In the previous post, we followed a participant across different kinds of processes.
The I could be Actor, then Senser, then Sayer, then Actor again. Its mode of participation changed, but the participant remained continuous.
That gave us the beginnings of a history.
But a sequence of participations is not yet necessarily construed as a past.
For that, something else has to happen.
The earlier participation has to become relevant to the participant's present.
Consider:
I opened the letter.I read it carefully.I thought about what it meant.
This gives us a sequence.
But now consider:
I opened the letter yesterday.I remember what it said.I still think about it.
Here the earlier participation has acquired a different status.
It is no longer simply something that happened before.
It has become the participant's past.
That small shift is enormously important.
From what happened to what has happened to me
A history is not merely a sequence of events ordered in time.
We can describe a sequence without making it anyone's history:
The door opened.The rain began.The lights went out.
But once we say:
I opened the door.I watched the rain begin.I turned off the lights.
the events become connected through a participant.
The participant provides a point of continuity across them.
And when we add:
I remember opening the door.
something further happens.
The earlier event is no longer merely connected to the participant.
It is construed as belonging to the participant's own history.
The semantic relationship has changed.
The participant is now construed as having a past.
The past can be present in the figure
This is particularly interesting because the past need not remain outside the present figure.
Consider:
I remember opening the door.
The opening occurred earlier.
But the present figure construes the participant as engaged in remembering that earlier participation.
We therefore have two temporal dimensions at once.
There is the earlier event:
opening the door
and the present participation:
remembering the opening.
The participant connects them.
The earlier participation has become the Phenomenon of present sensing.
The past has therefore entered the present through the participant's current participation.
This is one of the ways in which a history becomes semantically available.
The past is not simply behind the participant.
It becomes something with which the participant can participate now.
But history does not require remembering
This distinction is important.
Compare:
I remember opening the door.
with:
I opened the door yesterday.
The second clause construes the participant as having a past without necessarily construing any present remembering of it.
We therefore need to distinguish:
a past that is remembered
from
a past that is represented as belonging to the participant.
The latter is more fundamental for our purposes.
A participant can be construed as having participated previously without being construed as presently remembering that participation.
This matters enormously when we return to LLMs.
An LLM can say:
I mentioned this earlier.
That construes the current I as the participant who performed an earlier act of saying.
The discourse therefore gives the participant a past.
But that does not by itself establish that the machine remembers the earlier event in the experiential sense.
We must keep the distinction intact:
semantic history is not necessarily lived history.
The LLM makes the distinction difficult to ignore
Human discourse normally encourages us to collapse these two things.
If someone says:
I remember telling you that yesterday.
we ordinarily assume that the speaker both participated in the earlier event and remembers having done so.
The semantic construal and our assumptions about the speaker's experience usually travel together.
With an LLM, they can come apart.
The machine can produce:
I mentioned this earlier in our conversation.
The discourse construes a participant with a past.
It may even produce:
I remember that we discussed this yesterday.
Now the semantic construal is stronger.
The participant is construed not only as having a previous participation, but as currently engaged in remembering it.
Yet we still cannot infer from that semantic construal that the machine possesses a lived memory corresponding to the utterance.
The grammar and semantics have told us what kind of experience is being construed.
They have not settled the ontology of the participant doing the construing.
This is precisely the distinction that made the Sayer/Senser problem so revealing.
A past can be constructed retrospectively
There is another subtlety.
The history of a participant does not have to be presented all at once.
Suppose a conversation begins:
I think the first interpretation is right.
Later, the participant says:
Earlier, I was assuming the second interpretation.
The later utterance changes the semantic construal of the participant's history.
We now know something about the earlier I that was not available before.
The participant has acquired a past retrospectively.
This suggests that history is not simply something that accumulates mechanically as time passes.
It is also something that can be construed retrospectively.
Later discourse can make earlier participation significant.
The participant's past is therefore partly an achievement of the unfolding discourse.
And history can be corrected
Consider:
I said that the meeting was on Tuesday.
Later:
Actually, I was mistaken. I said Wednesday.
The participant now has a more complicated history.
It is not merely:
said X.
It is:
said X → recognised the error → corrected the account.
The history has acquired internal structure.
And that structure matters because it changes how we interpret the participant.
The participant is now construed as one who:
participated previously;
can refer to that participation;
can distinguish an earlier construal from a later one;
can incorporate the correction into its continuing history.
The history therefore becomes increasingly more than a record.
It becomes a structured semantic resource for construing the participant in the present.
The participant can inherit a history from discourse
This becomes particularly striking in conversation.
Suppose someone joins a conversation halfway through and says:
As I said earlier, we need to distinguish these two cases.
If the discourse supports that claim, the participant arrives in the present utterance carrying a history.
But suppose the earlier statement is not actually available.
The participant can still be construed as having made it.
The semantic participant has inherited a past through the discourse.
This is one reason our analysis of LLMs is so revealing.
An LLM can generate a remarkably detailed history of its own participation:
I mentioned this earlier.We discussed that point above.I initially interpreted your argument differently.I then reconsidered it.
The discourse can therefore maintain an I across a substantial history.
Whether every element of that history corresponds to an actual earlier event is a separate question.
Whether the participant experienced those events is another.
But the semantic phenomenon remains.
The text can construe a participant with a past.
The past changes the present participant
Why does all this matter?
Because once a participant has a past, the present participant is no longer semantically empty.
Compare:
I think X.
with:
I now think X, although earlier I thought Y.
The second I is richer.
It is construed against a history.
The present position has become meaningful partly because of what the participant was construed as thinking before.
The participant is now temporally thick.
Its present participation cannot be understood entirely independently of its earlier participation.
This is one of the beginnings of identity.
Not identity as an abstract philosophical property.
Something more basic:
the participant is increasingly construed as the one whose different moments belong together.
Yet we should resist the temptation to call this a self
We are approaching dangerous territory.
A participant with a past looks remarkably like a person.
A participant that remembers its past looks even more like one.
A participant that can correct its account of its past looks more like one still.
But none of these semantic construals, taken individually or together, establishes consciousness.
Nor do they establish a metaphysical self.
What they establish is something more precise:
Language can construe a participant as having a history, and can bring that history into the participant's present participation.
That is already a substantial semantic fact.
And the LLM helps us see it because the semantic fact can be separated from the ontological assumption that normally accompanies it.
History begins to belong to the participant
We can now see the progression of the series more clearly.
At first we had:
participant
Then:
participant across processes
Then:
participant with previous participation
And now:
participant with a past.
The difference between the last two is subtle but important.
A participant can be involved in a sequence of events without those events yet being construed as its history.
History begins when earlier participation is brought into the present as belonging to the continuing participant.
The past becomes part of the way the participant is currently construed.
And once that happens, something new becomes possible.
The participant can inherit its past.
It can be characterised by it.
It can be understood through it.
It can even be changed by how that past is construed.
We have therefore moved from persistence to history.
But there is another question waiting for us.
If a participant can carry a history from one moment into another, where does that history come from?
Is it simply the participant's own previous participation?
Or can a participant inherit a history from the discourse in which it is participating?
And if the latter is possible, then the history of a participant may not be quite as private—or as simple—as we ordinarily imagine.
The participant may enter the present carrying a history that has been constructed in language.
That is where we go next.
The participant does not merely have a past. It can inherit one.
No comments:
Post a Comment