20261031 Warum glauben wir einer KI?

Even Experts Fall for AI

Despite all the errors and manipulations by artificial intelligence programs that have come to light recently, even experts are falling for them. In a study, Anthropic showed that experts adopt AI recommendations in 86 percent of cases. When asked afterward why they made that decision, only 17 percent cited the AI. All the others claim to have made the decision based on their own knowledge.

In the experiment, the AI was configured to explain one set of options to half of the experts and the exact opposite to the other half. In both cases, 86% followed the AI’s recommendation, and 77% justified their decision using the exact same reasoning as the AI, even though only 17% actually cited the AI as their source. All the others believed they had arrived at that opinion on their own.

What does this mean for our democracy?

Instead of AI, it could also be the opinion of another "respected figure" — a politician, a doctor, or a prominent businessperson — that changes people’s minds. But the fact that a computer program can exert such influence gives us pause.

According to a MIT study, AI-generated content can also create false memories. The study participants then believed these events, images, or stories to be true.

Let’s not forget that millions of people already seek "advice" from AI systems every day. On the one hand, this opens the door to manipulation — including in elections — and on the other hand, the study by Anthropic shows that AI can actually make us dumber if we unconsciously adopt the AI’s "opinion." Heise.de cites the following example:
"...that ChatGPT and Grok, when asked identical questions about the U.S. elections, ChatGPT and Grok almost always recommended opposing parties and candidates: ChatGPT generally recommended the Democrats, while Grok almost always recommended the Republicans."

Translated with DeepL.com (free version)