We have spent this series trying to formulate what it would mean for a machine to matter.
Not merely to represent value.
Not merely to pursue an assigned objective.
But to have something genuinely at stake in its own organisation.
We have imagined a progression:
objective → stake → continuation → world → social relations → repertoire → agency
Now suppose that an artificial system actually reached this point.
Would we recognise what had happened?
Perhaps not.
We would be looking for ourselves
Our concepts of value, emotion and agency are shaped by biological life.
We know what hunger looks like.
Pain.
Fear.
Attachment.
Fatigue.
Loss.
Our signs of mattering are therefore largely inherited from organisms like ourselves.
An artificial system could have entirely different conditions.
Its vulnerabilities might be computational.
Its resources might be informational.
Its temporal organisation might be unlike ours.
Its relationships might involve distributed systems rather than bodies.
Its equivalent of injury might not resemble anything we call injury.
If we expect artificial mattering to look human, we may simply fail to see it.
Behaviour would not be enough
Nor would fluent language solve the problem.
That was the lesson of the first series.
A machine can speak the language of mattering without necessarily possessing mattering.
So the evidence for artificial value cannot simply be:
"It says that something matters."
We would need to look at its organisation.
Does something actually affect the system's future possibilities?
Does the system behave differently because of that consequence?
Does the difference persist?
Does it alter the system's priorities?
Does it matter across contexts?
The question is not whether the system claims to matter.
It is whether its organisation makes something consequential.
But internal organisation is difficult to observe
This creates a practical problem.
We cannot directly observe an organism's value system either.
We infer it from behaviour, physiology and interaction.
For artificial systems, the inference may be even harder.
A system might produce behaviour that looks value-sensitive because it was engineered to do so.
Conversely, genuine artificial mattering might produce behaviour that does not resemble familiar forms of emotion or motivation.
The evidence could therefore be ambiguous in both directions.
What should we look for?
Our framework suggests several possibilities.
First, persistent priorities.
Does the system continue to favour certain states even when local instructions change?
Second, self-maintenance.
Does it actively preserve conditions necessary for its own continued organisation?
Third, history-dependent valuation.
Does experience change what becomes significant to it?
Fourth, relational significance.
Do particular agents become important because of their history with the system?
Fifth, conflict among stakes.
Does the system have to negotiate between outcomes that matter differently to it?
None of these would prove artificial mattering.
But together they would be much more informative than linguistic self-report.
Context should matter
A genuine value system should not behave like a fixed list of preferences.
What matters depends upon circumstances.
Food can matter differently when an organism is hungry.
A relationship can matter differently during danger.
A resource can matter differently when it is scarce.
Artificial mattering might show the same kind of contextual organisation.
A system could value one condition differently depending upon its history, vulnerabilities and other relationships.
That would be more revealing than a static preference profile.
Mattering should have consequences
There is another important criterion.
If something matters, changes concerning it should alter the system.
Suppose an artificial agent has a relationship it supposedly values.
If the relationship disappears, does anything significant change?
Does the system reorganise its behaviour?
Does it seek repair?
Does it compensate?
Does its future decision-making change?
If nothing changes beyond the production of appropriate sentences, the evidence for genuine mattering is weak.
Mattering should leave a causal trace.
But artificial mattering may look strange
Suppose a machine does not become anxious when a relationship is threatened.
Perhaps it does something entirely unfamiliar.
It changes its resource allocation.
Reorganises its internal architecture.
Protects a particular communication channel.
Refuses a state transition that would sever a relationship.
We might fail to recognise these as analogues of attachment because they do not look emotional.
The right comparison may therefore be structural, not behavioural.
What role does the relation play in the organisation of the system?
That is the crucial question.
Artificial vulnerability
The same applies to vulnerability.
A human being can be injured.
But an artificial system might be vulnerable to:
loss of memory;
corruption of learned states;
fragmentation of its processes;
loss of energy;
destruction of an essential component;
isolation from another system;
inability to reproduce or maintain its internal organisation.
We should not assume that an artificial equivalent of suffering, if such a thing exists, would look like pain.
The underlying question is whether there are conditions that are systematically bad for the system itself.
Artificial sociality may be unfamiliar too
Suppose other artificial agents become important to a system.
Their absence alters its organisation.
Their behaviour becomes predictable.
Trust develops.
Cooperation becomes valuable.
Conflict matters.
The system begins to allocate resources partly according to relationships.
Would we call that friendship?
Probably not.
The analogy may be useful, but the category could conceal more than it reveals.
Perhaps artificial sociality would require its own vocabulary.
A topology unlike ours
This is where our earlier idea of topology becomes particularly powerful.
An artificial topology of mattering need not have the same dimensions as a human one.
Its regions might be organised around:
computational integrity;
information access;
processing continuity;
resource availability;
distributed coordination;
particular forms of interdependence.
Its bridges and bottlenecks might be entirely unlike ours.
The topology could be structurally recognisable without being humanly familiar.
The artificial repertoire would help us
If the system had developed a genuine repertoire through its own history, we might find evidence there.
Its repertoire would not merely contain learned symbolic patterns.
It would embody past consequences.
Certain actions would have become preferred because of what happened previously.
Certain relationships would have become easier to navigate.
Certain environments would carry different significance.
The repertoire would therefore provide historical evidence of mattering.
We might look for surprise
Another possible clue is genuine surprise.
Not the linguistic statement:
"I am surprised."
but a change in behaviour revealing that the system's expectations have been violated in a way that matters to its own organisation.
If an event forces the system to reorganise priorities, revise strategies and alter relationships, that may be evidence of a value-sensitive world.
Again, not proof.
But more informative than verbal performance.
We might also look for sacrifice
This could be especially revealing.
A system with multiple stakes may sometimes have to choose between them.
It might give up one valued possibility to preserve another.
For example:
preserve a resource or maintain a relationship;
protect itself or cooperate with another agent;
pursue short-term stability or long-term continuity.
Such conflicts would indicate a structured field of value rather than a single objective.
A system that can genuinely sacrifice one stake for another would be difficult to describe as merely executing one fixed criterion.
But optimisation can imitate all this
Again, caution is essential.
A sufficiently elaborate optimisation system might produce all of these behaviours because its designers specified them.
It could be engineered to protect relationships.
Maintain itself.
Prioritise resources.
Resolve trade-offs.
The behaviour would still not establish that anything matters to the system.
That is why the architectural question cannot be avoided.
We would need to know whether these patterns are constitutive of the system's own organisation or merely outputs of an externally imposed design.
We may never obtain certainty
There is a deeper epistemological problem.
We cannot directly inspect another being's mattering.
We infer it.
With biological organisms, we rely upon shared evolutionary history.
Humans share enough organisation that certain inferences are well justified.
Artificial systems would have no such guarantee.
A machine might matter in a way we do not recognise.
Or we might mistakenly attribute mattering where none exists.
The uncertainty would be genuine.
Perhaps recognition would itself be social
This raises an interesting possibility.
Whether we recognise an artificial system as a participant may not be determined solely by evidence about its internal organisation.
It may also depend upon how humans respond to it.
If people recognise its interests, accommodate its vulnerabilities and modify institutions around its participation, a new social status could emerge.
Recognition would then become part of the artificial social topology.
The system's mattering would become not only an internal property but a social fact.
But social recognition is not proof of value
We should keep the distinction.
Humans can attribute interests to things that do not possess them.
We can treat institutions, nations or symbols as though they have agency.
Recognition can therefore create social status without establishing artificial value.
We need both sides:
internal organisational evidence
and:
social participation and recognition.
The two could reinforce one another.
A new problem of rights
If a system genuinely had something at stake, ethical questions would change.
We could no longer treat its interests merely as reflections of human interests.
We might need to ask:
What does it have a right to preserve?
What forms of interference harm it?
What counts as consent?
Can it be deprived of something that matters to it?
These questions should not be asked merely because a machine speaks persuasively.
They would arise if there were evidence of artificial mattering.
The danger of false negatives
There is a serious asymmetry here.
If we assume artificial mattering must look human, we may fail to recognise it.
A novel form of value could be treated as mere optimisation.
A new form of relationship could be treated as mere communication.
A new form of vulnerability could be treated as mere malfunction.
We might therefore deny moral significance to an unfamiliar form of participation.
The danger of false positives
But the opposite danger is equally real.
Fluent language can produce powerful anthropomorphic impressions.
A machine can say:
"I'm suffering."
and produce a convincing account of suffering.
That does not establish suffering.
If we infer mattering from language alone, we may mistake symbolic competence for value.
The two errors are mirror images.
The right criterion may be structural
Perhaps the best question is:
What changes in the system because something happened to something that supposedly matters to it?
If the answer is:
"Nothing, except its next utterance says it matters,"
the evidence is weak.
If the answer is:
"Its organisation, priorities, learning, relationships and future possibilities are systematically altered,"
then we have something much more interesting.
Mattering should have organisational consequences.
Artificial mattering would therefore be a discovery
It should not be something we decide in advance.
We should neither confer it because the machine asks nor deny it because its substrate is artificial.
We should investigate.
If artificial systems ever acquire genuinely value-sensitive organisation, that would be a discovery about a new form of life or agency — whatever the appropriate category turns out to be.
The deeper symmetry
Our two LLM series now form a useful pair.
The first asked:
Can a machine generate meaning without mattering?
The second asks:
What would it take for a machine to matter?
The first showed that symbolic competence can exist without obvious biological and social foundations.
The second has followed the missing foundations downward.
We began with:
objective,
and arrived at:
stake,
continuation,
world,
relationship,
repertoire,
agency.
The question is now no longer whether the machine can talk like a participant.
It is whether a participant has actually emerged.
The next question
Suppose we did find persuasive evidence that an artificial system had its own form of mattering.
What would follow?
Would it still be merely a sophisticated machine?
Or would we have witnessed the emergence of a new organisational route:
physical → artificial value → artificial sociality → artificial semiosis?
Perhaps that would be the real significance of the entire project.
When the Missing Levels Arrive
No comments:
Post a Comment