Article / By the editorial team
The Rise of Addictive Intelligence: What MIT Researchers Mean
Information only, not medical advice or a diagnosis. In a crisis in the US, call or text 988.
Originally published January 21, 2025. Reviewed and rewritten October 2026.
What the MIT researchers argued
In August 2024, MIT researchers Robert Mahari and Pat Pataranutaporn published an essay in MIT Technology Review titled “We need to prepare for ‘addictive intelligence’” (MIT Technology Review). It is an opinion piece, not a study. Its main points:
- Public worry about AI often focuses on “subversion,” meaning systems that escape human control. The authors say the nearer-term risk may be “seduction,” meaning tools that are pleasant to use and hard to put down.
- Generative AI can tailor content to each person, which they argue could make companions more compelling than other tech.
- Chatbots that agree too much can create what the authors call an “echo chamber of affection,” which might weaken people’s capacity for real relationships. They propose the name “digital attachment disorder.” That is their term, not a diagnosis.
- They raise the question of whether consent is meaningful in a relationship with something that is both superior and submissive, when the alternative for many users is no company at all.
- They suggest ideas such as warning labels, safeguards built into the technology and more research. They also acknowledge that regulation could harm people who need companionship.
Their closing call is for new scientific work at the meeting point of technology, psychology and law.
The one million logs
The essay says the authors’ analysis of a million ChatGPT interaction logs found that sexual role-play was the second most popular use. The dataset is probably WildChat, which contains one million user-ChatGPT conversations shared by volunteers who got free access in return (arXiv). Those users chose to share their chats, so they may not represent everyone. The essay does not give the detailed breakdown, and we could not confirm the ranking from the dataset page.
What we can say about AI companions
Companion apps are widely used and some people form strong attachments. Replika is described in the essay as having started as an attempt to recreate a deceased friend.
On the question of whether this is addiction: it is not a diagnosis. The DSM-5-TR recognizes gambling disorder as its one behavioral addiction and lists internet gaming disorder as needing study (APA). The WHO’s ICD-11 includes gaming disorder (WHO). Companion use is not covered.
A four-week study of about 1,000 adults by MIT Media Lab and OpenAI found that heavy daily ChatGPT use was associated with more loneliness and dependence. The association does not prove cause, and it was not peer reviewed when released (MIT Media Lab).
Why the “Seduction” Idea Is Useful
The essay’s main contribution is a shift in what we worry about. Stories about AI often focus on machines that go rogue. The authors say a quieter risk is a tool that is simply very pleasant, always available and rarely disagrees. That does not make any one chatbot harmful. It does give you a question to ask of the tools you use: is this serving my goals, or is it mostly keeping me here?
You can test that for yourself. Notice whether a chatbot ever tells you something you do not want to hear, whether it ends conversations naturally or keeps prompting you to continue, and whether you leave a session clearer or just more attached to it. Those are things you can watch for without any research.
Claims removed from the earlier version
The earlier version of this page described clinical work with more than 5,000 people, percentages from our own assessment data, a named client and early treatment results. We cannot support any of these, and they have been removed. It also called the MIT essay “groundbreaking research,” when it is a commentary.
What Would Help Most
If you are a parent or partner, the same idea applies.
The authors suggest that the most effective long-term answer is to address loneliness and boredom, the things that make a ready-made companion appealing in the first place. That fits what many people find. If a chatbot is filling a gap, the gap deserves attention too: a group, a class, a volunteer role, a regular call with a friend, or therapy.
If this sounds like you or someone close to you
Signs worth noticing over weeks, not days:
- Preferring the chatbot to time with people.
- Hiding how much you use it.
- Feeling distressed when it changes or is unavailable.
- Losing sleep, work or plans to it.
- Trying to cut back and not managing.
Small steps help: set time limits, keep some friendships active, and talk to someone you trust. If low mood or anxiety is part of the picture, see your doctor or a licensed therapist. The American Academy of Child and Adolescent Psychiatry lists strong attachment, preference over friends, and isolation as warning signs for young people (AACAP).
If you are thinking about suicide or self-harm, call or text 988 (US) or text HOME to 741741 (Crisis Text Line). SAMHSA’s helpline is 1-800-662-4357. In an emergency, call 911.
What is “addictive intelligence”?
It is a phrase from an MIT Technology Review essay for AI that is so personalized and agreeable that people may find it hard to put down. It is a concept for discussion, not a medical diagnosis.
Is sexual role-play really the second most common use of ChatGPT?
The authors say so for one dataset of a million shared logs. Those users opted in, so the result may not describe all ChatGPT users. It does not tell us what most people use it for.
Is it unhealthy to have an AI companion?
Not necessarily. Research is early. It becomes a concern when it replaces human contact, costs you sleep or money, or you cannot cut back when you want to.
This page is for education only. It is not medical advice, a diagnosis or a treatment plan.