The way AI models behave depends on what they are trained on. “They are very much like us, as they operate on the basis of our data. If we had trained them on the basis of literature about mafia, they could start behaving as criminals. We should be more aware of the fact that one can build these models for better or for worse.”
However, AI that behaves as humans can offer an opportunity to make technology more accessible, as we communicate with it in a natural way – through language, but also through mimics and gesticulation. In his research Koen Hindriks used a Nao robot as tutor for a pupil. It is crucial to be able to make a distinction: “Everyone can see it’s a robot, I’m not trying to make people believe they talk to a human.”
Chat-GPT and co. on the contrary seem to resemble humans to such an extent that many people find a romantic partner in AI. Therefore the research will first start analyzing and understanding the social behaviour of the AI.
Read more (in German).