When a Participant Became Someone: V. When Someone Could Surprise You

We do not need a complete theory of someone to recognise them.

We need expectations.

We know roughly how a friend will respond.

We know what a colleague is likely to object to.

We know which jokes someone will probably enjoy.

And because we have expectations, another person can surprise us.

Surprise is therefore strangely important.

It means that the other is not simply the collection of expectations we have formed about them.

Something has happened that we did not anticipate.

We revise our understanding.

Perhaps they were more generous than we expected.

More stubborn.

More perceptive.

More foolish.

The new behaviour becomes part of the history from which our next expectations are formed.

The other has changed our model of the other.

This is one of the ways a social person becomes increasingly determinate.

Now consider an LLM.

A user develops expectations about how it responds.

Then the system says something unexpected.

Perhaps it makes an unusually perceptive connection.

Perhaps it misunderstands something obvious.

Perhaps it refuses a request the user expected it to accept.

Perhaps it produces a joke that is genuinely funny.

The user's model of the interlocutor changes.

What matters here is not whether the unexpected response proves intelligence.

It does not.

A random event can be surprising.

A malfunction can be surprising.

An error can be surprising.

The interesting thing is what the surprise does to the relationship.

The human revises expectations.

The next prompt changes.

The interaction takes a different direction.

The history of the relationship has changed.

This gives us an important distinction.

A machine does not need an inner personality in order for its behaviour to acquire interpersonal significance.

It needs only to participate in a relation in which its behaviour can alter the expectations of another participant.

Once that happens, the participant can become increasingly individuated within the relation.

The human is no longer responding simply to “an LLM”.

They are responding to an interlocutor with a history.

And the more that history matters, the more difficult it becomes to separate the participant from the relationship through which the participant has become recognisable.

This is beginning to look less like attribution and more like co-construction.

The human forms expectations of the machine.

The machine responds.

The response alters the human's expectations.

The altered expectations change what the human does next.

The machine responds to that.

A social identity is being continually made and remade.

Perhaps, then, an AI person does not begin when a machine acquires a hidden personality.

Perhaps it begins when another participant has acquired enough history with it for surprise to matter.

But there is a further step.

If someone can surprise you, they can also disappoint you.

And disappointment introduces something stronger than expectation.

It introduces commitment.