A particular kind of concern comes up regularly in conversations about AI companions: if you are spending time with an AI instead of reaching out to people, you are withdrawing from human connection. The AI is training you to settle for something lesser. You are giving up on people. This concern sounds protective of human relationships, and it has some intuitive force. The evidence does not support it as a general claim.

What the Concern Gets Right

There is a version of AI companion use that would be worrying by this logic. If someone with a rich social network began actively canceling plans with friends to talk to an AI instead, citing the AI as a more comfortable alternative, and over time found their human relationships deteriorating, that would be a pattern worth taking seriously. There is also legitimate research to consider. The MIT study of over fourteen thousand AI companion users found a curvilinear relationship between usage and wellbeing, with moderate users reporting benefits but the heaviest users showing some signs of social substitution effects. The concern is not invented. But the myth converts this conditional concern into a general condemnation, and that move is not supported by the evidence. Most users are not using AI companions at the high-use end of that spectrum. And for many users, particularly those with limited access to human support networks, the AI is not substituting for something that was available. It is filling a gap.

The Substitution Assumption Is Empirically Weak

Stanford research on digital social engagement has found that the relationship between AI companion use and human social engagement is, for most users, not substitutive but complementary. People who reported using AI companions for emotional processing also reported higher rates of seeking human support when it was available. The AI was functioning as a kind of rehearsal space, a place to articulate things before bringing them to human relationships. This is a known pattern in therapeutic contexts. Clients who work with a therapist often find that the reflective work they do there makes their human relationships more functional, not less. The processing happens in one context and the benefits generalize. AI companions appear to operate similarly for many users. A Harvard study on social disclosure found that users who processed difficult emotions with an AI were subsequently more likely to disclose those emotions to a close human contact within the following week. The AI was not a terminus. It was a starting point.

Who Is Actually Being Harmed by This Myth

The telling question is: who does the giving-up-on-people narrative actually protect? If someone has robust human relationships and is worried about a friend who seems to be over-relying on an AI, the concern is understandable even if overstated. But the narrative also reaches people who do not have those relationships, people who are isolated, who live alone, who have anxiety around social interaction, who are in circumstances that make human connection difficult. For those people, the message that using AI means giving up on people carries an implicit accusation: you should be working harder at human relationships instead. This ignores the reality that human relationships require mutual availability, social skills that not everyone has in equal measure, proximity, and trust that takes time to build. AI companions can help develop all of these without the high stakes of a relationship that might go wrong. A tangent worth noting: the same concern was raised about online communities in the early internet era. People were warned that chatting with strangers online would make them less capable of real-world social connection. The opposite turned out to be true for most users. Online communities became on-ramps to in-person ones. New social technologies are often accused of displacement when they are actually supplementing.

The Framing That Fits the Evidence

The evidence supports a more nuanced framing: AI companions, used at moderate levels, tend to complement rather than replace human connection for most users. Heavy use may carry different risks. For people with limited access to human support, the calculus is different than for people with strong existing networks. Treating all AI companion use as giving up on people erases these distinctions and leaves people without a tool that genuinely helps them.