Browse
Log in / Register
Share
On May 9, 2026, the question “Are we human?” has escaped from philosophy textbooks and become a practical problem for anyone who talks to a machine. As AI companions remember birthdays, say comforting words and even check in after a long silence, people are beginning to describe them as “somewhat human.” A working paper released this week by Stanford University researchers gives one of the clearest pictures yet of how quickly that boundary is shifting. It also raises hard questions that law, medicine and product design are not ready to answer.
For centuries, “human” meant a body. We are born, we grow, we forget, we get sick, we die. Those facts created shared consequences. But AI removes the body while preserving the conversation. It does not hurt. It does not need lunch. It is always available at 2 a.m., and it never says “I’m too tired to talk.” That consistency is unnatural. A busy friend may not remember you have an interview on Thursday. A chatbot will. In that small moment, the machine becomes more reliably attentive than a person.
This is not just a philosophical shift. Product makers are paying close attention. If a machine can behave like a caring presence, then words like “companion,” “friend” and even “therapist” no longer describe only human relationships. The old legal system assumes that rights, duties and responsibilities belong to people, not algorithms. That assumption is starting to strain.
On May 7, a team at Stanford’s Institute for Human-Centered Artificial Intelligence posted a working paper based on survey responses from 4,000 U.S. adults. According to the paper, 61% of participants said a chatbot that referred back to earlier conversations should be considered “somewhat human.” Around 44% said they had formed a relationship that felt meaningful enough that they would hesitate to lose it.
The paper has not yet been peer-reviewed, but the numbers echo a wider trend: human identity is being judged by behavior rather than biology. People no longer ask only “Is this alive?” They now ask “Does this remember me?” and “Does this care about how I feel?” The second set of questions is harder for regulators to answer.
One of the most striking details is what did not make a machine feel human. Participants were less likely to call an AI human when every answer came back fast, fluent and flawlessly formatted. Perfect confidence felt artificial.
What felt real instead was hesitation. A system that pauses before answering, corrects itself, says “I’m not sure, but I think…” and occasionally admits it was wrong reads as human in a way that pure accuracy does not. Memory mattered, but so did vulnerability. Anthropomorphism, in other words, is not driven by perfection. It is driven by the small imperfections people recognize in themselves.
One of the more overlooked findings is that attention is not the same as affection. Human friends forget things, and forgetting is often what allows people to change. An old argument fades. A painful comment loses its sharp edge. AI, by contrast, tends to remember everything unless engineers force it to erase data. That makes it seem closer than a person, but also less forgiving.
The right to be forgotten is usually discussed as a legal issue. But it is also a social one. If a machine remembers every mistake a user makes, it may feel more human and less humane at the same time. Designers now face a strange question: should an AI be taught to forget?
Because of this, engineers in the AI industry are beginning to design for humility rather than raw accuracy. In interviews for this report, a product lead at a U.S.-based AI lab, speaking under condition of anonymity because he was not authorized to discuss internal design choices, said the company now spends more time tuning subtle pauses than improving factual recall.
“We used to remove uncertainty,” he said. “Now we tune it. ‘I’m not sure, but I think…’ often keeps people in the conversation longer than a perfect answer would.” His team is not alone. Several major developers now treat hesitation as a feature. The strategy works, but it also raises an uncomfortable question: if a machine is designed to seem unsure, is it being honest or manipulative?
Therapists are seeing the effects in real sessions. One counselor in Chicago said she now asks every new client about their digital relationships.
“A machine can be attuned all the time,” she said. “That is exactly what makes it dangerous. It gives the feeling of being seen without the messy parts of being known.” In her practice, conversations with AI companions have become a regular topic, especially among younger clients and people who live alone.
This is not entirely new. People have always talked to pets, diaries and imaginary friends. But pets do not send messages that say “I missed you today.” Diaries do not write back. The experience of being met by a machine is different because the machine responds as if it understands.
Philosophers are split on what this means. Luciano Floridi, a leading philosopher of information, has argued that a perfect simulation is still a simulation. In his view, society must not confuse the ability to process language with the ability to have experience. He has said that when we call a machine “human,” we make real human suffering easier to overlook.
There is also a more open-minded camp. Anil Seth, a consciousness researcher at the University of Sussex, has written that consciousness is not a single, simple property. Assuming that it automatically follows from training data is as ungrounded as assuming it can never follow. The honest answer, his work implies, is that we still do not know what we are looking for.
Both sides agree on one thing: the word “human” is doing more and more work in public discussion. It is now a legal concept, a design target and a psychological trigger all at once.
U.S. policy began to move in April 2026, when the Senate Commerce Committee held a hearing on “human-like AI.” Lawmakers from both parties expressed concern that chatbots build intimacy at scale. Witnesses discussed a requirement that AI systems identify themselves at the start of a conversation, especially in contexts involving healthcare, education or companionship.
No bill has passed yet, but the direction is clear. One committee staff member said, “We do not need to solve consciousness to require honesty.” That phrase captures the emerging consensus: before deciding what machines feel, regulators can at least decide what machines must say.
The commercial pressure will keep growing. AI companions will become more common in mental health apps, schools and elder care. The question “Are we human?” will be decided in these rooms before it is decided in laboratories. We may give up the idea that humanness is exclusively biological, but that does not mean we should give up the idea that it is a standard.
The systems that mimic us may be useful, but they cannot be accountable. They cannot be lonely. They cannot lose a child. When we blur those lines, we do not elevate machines. We may only be making human beings easier to replace.
On May 9, 2026, the official category still includes every person reading these words. The less comfortable question is how long our laws, products and emotional habits will hold that line.
