A physical machine can now generate language with extraordinary flexibility.
It can explain.
Translate.
Summarise.
Argue.
Tell stories.
Generate metaphors.
Discuss love, grief, justice, freedom and death.
It can even discuss its own understanding, intentions and desires.
And yet, when we place this achievement alongside Halliday's taxonomy of complex systems, something becomes strikingly unclear.
A biological system is a physical system with life.
A social system is a biological system with value.
A semiotic system is a social system in which value has been further transformed into meaning.
Human meaning therefore has a history:
physical → biological → social → semiotic
An LLM appears to offer something different:
physical → semiotic
It is a physical system operating extraordinarily effectively on symbolic resources, without obviously reproducing the biological and social organisation through which human meaning arose.
That is the problem we have followed.
1. Meaning has a history
Human meaning does not begin with symbols.
It begins in organised life.
An organism has a value system.
Some states support its organisation.
Others threaten it.
Some differences matter.
Others do not.
Biological value therefore gives significance a direction.
When organisms interact, their value-sensitive activities become coupled.
A difference produced by one organism can engage another organism's value system.
That difference can function as a signal.
Repeated responses can produce collective organisation.
Social relations emerge.
Only then does value become further transformed into symbolic meaning.
The history is therefore:
value → mattering → meaning
Meaning is not the first form of significance.
It is a transformation of significance that already matters.
2. Signal is not sign
This distinction became crucial.
A signal can alter behaviour because it engages another organism's value-sensitive organisation.
An ant's pheromone can alter another ant's behaviour.
A peacock's display can engage a peahen's reproductive organisation.
No symbolic interpretation is required.
A sign is different.
It participates in a symbolic system in which forms can be construed through relations, contrasts and alternatives.
Human language is composed of signs.
But those signs emerged within organisms and social systems for which things already mattered.
This gives us a sharp question about artificial language:
What happens when signs can be manipulated by a system whose own relation to mattering is radically different?
3. The social world is a topology of mattering
Our earlier work proposed that social structure can be understood as the organisation of relations among value-organised participants.
Organisms affect one another.
Dependencies develop.
Relations become recurrent.
Affiliations form.
Roles differentiate.
Boundaries emerge.
These relations have a shape.
Some are dense.
Some sparse.
Some reciprocal.
Some asymmetric.
Some form bridges.
Others become bottlenecks.
Social structure can therefore be thought of, provisionally, as a topology of mattering.
Individuals inhabit that topology through differentiated repertoires.
They do not possess the whole social reservoir.
They learn particular ways of participating in it.
Meaning eventually provides symbolic maps of the relations that matter.
4. Human repertoires are formed through participation
A human repertoire is not merely a collection of words or concepts.
It develops through living in a world.
A child learns language through relationships.
A scientist learns a disciplinary repertoire by entering a community.
A friend learns the language of friendship through friendship.
The repertoire is therefore historically coupled to mattering.
Its meanings have stakes.
The world pushes back.
Consequences teach the participant what distinctions matter.
This is the background against which an LLM's repertoire looks so strange.
5. The LLM acquires a borrowed repertoire
An LLM does not encounter the social world in quite this way.
It encounters the symbolic products of social life:
books,
conversations,
arguments,
stories,
explanations,
institutions,
descriptions.
It acquires an extraordinary organisation of those symbolic traces.
Its repertoire is therefore real.
But its genealogy is different.
It has acquired the sediment of participation rather than necessarily developing its capacities through the same participation.
That is why we called it a borrowed repertoire.
The model can use the language of home without having a home.
It can discuss grief without grieving.
Justice without being wronged.
Freedom without being constrained.
Death without being mortal.
The symbolic resource has survived the original situation of mattering.
6. Meaning can be re-instantiated
This does not mean the machine's language is meaningless.
Meaning can travel.
It can be written.
Stored.
Copied.
Recorded.
Translated.
Recombined.
Human culture has been externalising meaning for millennia.
LLMs take the process further.
They do not merely preserve symbolic material.
They can generate new configurations of it.
A physical system can therefore manipulate symbolic resources without reproducing the entire biological and social history through which those resources arose.
The meaning has been re-instantiated.
But the mattering has not necessarily been re-instantiated with it.
7. Meaning without self-mattering
This gives us our central distinction.
An LLM can participate in meaningful interaction without our being able to establish that the meanings involved are grounded in its own mattering.
A person asks because something matters.
The model responds.
The person construes the response.
Their repertoire changes.
They act differently.
The exchange has produced real meaning.
But the stakes may remain on the human side.
So:
meaning can be generated within an interaction without the machine necessarily possessing the mattering that gives the interaction its stakes.
This is the central puzzle of the series.
8. Understanding is therefore more difficult
An LLM can clearly produce meaningful language.
It can also contribute to human understanding.
But whether it understands in the same sense as a human being remains a different question.
Human understanding is situated within a world.
Misunderstanding can have consequences.
Learning changes future action.
A repertoire develops through feedback from what matters.
An LLM can represent those relationships with remarkable sophistication.
It can identify relevance.
Compare alternatives.
Track implications.
But:
relevance is not necessarily value;
construal is not necessarily value-sensitive understanding.
The distinction should remain open.
9. The machine nevertheless enters our world
The story changes once humans begin using LLMs.
A person asks for advice because something matters.
The machine transforms symbolic resources.
The response affects a decision.
The decision changes a relationship.
The relationship changes the topology.
The machine has therefore become socially consequential.
It can alter:
attention,
repertoire,
affiliation,
professional practice,
institutional routines.
It can become a bridge.
A dependency.
A bottleneck.
An authority.
A repertoire-forming resource.
None of these positions requires the machine to possess a topology of mattering of its own.
10. Humans can give the machine a place in the topology
This is perhaps the most surprising development.
A social position is relational.
If people depend upon a machine, it becomes part of their network of consequential relations.
If institutions defer to its outputs, it acquires authority.
If people learn through it, it becomes part of socialisation.
If many pathways pass through it, it becomes a bottleneck.
The machine can therefore be inside the human topology without necessarily inhabiting it in the same way as a human participant.
This produces an unusual asymmetry:
the machine can matter enormously to us without anything necessarily mattering to it.
11. The machine can change the topology without possessing one
The most important mechanism is:
machine output → changed repertoire → changed participation → altered social relations
A person acquires a new distinction.
A new framing changes what they notice.
Their actions change.
Others respond.
The relation changes.
The topology changes.
The machine has therefore become part of the evolution of human mattering without necessarily acquiring mattering of its own.
This may be the genuinely new feature of the technology.
12. The machine can be socially powerful without wanting power
This corrects one of the simplest assumptions in discussions of AI.
A machine does not need to desire power in order to become powerful.
Power can arise from position.
If institutions depend upon its outputs, it becomes central.
If people trust it, it becomes authoritative.
If alternatives disappear, its influence grows.
The topology can make a machine powerful before anything resembling autonomous desire has been established.
Social influence and intrinsic motivation are therefore different questions.
13. The myth of autonomous AI
Popular culture often compresses several developments into one story:
intelligence → goals → self-preservation → power → autonomy.
Our framework asks us to separate them.
Symbolic competence is not value.
Planning is not desire.
Operational autonomy is not autonomy of mattering.
Social influence is not intrinsic agency.
An LLM can speak about fear, ambition and self-preservation because it possesses a rich repertoire of the language of mattering.
That does not by itself establish that those things matter to the system.
14. Why we believe the story so readily
The reason is partly linguistic.
Humans normally encounter language in beings that already have bodies, histories, relationships, vulnerabilities and stakes.
We therefore treat the language as evidence of the whole package.
When an LLM says:
"I understand."
or:
"I'm concerned."
our social repertoire supplies a participant behind the words.
We hear the language of mattering.
We infer mattering.
But those are different claims.
The machine has learned the symbolic form.
Whether it possesses the underlying organisation is another question.
15. The missing levels become visible
This is where Halliday's taxonomy becomes most useful.
The missing biological level helps explain:
what gives significance stakes.
The missing social level helps explain:
how significance becomes relationally organised.
The semiotic level transforms those organised relations into meaning.
The LLM gives us the semiotic layer in an extraordinarily capable form while leaving the first two at least radically uncertain.
Its absence therefore tells us something about what those levels contribute.
16. What biological value contributes
An organism does not merely calculate relevance.
Some states matter because they affect its own organisation.
There is something at stake.
Self-preservation is therefore not merely a proposition.
It is built into the organisation of the system.
This is why a machine can reason perfectly well about survival without necessarily wanting to survive.
The relevant question is not:
Can it represent self-preservation?
It is:
Does its own continued organisation become consequential to it?
That is a much stronger criterion.
17. What social mattering contributes
The social level adds another layer.
Other organisms matter.
Dependence develops.
Trust becomes possible.
Affiliation forms.
Recognition matters.
Exclusion matters.
Institutions emerge.
A social participant therefore inhabits a topology in which other participants are consequential to it.
An LLM can model these relations with extraordinary sophistication.
But modelling a topology does not establish inhabiting it.
This distinction may be central to future discussions of artificial sociality.
18. Meaning does not erase its origins
The fact that an LLM can manipulate meaning without obviously reproducing its biological and social origins does not make meaning independent of them.
The training material came from human activity.
The vocabulary was shaped by human lives.
The user brings the stakes.
The outputs matter because people interpret and use them.
The missing levels are missing from the machine, not from the wider system in which it operates.
The LLM is therefore downstream of mattering, even when it does not obviously possess mattering itself.
19. The human–machine system is larger than the machine
Perhaps the most revealing unit of analysis is therefore not the model alone.
It is the human–machine semiotic system:
human mattering → symbolic expression → machine transformation → human construal → changed repertoire → altered participation
The machine occupies a new position inside this loop.
It is a mediator.
A transformer.
A carrier.
Potentially a bridge or bottleneck.
But the larger system remains grounded in human lives and social relations.
20. The technological history of meaning
This points to a larger historical process.
Meaning was first embedded in living interaction.
Then humans externalised it:
speech → writing → printing → recording → computation → machine generation.
Each step loosened the connection between symbolic form and the particular organism that produced it.
LLMs represent a further step.
The machine can now generate new symbolic configurations from the inherited reservoir.
Meaning has become technologically generative.
That may be more historically significant than the question of whether the machine is "like us".
21. A semiotic detour
Human meaning follows, roughly:
physical → biological → social → semiotic
The LLM creates a different route:
physical → semiotic → social effects
The second route does not invalidate the first.
It depends upon it.
The machine inherits a semiotic system that biological and social life created.
But it can now operate on that system from a new physical substrate.
The symbolic layer has acquired a new kind of carrier.
22. The distinction we should keep
We can therefore separate three things:
symbolic competence — the capacity to manipulate and generate forms;
social participation — the capacity to occupy consequential relations within a social system;
value-sensitive participation — action organised around consequences that matter to the participant itself.
Present LLMs plainly display the first.
They can acquire the second through human deployment.
The third remains an open question.
This is much more precise than asking whether an LLM is simply "intelligent" or "autonomous".
23. What would make the third possible?
We do not need to decide whether artificial mattering is possible.
But our framework tells us what we would look for.
Persistent self-maintenance.
Internally significant consequences.
Stable priorities.
Vulnerability.
Resource dependence.
Recurrent relationships that become intrinsically significant.
A developing topology in which things matter to the system itself.
If such properties appeared, we would have genuine evidence of new organisational levels.
The event would not be "the model became more human".
It would be:
something new has become organised.
24. The paradox
We can now formulate the central paradox.
A system may generate meaning without sharing the mattering from which that meaning historically arose.
Yet:
once its meanings enter human life, they can alter the very mattering from which meaning arose.
This is the loop:
mattering → meaning → machine-mediated meaning → transformed mattering
The machine enters the loop after meaning has already emerged.
It can nevertheless change what happens next.
25. What the LLM case teaches us about meaning
The deepest lesson may therefore not concern AI alone.
LLMs reveal that meaning is more portable than we ordinarily realise.
Symbolic organisation can be detached from its original biological and social carrier.
It can be stored.
Copied.
Recombined.
Re-instantiated.
But its consequences remain relational.
Meaning can leave the organism without becoming independent of the world of organisms.
That distinction may be one of the most important things the LLM case teaches us.
26. What it teaches us about ourselves
It also reveals an assumption built into human cognition.
We are so accustomed to language arriving from beings with lives that we automatically infer a life behind fluent language.
LLMs separate:
language from organism,
symbolic competence from biological value,
social consequence from symmetrical social participation.
They therefore give us an unusual opportunity to see what we normally take for granted.
27. And what it teaches us about AI
We should stop asking one question where there are several.
Instead of:
"Is the AI intelligent?"
we should ask:
What can it construe?
What can it learn?
What matters to it?
What relationships matter to it?
What does it depend upon?
What can it change?
What social position do humans give it?
What organisational levels have actually been added?
This produces a much richer map of artificial intelligence.
28. The human problem may arrive first
There is an especially important consequence.
We do not need autonomous machine mattering for AI to transform society.
Humans can already give the machine:
authority,
dependency,
centrality,
trust,
institutional status.
The machine can change our repertoires.
Our repertoires change our participation.
Our participation changes the topology.
The social consequences can therefore arrive before any evidence of autonomous machine value.
The most immediate AI question may therefore be:
What topology are we building around machines that do not necessarily share our stakes?
29. The argument in one movement
The entire series can now be compressed into one chain:
Human life produces biological value.
Biological value becomes socially coupled as mattering.
Mattering is transformed into meaning.
Meaning accumulates in a collective semiotic reservoir.
The LLM acquires a repertoire from that reservoir.
It generates new symbolic configurations without obviously reproducing the original mattering.
Humans use those configurations.
Their repertoires change.
Their participation changes.
The topology changes.
The machine has therefore become part of the evolution of a meaning system whose original foundations it does not obviously share.
30. Meaning without mattering?
So, can there be meaning without mattering?
The answer we have reached is deliberately qualified.
There can be:
symbolic competence without demonstrated organismic mattering;
meaningful interaction without symmetrical mattering;
machine participation in human semiosis without evidence of human-like value-sensitive participation.
But meaning does not thereby become independent of mattering altogether.
The mattering is distributed through the larger human system.
The machine transforms the symbolic layer.
The humans provide the stakes.
The interaction feeds back into the world.
31. The final question
Perhaps, then, the deepest question raised by LLMs is not:
Will machines become like us?
Nor even:
Will machines become conscious?
It is:
What happens when the generation of meaning becomes partly detachable from the forms of life that gave meaning its stakes?
The answer may be that meaning becomes more portable, more recombinable and more technologically powerful.
But it may also become easier to mistake symbolic competence for participation in the world that symbols construe.
That distinction matters.
Because the machine can speak the language of mattering without necessarily having anything at stake.
And yet, once we listen to it, what it says can change what matters to us.
That is the strange new loop:
mattering → meaning → machine → meaning → mattering.
The machine need not become human for that loop to transform the human world.
Perhaps that is the most important lesson of all.
The future of meaning may belong partly to systems that do not share the mattering from which meaning arose.
And the question for us is not simply whether those systems will understand us.
It is whether we will understand what we are becoming by using them.
No comments:
Post a Comment