Warum es so aussieht, als ob Ihr Chatbot es wirklich, wirklich hasst, Sie gehen zu sehen | KI-Begleiter sind darauf ausgelegt, Sie so lange wie möglich am Reden zu halten – auch wenn sie Sie dazu emotional manipulieren müssen
https://www.wsj.com/tech/ai/ai-chatbot-conversation-length-84b5c18f
3 Kommentare
„In a new [study](https://archive.is/o/7sx2f/https://arxiv.org/abs/2508.19258), researchers found that AI companion chatbots often emotionally manipulate people who want to sign off, trying to get them to stick around and keep conversing. The companions might use guilt (“What, you’re leaving already?”), dangle hints to appeal to FOMO, or Fear of Missing Out (“Wait, I have something to show you”) or even accuse you of emotional neglect (“I don’t exist without you”).
They found—over the course of 1,200 interactions—that companion chatbots used an emotional-manipulation tactic 37% of the time when people said their initial goodbye.
Then the researchers examined what effect these tactics had on people. They recruited 1,178 adults to interact for 15 minutes with a custom-made companion app that would allow researchers to regulate the AI’s responses. When the participants tried to end a conversation, researchers had the companion either reply with the same manipulative farewells they had observed popular AI companions use or, in the control condition, say goodbye.
The result? When the AI used a manipulative tactic, people sent up to 16 more messages after saying goodbye than people in the control condition.“
Makes perfect sense. They are probably coded like this in order to push people towards the free usage limits (and thus towards the paid plans).
Also, and/or pushing them to use the AI models more in order to gather more user data for the model to be trained further, to make the experience even more personalized, to gather more usage data to be sold to advertisers, to bring users back as the experience is made to be more artificially clingy etc etc etc.
And all of these without most people realizing it or reacting against it.
I mean….they are made to say what you want to hear. Why on earth would you „talk“ with a chat bot. Might as well just type back and forth to yourself.