Compare two answers. One says, 'Incorrect input.' The other says, 'It looks like the date format may be causing the problem. Try entering it as DD/MM/YYYY.' The second system often feels not only friendlier but smarter.

Humans judge competence partly through communication. Clear explanation, responsiveness and appropriate tone are signals we use with other people. When AI displays these cues, some of the same social machinery activates.

This is anthropomorphism: attributing human-like qualities to non-human things. We name cars, swear at printers and thank voice assistants. Conversational AI gives this tendency far more material to work with.

Warmth can increase comfort and engagement, but it can also inflate trust. A confident, empathetic answer may feel more reliable even when its factual quality is unchanged.

Designers therefore face a trade-off. A cold system is unpleasant and hard to use. An excessively human system can blur boundaries, especially in emotional or advisory contexts.

The right tone probably depends on task. A customer-service bot can be warm. A medical support tool should be compassionate but unusually clear about uncertainty and limits. A financial system should not use friendliness to make risky suggestions feel safer.

Users can protect themselves by separating two questions: 'Did this answer feel good?' and 'Was this answer well supported?' Human beings often combine them without noticing.

Friendly AI feels intelligent because social fluency is one of the ways we have always recognised intelligence in people. The machine benefits from that old shortcut.