What can't artificial intelligence do?

2026-07-18 17:03:00, Tech CNA

"AI can calculate, but humans have to make the final decision," says Maximilian Kiener from the Institute for Ethics in Technology at Hamburg University of Technology.

Whether it's career decision-making assistance or a substitute therapist, artificial intelligence (AI) seems to have the right answer to almost every problem.

AI developers almost always design these virtual assistants to be kind and non-judgmental.

But this is just a dangerous illusion that users too often believe, says Maximilian Kiener.

"What's not working are the value judgments. What career path should I pursue? Should I become a lawyer, should I become a doctor? There's no single answer. The problem that keeps coming up is that we want to compress all the complex problems into an AI format," he says.

Topics like loneliness or education are often social issues. This is not a task that AI can technically solve. Maximilian Kiener does not see any innovation in some new AI applications.

"Things that are simply new are not innovations. For example, I have a very hard time identifying fake images of children that could be described as innovations."

According to the scientist, ethics does not hinder the development of artificial intelligence, but only makes it truly useful for people. It is optimal when AI and humans work together from the very beginning to find solutions together, says Kiener.

Matthias Uhl, professor of business and social ethics at the University of Hohenheim, developed a Socratic chatbot: " Socratic AI differs from conventional AI in that it asks questions and helps me find my own solution to the problem, rather than giving me direct answers."

The team from the University of Hohenheim and the University of Bamberg tested this language model in several experiments. The result: The adapted chatbot was able to help users make more morally balanced decisions.

But the team also discovered a hurdle during implementation, says Matthias Uhl. "AI that asks additional questions is chosen less often. When people have a choice, they prefer to get direct advice."

While private users often trust AI immediately, experts are often more skeptical. For example in medicine: AI systems can analyze X-rays reliably and quickly. However, many doctors have difficulty accepting the technology.

Those who know too much dispute AI analyses, even when they are more accurate - scientists have discovered. New AI can help, believes Matthias Uhl.

"If the system itself shows its uncertainty, this can lead to greater user trust in the advice. For example, when AI says it is 90 percent confident in its classification."

Medical experts are expected to be able to test the new AI system as early as next year./DW