In the program “Breitband” on Deutschlandfunk Kultur, Hendrik Heuer, senior professor of the research program “Trustworthy Artificial Intelligence” at the Center for Advanced Internet Studies (CAIS), assesses the release of ChatGPT-5 and the controversial reactions to it. In the interview, Hendrik Heuer first points out the limited significance of common benchmark tests, in which the new model achieves very good results. Although such standardized comparative tests allow performance evaluations between AI models, they provide only limited reliable information for practical use. Since AI systems are trained with large amounts of internet data and often tested with tasks whose solutions are also available online, there is a risk of reproducing familiar content rather than demonstrating real problem-solving skills. Hendrik Heuer therefore advocates independent testing and research by external actors, such as journalists, comparable to testing mechanisms in the automotive industry.
In addition, Prof. Heuer addresses the debate surrounding the emotionality of ChatGPT. Many users missed the “friendly” or “personal” tone of previous versions in the new model. These reactions showed how strongly AI systems can create emotional bonds and trust. In Heuer’s view, OpenAI’s decision to initially scale back these effects and later partially revise them misses the real problem. Instead of merely relativizing risks after the fact, AI systems must be designed responsibly from the outset.
Findings from the field of human-computer interaction research show that user-centered design decisions have a significant influence on how AI is perceived and used. Trust should not arise by chance, but must be consciously and responsibly considered.