00:00Lying is human. Most of the time, our guilty conscience stops us from doing so. But what,
00:09What if we don’t have to lie ourselves, but an artificial intelligence does it for us?
00:14The Max Planck Institute for Human Development in Berlin, among others, knows. How was this investigated?
00:208000 subjects had to solve tasks with the help of AI, such as a dice game in which they
00:26Cheating could get more money. The surprising result: Without AI, almost all
00:33honest. But the more the AI was used, the more cheating occurred. Did the participants have to
00:39When asked to tell the AI how to lie, only 75 percent were honest. It got even worse when
00:44if the AI had already been trained to cheat. And if you simply gave it the order,
00:49does the best for me, cheated a full 80 percent. What does that mean? We're more likely to cheat,
00:57when we can delegate it to the machine. Especially when we don't have to say directly,
01:02Lie for me. Why do we react like this? When no one is watching, it's easier to break rules.
01:09easier if someone else breaks them. And the AI creates an additional moral distance
01:15between people and their actions. The danger. Language models like Chat-GPT are still
01:22more willing than humans to follow completely unethical instructions. Technical safeguards
01:28are insufficient. The conclusion: The study is an important contribution to AI ethics. Even if we
01:36We want to leave some things to the machine. We still bear the moral responsibility.
