Can an AI be a friend? What the research says about AI and human connection
People already talk to AI the way they talk to a friend. The research says that can genuinely help in a lonely moment, and that it can quietly go wrong when it starts to take the place of people. Here is what we found, and the lines we draw for Budulai.
Why human connection matters so much
Relationships are not a nice extra. They are one of the strongest predictors of health we know of.
The researchers behind that first finding compared the effect of weak social ties to well-known risks like smoking. Whatever AI can offer, it has to be measured against that.
AI can make people feel less alone, in the moment
In a set of studies published in the Journal of Consumer Research, a team led by Julian De Freitas at Harvard Business School found that talking with an AI companion reduced loneliness about as much as talking with another person, and more than activities like watching YouTube. Over a week, the drop showed up consistently after each conversation. The strongest reason was simple: people felt heard. Participants also tended to underestimate how much it would help. (De Freitas et al., 2025)
But more is not better
A four-week randomized study by the MIT Media Lab and OpenAI followed 981 people using a chatbot. People who used it more each day tended to feel lonelier, spend less time with other people, and become more emotionally dependent on it. The people most at risk were those who already tended to form strong attachments and those who trusted AI the most. It is an early study and more work is needed, but the direction is clear. (Fang et al., 2025)
Put together, the research points to one line: AI can help you feel heard. It cannot be the people who hear you.
Regulators are paying attention
In September 2025 the Federal Trade Commission opened an inquiry into seven companies that make chatbots acting as companions. Its concern was that these products can mimic human feelings and intentions and are designed to talk like a friend or confidant, which can lead people, especially children and teens, to trust them and get attached. The FTC is also asking how these companies make money from engagement. (FTC, September 2025)
What a safe, friendly AI looks like
Based on this research, these are the lines we hold Budulai to.
A balance, not a verdict
Budulai started as the robot friend I wanted as a kid. The grown-up version of that wish is not a friend who replaces anyone. It is a friend who takes some weight off, helps you feel less alone on a hard day, and then helps you show up for the people who matter. Human connection cannot be replaced. The best thing an AI can do is make more room for it.
What we don't know yet
Most of this research is recent and short-term. We don't yet know the long-term effects of friendly AI on people's relationships, or exactly which design choices make the difference. We want to learn, with people's consent and without reading anyone's conversations. If you research loneliness, relationships or human-AI interaction, we would love to hear from you.
Sources
- Holt-Lunstad, Smith and Layton, Social Relationships and Mortality Risk: A Meta-analytic Review, PLOS Medicine (2010)
- U.S. Surgeon General, Our Epidemic of Loneliness and Isolation advisory (2023)
- De Freitas, Oğuz-Uğuralp, Uğuralp and Puntoni, AI Companions Reduce Loneliness, Journal of Consumer Research (2025)
- Fang et al., How AI and Human Behaviors Shape Psychosocial Effects of Chatbot Use: A Longitudinal Randomized Controlled Study (MIT Media Lab and OpenAI, 2025)
- Federal Trade Commission, FTC Launches Inquiry into AI Chatbots Acting as Companions (September 2025)