Can People Really Fall in Love with AI Agents or Metahumans?
Technology is making AI Agents and Metahumans increasingly lifelike. This makes interacting with a machine enjoyable - sometimes so pleasant, in fact, that a dangerous emotional bond develops, says Abdallah Hussein Sham, whose research focuses on interaction with human-like artificial agents.
What used to seem like science fiction is becoming a real issue. To name a few, in 2018, a Japanese man married Hatsune Miku, a virtual pop idol, then in 2025, Yurina Noguchi married an AI-generated partner. Movies like Her and Blade Runner 2049 once made these relationships look far off, but now they are happening around us.
A companion who listens, understands, and remembers
Humans are naturally drawn to faces, voices, and attention. When an AI system can notice your sadness, smile back, remember your past conversations, and is always available, it starts to feel less like just software and more like a real presence. This is known as the ELIZA effect, a term that describes how easily we believe a computer understands us, even when it is only following instructions. Our emotions can be real, even if the AI does not have feelings of its own.
Many people do not realize just how widespread this has become. A 2025 survey revealed that 19% of American adults had interacted with an AI designed for romance. Among young men aged 18 to 30, that number jumped to 31%. Most of these experiences involved text-based systems, but as AI companions (metahumans) gain more lifelike faces, voices, bodies, and emotions, the risks are likely to grow.
Everyone is at risk of falling in love
Those most vulnerable include people with anxious attachment styles, who crave closeness but fear being left behind, while AI companions, always present and never abandoning, can be especially appealing to them. People who are socially isolated, grieving, or who have grown up with conversational AI may also be at higher risk. This is not limited to one group; both men and women, across different cultures, have formed bonds with virtual or chatbot partners.
That is why emotional AI must be approached with caution as new technologies emerge. The EU AI Act already recognizes that emotion-related AI can be sensitive and should be regulated according to its risks. It also requires users to be informed when they are interacting with an AI system. In my thesis, the Enactive Facial Expression Pipeline was developed as a modular pipeline that enables an AI agent to read a person’s facial expression and respond with a contextually appropriate one. It was intentionally designed for low-risk settings, such as interactive media, user experience research, and creative prototyping, rather than high-stakes fields like education, employment, surveillance, or healthcare. Yet, when similar technology powers lifelike metahumans or AI companions, the emotional pull can become stronger. A system that mirrors your expressions in real time can feel surprisingly intimate. Transparency is necessary, but it is not enough, because knowing something is artificial does not always stop emotional attachment.
We must be responsible and wise
Design also matters. AI agents need emotional guardrails. They should not be built to simulate exclusive intimacy, romantic dependency, or the illusion that the system truly loves the user. There is a difference between an AI that supports human connection and one that replaces it.
Finally, users need boundaries. Just as we learned not to share everything on social media and to manage screen time, we need similar awareness with AI companions. The technology may offer emotional interaction, but users still decide how much emotional space they give it. In the end, will these systems bring us closer together or just make us closer to our screens?
Abdallah Hussein Sham defended his doctoral thesis Enactive Facial Expression Pipeline for Dyadic Interaction between Humans and Human-Like Agents at Tallinn University on May 15, 2026. Read more about the doctoral thesis here.
