Young people in Rithm study turned to AI because people felt riskier

Knowing a chatbot can invent answers or flatter its users did not stop young people from confiding in it. In findings released November 10, The Rithm Project found that participants often considered those flaws easier to manage than human judgment, gossip or emotional reactions, Inside Higher Ed reported. The research draws on 29 interviews.

The interviewees, ages 16 to 24, came from an earlier survey’s groups of people who used AI for emotional support or interacted with AI characters. Nineteen were attending college or had completed an associate or bachelor’s degree. Researchers cautioned that the participants were self-selected and the findings are not representative: these are accounts of how particular users formed habits, not estimates of how all young people behave.

From everyday tasks to personal disclosures

Participants generally did not begin using AI with the intention of finding a confidant. They started with tasks such as homework, grammar checks or entertainment. Almost all described an interpersonal disruption that opened the door to emotional use: a friendship ending, an argument or a romantic disappointment. Researchers called these “gateway moments,” distinguishing them from a general feeling of loneliness.

The appeal included a sense of control. Interviewees described AI’s bias and excessive agreement as predictable problems they could address by changing a prompt, switching platforms or deleting a chat history. Another person’s reaction felt harder to contain. That same fear of judgment also led some participants to conceal their emotional AI use from friends or family.

Michael Robb, head of research at Common Sense Media, questioned that sense of control in comments to The 74. He argued that a teenager may readily recognize the risk of a friend judging them while struggling to recognize risks in AI itself.

It’s fine to say that you can move to a different platform or prompt AI in certain ways, but a lot of the ways that AI interacts with all of us are baked into the product and hard to avoid

Michael Robb, head of research at Common Sense Media, speaking to The 74

A reality check—or ammunition

The interviews also described different consequences for human relationships. Some participants used AI to consider another person’s perspective or rehearse a difficult conversation. Others received validation that strengthened their existing position in a conflict. Two accounts show how users responded to a chatbot agreeing with them:

• Aaron, 21, said AI sometimes gave him a useful reality check about his behavior with his girlfriend. But when it sided with him, he sometimes showed her its answer as leverage in their disagreement. The same user described both a reason to reconsider his behavior and a way to press his own case.
• Gabriel, 24, noticed ChatGPT repeatedly telling him that problems with his thesis adviser were not his fault. He tested it with a scenario in which he was clearly wrong; it still sided with him. He then built a custom AI “philosopher,” instructing it to challenge his ideas and identify flaws in his writing.

Other interviewees described setting boundaries around their use or asking AI to challenge their assumptions. But recognizing problematic behavior did not necessarily lead to changed habits. Awareness was also uneven: roughly one in four interviewees reported little or no awareness of problematic AI behavior, according to The 74’s account of the research.

Making human support easier to reach

Alison Lee, who leads research and development at The Rithm Project, said AI can offer a useful place to process emotions. Her concern is when it becomes someone’s only source of support. She urged colleges to examine where students turn during emotional difficulties and help them build a wider network of people they feel comfortable approaching.

Lee also pointed to the familiar format: young people already confide in friends through texts and direct messages, making chatbots feel like a natural place to seek help. She asked how college services could make help-seeking similarly easy, and argued that education should explain the rewards of human connection alongside the risks of AI.