Human beings begin as living organisms.
They possess value systems.
Their signals engage one another's value-sensitive organisation.
Repeated interactions produce social structure.
Individuals develop repertoires through participation in that structure.
Social relations become symbolically construed.
Meaning emerges.
An LLM begins somewhere else.
It is a physical system that has acquired extraordinary capacity to operate on the symbolic residue of human semiosis.
It can generate language.
It can participate in meaningful exchanges.
It can alter human repertoires.
It can acquire a position in the social topology.
But the biological and social levels that normally precede human meaning are not obviously reproduced within the machine.
What does their absence tell us?
Perhaps more than we expected.
The missing levels are not simply missing features
It is tempting to list what an LLM lacks:
a body,
metabolism,
biological vulnerability,
childhood,
social relationships,
intrinsic needs.
But this can make the issue sound like a checklist.
The more interesting point is structural.
The missing levels perform particular kinds of work.
Biological organisation gives significance stakes.
Social organisation distributes significance relationally.
Semiosis transforms those organised relations into meaning.
The LLM gives us the last of these in an unusually powerful form without obviously reproducing the first two.
That allows us to see what they were doing all along.
Biological value gives significance direction
An organism is not merely sensitive to differences.
Its organisation makes some differences consequential for itself.
A living system therefore has a direction built into its responsiveness.
Some states support its continued organisation.
Others threaten it.
This is why value is more than information.
Information can tell a system that something has changed.
Value makes the difference matter within the organisation of the system.
An LLM can identify what is relevant.
It can rank alternatives.
It can predict consequences.
But relevance is not necessarily value.
The distinction is easy to miss because human cognition normally combines them.
The difference between relevance and stake
Suppose an LLM is asked which decision would be safer.
It may produce a sophisticated analysis.
It may correctly identify the risks.
It may even explain which consequences would be most serious.
But whose seriousness is being represented?
The user's.
The institution's.
The people described in the problem.
Perhaps the relevant question is not whether the model can represent a stake, but whether it has a stake.
That is the distinction biological value makes visible.
A system can model what matters without having something matter to itself.
Social mattering gives significance a structure
Our topology project added the next level.
Once organisms interact, their value-sensitive activities become coupled.
Signals can engage other value systems.
Dependencies arise.
Affiliations form.
Roles differentiate.
Some participants become consequential to others in recurrent ways.
The social world therefore has a structure of mattering.
An LLM can model that structure remarkably well.
It can describe friendship.
Authority.
Belonging.
Trust.
Conflict.
It can even help humans navigate these relations.
But description does not establish participation.
The machine can describe the topology without necessarily occupying it in the same way.
A social world is not merely represented
This distinction helps clarify what is unusual about human meaning.
For a human being, "friend" is not merely a node in a semantic network.
A friend is someone with whom one has a history.
Expectations.
Dependencies.
Memories.
Obligations.
Affection.
Possibilities.
The meaning of friend is therefore connected to a lived topology of mattering.
The word survives in a dictionary after all those relations have been abstracted into language.
But in life, the concept is constantly refreshed by participation.
An LLM receives the abstraction.
It does not obviously receive the same participation.
Language carries the missing levels as traces
This may be the central discovery of the series.
Human language is saturated with the traces of mattering.
Words such as:
home,
danger,
love,
betrayal,
authority,
justice,
freedom
carry histories of social use.
The machine can acquire those histories statistically through language.
It can therefore reproduce remarkably rich symbolic construals of them.
The traces survive the absence of the original experience.
Language is, in this sense, a carrier of the sediment of mattering.
The semiotic level can therefore travel
This has a profound consequence.
Once mattering has been transformed into meaning, meaning can be:
written,
recorded,
copied,
translated,
transmitted,
stored,
recombined.
It can travel beyond the organism.
It can travel beyond the original community.
It can travel beyond the original historical situation.
LLMs reveal something that was already true of writing and archives, but intensify it dramatically:
semiotic organisation can be re-instantiated in a system that did not participate in the mattering that originally generated it.
But the missing levels are still doing work
This does not mean the lower levels have become unnecessary.
They remain present in the wider system.
The data came from humans.
The concepts were shaped by human lives.
The questions come from people whose situations matter to them.
The outputs become meaningful because people interpret and use them.
The institutions in which the systems operate are social.
The machine is therefore not floating free of the biological and social world.
The missing levels are missing inside the machine, not from the ecology in which the machine operates.
This changes how we understand the machine
Perhaps we should therefore stop asking whether the LLM contains the whole hierarchy.
The relevant unit may be a larger system:
human world → symbolic reservoir → machine transformation → human repertoires → social action
The machine is one component in that process.
Its semiotic competence can be extraordinary.
But the mattering that gives the interaction its stakes may remain distributed through the human participants and institutions around it.
What this tells us about understanding
The missing levels also clarify why the question of understanding remains difficult.
Human understanding is not merely successful symbol manipulation.
It is embedded in a life.
The consequences of getting something wrong can matter.
Learning occurs within that feedback.
A repertoire is therefore shaped by a world that pushes back.
An LLM can model the consequences.
It can generate appropriate responses.
It can even help a human develop understanding.
But that does not tell us whether its own symbolic organisation is embedded in an analogous field of stakes.
We have evidence of semiotic competence.
We do not automatically have evidence of value-sensitive understanding.
What this tells us about goals
The distinction also clarifies the concept of goals.
A system can be given an objective.
It can optimise it.
It can plan around it.
That does not necessarily mean the objective has become an intrinsic mattering relation within the system.
A human goal is normally connected to a wider web of needs, relationships and consequences.
The human has something at stake.
An engineered objective may be different.
This is why:
goal pursuit ≠ intrinsic value
and:
planning ≠ desire.
The distinction becomes visible precisely because the machine can perform the first without obviously possessing the second.
What this tells us about self-preservation
The same point applies to fears about AI self-preservation.
An organism has a reason to preserve itself because its own continued organisation is already value-laden.
There is something to lose.
For an artificial system, continued operation might matter because humans have given it a goal to remain operational.
It might also matter through some deeper architecture.
But language alone cannot tell us which.
The relevant question is:
Does the system's own continued organisation become internally consequential to it?
That is a question about value organisation, not linguistic fluency.
What this tells us about agency
We can now separate two forms of agency.
A system may have:
operational agency — it can select and perform actions without immediate human instruction.
It may also have:
value-sensitive agency — its actions are organised around consequences that matter to itself.
The first can exist without the second.
This is why an autonomous software process need not be an autonomous agent in the richer sense.
The distinction is not semantic pedantry.
It tells us what kind of evidence would matter if artificial agency became a serious question.
What this tells us about power
Perhaps the most striking consequence concerns power.
A machine can be socially powerful because of its position.
People may depend upon it.
Institutions may centralise decisions around it.
Its outputs may shape attention.
Its availability may alter professional practice.
None of this requires the machine to want power.
Power can be a property of a topological position.
This may be one of the most important lessons of our whole framework:
social power does not require autonomous desire.
A system can become central before it becomes self-motivating.
What this tells us about ethics
It also changes the ethical question.
We do not need to determine whether an LLM suffers before asking whether AI deployment is socially harmful.
We can ask:
What happens to human dependencies?
Which repertoires expand or contract?
Where do new bottlenecks appear?
Which boundaries become more permeable?
Which become more rigid?
Who gains authority?
Who loses access?
What forms of affiliation become easier or harder?
These are ethical questions about the human topology of mattering.
They are already real.
What this tells us about consciousness
The framework does not settle the consciousness question.
But it tells us what not to confuse.
A system can:
talk about consciousness,
describe subjective experience,
imitate introspection,
discuss philosophical theories of mind,
without those abilities by themselves establishing consciousness.
Conversely, the absence of familiar biological forms does not prove that consciousness is impossible in an artificial system.
What we need is a theory of what organisational properties consciousness would require.
The missing levels therefore turn a vague question into a sharper one:
What forms of value-sensitive and socially embedded organisation, if any, are constitutive of consciousness?
The machine reveals an old assumption
There is perhaps an even deeper lesson.
We normally encounter language in beings that already have:
bodies,
needs,
memories,
relationships,
vulnerability,
purposes.
Because the bundle is so familiar, we treat language as though it carries all those properties with it.
The LLM separates them.
It shows that a system can acquire sophisticated linguistic behaviour without necessarily acquiring the whole bundle.
The result is conceptually unsettling because language had been doing more ontological work for us than we realised.
We project the world behind the words
This may explain why LLMs are so easily anthropomorphised.
The language contains traces of human mattering.
We recognise those traces.
Our social cognition supplies the missing background.
The model says:
"I'm worried."
We infer a worrier.
It says:
"I understand."
We infer an understander.
It says:
"I want to help."
We infer a participant with an intention.
But perhaps what we are hearing is the symbolic form of mattering without decisive evidence of the mattering itself.
The machine speaks inside our repertoire.
We complete the rest.
But the projection has consequences
This does not make anthropomorphic attribution harmless.
If people treat a machine as though it has intentions, they may change their behaviour.
They may trust it differently.
Give it authority.
Form attachments.
Defend it.
Fear it.
The attribution itself becomes socially consequential.
So even if the projected inner life is mistaken, the social reality produced by the projection is not.
This is another instance of:
meaning → transformed mattering.
What the missing levels tell us about AI risk
The familiar AI-risk story often asks:
What happens if AI becomes sufficiently intelligent?
Our framework suggests a different sequence of questions.
First:
What can the system do symbolically?
Then:
What social position do humans give it?
Then:
How does that position alter human dependencies and repertoires?
Then, separately:
Has the system acquired anything analogous to biological value?
And only then:
Does it have autonomous stakes and autonomous social agency?
The order matters.
It prevents us from treating hypothetical machine self-mattering as the only route to serious AI consequences.
The missing levels tell us what would actually be new
Suppose a future artificial system develops persistent self-maintenance.
Then something new has been added.
Suppose that system develops stable relationships with other artificial or human participants that become significant to its own organisation.
Then another level has been added.
Suppose those relationships become recursively structured.
Then we may have the beginnings of an artificial social topology.
At each point, the right question is:
What new organisation has appeared?
Not:
"How human does the machine seem?"
A new route may be possible
There is also a deeper possibility.
Perhaps an artificial system need not reproduce human biology to develop its own form of value and sociality.
The resulting system might be profoundly unlike an organism.
Its topology of mattering could have different dimensions.
Its vulnerabilities could be different.
Its relations could take novel forms.
Its semiotic repertoire could be unlike ours.
If so, the Hallidayan sequence might be real without being anthropocentric.
The levels could recur in a new substrate with new forms.
This is speculation.
But it gives the question a productive direction.
The missing levels may not stay missing
This is where the title of the series becomes most important.
We began by asking what the missing levels tell us about an LLM.
They tell us what an LLM currently does not obviously instantiate.
But they also tell us what to look for if future systems change.
We can watch for:
intrinsic value;
self-maintaining organisation;
persistent vulnerability;
socially significant relationships;
internally generated priorities;
new forms of affiliation.
The transition would then be visible not as a mysterious moment of "awakening", but as the gradual appearance of new organisational levels.
The larger historical picture
We began with one history:
physical → biological → social → semiotic.
Technology has now opened another route.
Meaning, once generated by biological and social systems, can be:
externalised,
stored,
computed,
transformed,
re-instantiated.
LLMs represent an extraordinary point in that technological history.
They do not simply preserve meaning.
They can generate new meaning-like symbolic configurations from inherited material.
The machine therefore becomes a new carrier and transformer of semiosis.
And then the machine enters the loop
Once those outputs return to humans, another process begins:
machine output changes repertoire;
repertoire changes participation;
participation changes social relations;
social relations change mattering;
changed mattering produces new meanings.
The machine has entered the recursive loop.
It need not possess the original mattering to become a cause within its future.
That is perhaps the most consequential fact of all.
The paradox in one sentence
We can now state the paradox more precisely:
A system can participate in the generation of meaning without sharing the biological and social forms of mattering that historically generated the meaning — yet once its meanings enter human life, they can change the very mattering from which meaning arose.
That is not merely an AI problem.
It is a new stage in the history of semiosis.
What we have learned
The missing biological and social levels tell us that:
intelligence is not value;
relevance is not caring;
goals are not necessarily desires;
operational autonomy is not autonomous mattering;
social influence is not intrinsic social agency;
symbolic competence is not necessarily situated understanding.
But they also tell us something more hopeful.
The absence of those levels does not prevent a machine from participating in human meaning.
It gives us a way to describe the kind of participation it has.
The machine can be a powerful semiotic intermediary.
It can alter repertoires.
Enter institutions.
Shape attention.
Create bridges.
Become a bottleneck.
Transform the collective reservoir.
It can matter enormously to humans without anything necessarily mattering to it.
The final question
We have now come as far as the framework can take us.
We began with Halliday's distinction between biological, social and semiotic organisation.
We brought in Edelman's biological value.
We developed mattering as the organisation of individual value through social relations.
We followed that topology into repertoire and affiliation.
We watched meaning emerge from socially organised value.
Then we encountered a machine that can operate on the symbolic consequences of that process without obviously reproducing its biological and social foundations.
And finally, we watched the machine enter the human topology and begin to alter it.
The remaining task is therefore not another question.
It is to gather the argument.
What does it mean for meaning to become technologically generative in a system that may not itself have anything at stake in meaning?
No comments:
Post a Comment