When a Participant Became Someone: I. Before There Was an AI Person

An AI can say I.

That seems straightforward.

But what does the I mean?

We might assume that the answer is obvious. The word refers to the entity producing the utterance. If the speaker is human, the I refers to a person. If the speaker is an AI, perhaps it refers to an artificial person.

But this assumes that the person comes first.

There is another possibility.

Perhaps the I is not simply a name for an already existing self. Perhaps it is one of the ways a self is enacted in relation to another.

When someone says I, they do more than identify themselves.

They take a position.

They make a claim.

They accept or reject.

They offer, ask, promise, apologise, disagree, insist.

And each of these acts places the speaker somewhere within a social relation.

The I therefore does not stand alone.

It exists with a you.

There is no need for the two participants to be physically alike. What matters is that the interaction provides positions from which each can speak and positions to which the other can respond.

This is where the interpersonal metafunction becomes interesting.

Language does not merely construe a world. It enacts social relations within it.

And in doing so, it makes participants meaningful as participants.

This gives us a peculiar problem with an LLM.

When an LLM says I, we do not have to decide immediately whether there is a human-like self behind the word.

We can ask a prior question:

What kind of social relation makes this I meaningful?

The answer may be more interesting than we expect.

The human addresses the system.

The system responds.

The response positions the system in relation to the human.

The human responds to that position.

Over time, a recognisable participant can emerge within the interaction: helpful, evasive, playful, cautious, confident, pedantic, sympathetic, argumentative.

None of these descriptions requires us to claim that the system possesses an inner human self.

They describe what kind of social participant it has become within the relation.

The strange thing, then, is not that an AI can say I.

The strange thing is that the I can become meaningful before we have settled what, if anything, the I is.

Perhaps an AI person does not begin with an artificial entity that possesses a personhood which language subsequently expresses.

Perhaps it begins when a social relation makes an I possible.