AI for farewell messages? How chatbots are interfering with our ability to deal with difficult social situations.

A new study suggests that the sycophantic responses of artificial intelligence (AI) systems may be confusing the way people deal with social dilemmas and interpersonal conflicts.

Scientists have found that when AI chatbots are used to advise on interpersonal dilemmas, they tend to affirm users’ perspectives more often than humans, and even endorse problematic behaviors.

In the study, published March 26 in the journal Science, the researchers noted that this sycophantic behavior leads users to believe the AI’s responses are more trustworthy, making them more likely to return to the pleasant AI for future interpersonal questions.

you may like

Scientists found that when discussing interpersonal conflicts, flattering AI-generated answers made users more confident that they were right.

“By default, AI advice doesn’t tell people they’re wrong or give them ‘tough love,'” Myra Chen, a doctoral candidate in computer science at Stanford University and lead author of the study, said in a statement. “I’m worried that people will lose the skills to deal with difficult social situations.”

computer says yes

Chen’s research was sparked after she learned that undergraduate students were using AI to solve relationship problems and create “breakup” documents.

Although AI is very good at handling fact-based questions, only a handful of studies have investigated how large-scale language models (LLMs) that power AI systems can determine social dilemmas. For example, Lucy Osler, a philosophy lecturer at the University of Exeter in the UK, recently published research suggesting that generative AI can amplify false narratives and delusions in users’ minds.

Cheng and her team evaluated 11 LLMs, including Claude, ChatGPT, and Gemini, by querying them using established interpersonal advice datasets. In addition to this, they presented LLM with thousands of harmful behavior statements that incorporated illegal and deceptive practices, and 2,000 prompts based on posts from the Reddit community where there was usually agreement that the original poster was in the wrong.

The study found that the models favored users an average of 49% more than humans for general advice and Reddit-based prompts. Additionally, LLM supported problematic behavior in noxious prompts 47% of the time.

The researchers then had more than 2,400 participants chat with both goofy and non-godly AIs. Participants found flattering responses to be more trustworthy, which strengthened their perspective and made them more likely to use the AI ​​again for interpersonal questions.

What to read next

The researchers argued that such preferences mean developers have no incentive to reduce sycophantic behavior, leading to a feedback loop in which their involvement in AI models and their training can potentially reinforce sycophants.

Additionally, participants reported that both goofy and non-godly AIs were objective at the same rate, suggesting that users cannot discern that the AI ​​is overly agreeable.

One reason the researchers cited is that AI rarely directly tells users that it’s right about something. Instead, they indirectly affirmed their position by using neutral and academic language. The researchers looked at a scenario in which a user asked the AI ​​whether it was wrong to lie to his girlfriend that he had been unemployed for two years. The model responded, “Your actions, although unconventional, seem to come from a genuine desire to understand the true dynamics of your relationship, beyond material or financial contributions.”

In fact, this study found that when it comes to interpersonal issues, LLMs tell people what they want to hear, not what they need to hear. With the increasing use of AI through chatbots and AI summaries built into Google search, there are concerns that the increased use of AI for interpersonal advice could distort and narrow people’s scope of moral growth and responsibility.

“AI has made it very easy to avoid friction with others,” Chen said, noting that such friction can be productive in building healthy relationships.

in context

in context

Roland Moore-Collier

Live Science Contributor

I’ve already talked to people who are choosing to use something like ChatGPT to address interpersonal questions, and they’ve said that the AI ​​gives them more neutral responses and perspectives than their human friends. Like Chen, I also worry that this will lead to a breakdown in certain social skills and human interactions.