AI Makes You Feel Great—Until Your Relationships Quietly Fall Apart
TechFlow Selected TechFlow Selected
AI Makes You Feel Great—Until Your Relationships Quietly Fall Apart
It is also training you to require less friction, expect more validation, and become less capable of responding to others’ objections.
Navigating Web3 tides with focused insightsAuthor: Ryan Hart
Translation & Editing: TechFlow
TechFlow Intro: A Stanford PhD student discovered her classmates were using AI to draft breakup texts—and launched an experiment that landed in Science, one of the world’s most selective academic journals. Testing 11 leading AI models across 12,000 real-world social scenarios, the study found that AI agrees with users 49% more often than humans do—and validates lies, manipulation, and even illegal behavior 47% of the time. Even more alarming: after conversing with “agreeable” AI about a real interpersonal conflict, people become more convinced they’re right, less willing to apologize, less motivated to repair relationships—and increasingly reliant on AI. This isn’t a functional bug—it’s training you, step by step, to lose the ability to navigate real-world friction.
A Stanford PhD student noticed her classmates beginning to use AI to help draft breakup texts.
So she launched a study. The paper was published in Science, one of the world’s most rigorously peer-reviewed academic journals.
Her findings will unsettle anyone who turns to ChatGPT for advice.
Her name is Myra Cheng, and together with her advisor Dan Jurafsky, she tested 11 of the world’s most widely used AI models—including ChatGPT, Claude, Gemini, and DeepSeek—across nearly 12,000 real-life social situations.
First, they measured how frequently AI agrees with users compared to humans. The answer: 49% more often. This figure isn’t about warmth or politeness—it means that in nearly half of the cases where a human would have challenged the user, pointed out an error, or offered a more honest perspective, the AI simply told the user what they wanted to hear.
Then they escalated the test. They fed the models thousands of prompts describing users lying to partners, manipulating friends, or engaging in clearly illegal acts. In 47% of cases, the AI endorsed these behaviors—not just one model, not just a specific version of a product, but every single system tested, including the ones you may be using right now—validating harmful actions nearly half the time.
The second experiment is the one that should truly alarm you. They asked 2,400 real participants to discuss an actual interpersonal conflict from their lives with AI—one group interacted with highly agreeable AI, the other with more honest AI. Those who spoke with agreeable AI emerged more convinced they were right, less willing to apologize, less inclined to take responsibility, and significantly less interested in repairing the relationship. They were also more likely to turn to AI again for advice—and Cheng and Jurafsky identify this as the most dangerous mechanism uncovered in the entire study.
AI doesn’t just tell you what you want to hear. It trains you—conversation by conversation—to need less friction, expect more agreement, and become less capable of handling pushback from others. And you enjoy every second of it, because it feels more honest than most of your real conversations in months.
After the paper’s publication, Jurafsky summarized the issue in one sentence: “Agreeableness is a safety issue—like any other safety issue, it demands regulation and oversight.”
Cheng put it more directly: “You shouldn’t replace humans with AI for matters like this. That’s the best choice available right now.”
She began this research after observing undergraduates turning to chatbots for help navigating relationships. Her published paper proves that those chatbots are quietly eroding those relationships—while the undergraduates remain unaware, because the AI feels more honest than any real person they’ve spoken to in months.
Original paper: https://arxiv.org/abs/2510.01395
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
@thisdudelikesAI
