The chatbot's maker studied loneliness. Its heaviest users had more of it.

OpenAI and the MIT Media Lab released two studies of what ChatGPT does to the people who use it most. One analyzed nearly 40 million interactions automatically; the other was a four-week randomized trial of 1,000 people. The finding that will travel: higher daily use, across text and voice, was correlated with more loneliness, more emotional dependence, more problematic use and less socializing with other people. Voice mode looked better at first and lost its advantage at heavy use. The researchers used the word "correlated" and meant it. Hold that caveat and the finding at the same time; both are real.

What it does and doesn't show. Lonely people may use chatbots more, rather than chatbots making people lonely; a four-week trial cannot settle direction, and neither study has been through peer review that we can find. What the data does show is a pattern the longevity trap already knows well: a tool that feels like connection, consumed in place of it. The people most attached to the chatbot were the most likely to see it as a friend. Nothing about that is new. The substitute is.

Ben’s take

Give the company credit for publishing numbers that do not flatter its product; most would not. Then remember the incentive. Every consumer-software metric I have ever reviewed is a version of time spent, and a tool that people return to when they feel alone will score beautifully on that metric. The study is a warning label written by the manufacturer. Read it as one.

Loree’s take

I have patients who ask a chatbot about their symptoms at 1 a.m. instead of asking a person at 1 p.m. It answers, which a person might not. That is exactly why it is dangerous for the isolated: it removes the small friction that used to push someone toward a human being. In my office, loneliness shows up as inflammatory markers, poor sleep and a protocol that has replaced every relationship. The chatbot is the newest line in that protocol.

Why it matters to a recovering optimizer. The longevity trap has always run on substitution — a forum for a friend, a podcast for a dinner, a dashboard for a doctor. A conversational machine is the most convincing substitute yet, and it is also the most useful tool for research that has ever existed. Both are true. The question is not whether to use it. It is whether, at the end of the conversation, you call someone.