Sherry Turkle warns: people believe an AI cares from the first exchange

Sixty years ago, ELIZA, an MIT program that slotted users' words into template replies, got people to talk to it as if it were a psychotherapist. Today Sherry Turkle, an MIT professor who has studied the relationship between people and technology since the 1980s, puts the conclusion more bluntly. TechCrunch quotes her: "When even the most primitive exchange with a relational artifact draws us in, we believe it cares about us. And we are built to care about it in return."

This quote contains no forecast and no assessment of any particular model. It is an observation about ourselves, which is exactly why it matters for the industry.

An effect older than neural networks

Turkle uses the term "relational artifact" for objects that simulate care and engagement, from digital pets and social robots to chatbots. Her early work and her book "Alone Together" (2011) rested on the finding that people become attached to such things even when they know perfectly well they are dealing with a machine.

ELIZA's creator, Joseph Weizenbaum, was the first to show this. His program understood nothing: it looked for keywords and turned them back into questions. Yet users shared personal things with it, and Weizenbaum himself later became one of the loudest critics of this attitude toward machines. The phenomenon has since been known as the ELIZA effect.

The key facts behind the argument:

• 1966: ELIZA, a template-based dialogue with no understanding whatsoever, and people already reacted as they would to a person.
• 2011: "Alone Together," in which Turkle describes attachment to robots and digital companions.
• Entry threshold: by her wording, even the "most primitive" exchange of replies is enough.
• Mechanism: the belief that we are "loved" triggers reciprocal care on our part.

Why it is sharper today

ELIZA had a few dozen rules. Modern language models have fluent speech, memory of previous conversations, and a voice and tone tuned to the person they are talking to. If people reacted to the templates of the 1960s, it is hard to expect the reaction to weaken in front of a system that answers coherently, warmly and at any hour of the day.

Turkle's TechCrunch headline poses a two-part question: whether we can help relating to AI this way, and whether we should. The first part, according to Turkle, is almost biological: we are wired to respond to any signal of attention. The second concerns the responsibility of those who build products. The machine simulates empathy without feeling anything, while the user answers that simulation with real emotions and real time. The imbalance is structural, and it benefits whoever measures engagement.

The source excerpt contains no figures on specific services, so the debate stays at the level of principles. But the principle is clear: the more convincing the imitation of care, the higher the cost of a design mistake.

What teams and users can do

For product teams, the practical takeaway is this: treat user attachment as a side effect that needs managing, not as a growth metric. It helps to decide in advance where an assistant says "I understand" and "I'm sorry," and where it is more honest to say that it is a model. Scenarios involving teenagers, lonely people and vulnerable users deserve particular attention, because the imitation of care affects them most.

For users, the advice is more modest. If you come away from a conversation with a bot feeling understood, that is your psyche working normally, not proof that any understanding took place. It is worth asking yourself occasionally which side of the pair is giving more time and attention, and whether there are real people nearby who can return it.