. In her view AI only triggers them, so people should be wary of the illusion that it truly empathises with them.

David Salomon mentioned a trend according to which people increasingly use AI as a friend or therapist. According to Kateřina Lesch, this points to loneliness and shyness, because talking to a box is easier than talking to a person. Adam Hanka sees it differently: a new group of users with different needs has come to AI.

Children, over-adoption and AI as a coach

Kateřina Lesch mainly teaches her young children to name their own emotions. She does not expose them to AI yet, because she wants them to experience a campfire on the banks of the Sázava river before generated colouring books. She does not want to put them at a disadvantage in the digital world, though, as it belongs among the skills they will need. She warns against over-adoption, when people automate everything including close relationships.

She also sees the good sides of AI, from help with routine work to drug development. She currently uses it most as a coach. The system prompt of her agent contains the results of her personality tests and a description of her position, and before a meeting she has it pick her presentation apart. At the level where she works, she says, she has few opportunities to be mentored, because there are very few women there.

You can listen to the whole conversation on the Uhlík vs Křemík podcast.

The episode is in Czech. Czech page with chapters, quotes and the full transcript