tegwick
3fbcca209d
Integrate research/foundations notes 001–005 into Terminology, Claims, bibliography, concept notes, thought experiments, and open questions. Write phase-i-foundations-synthesis.md, finish SOUL-WP-0002, and mark SOUL-WP-0003 ready for interaction research.
47 KiB
Executable file
Research Note 006 — Social Mind and Co-Adaptive Model Building
How Beings Construct, Revise, and Act Through Models of One Another
Status: Exploratory research note
Research cluster: C — Social Mind and Co-Adaptive Model Building
Soul Frame concepts: Mind, Being, self-model, other-model, relationship, common ground, recognition, personhood, social Beinghood
Builds on: Cluster A — Autonomous Individuality; Cluster B — Identity and Continuity
Primary objective: Ground and refine the Soul Frame concept of co-adaptive model building.
1. Research question
Clusters A and B established two foundational questions:
What makes an informational process an autonomous individual?
and:
What makes that individual continue as itself through time?
Cluster C turns toward interaction.
Its central question is:
What happens when one continuing Being encounters another continuing Being and each begins adapting to the other?
Soul Frame has provisionally described Mind as:
adaptive model building and control
and social interaction as:
co-adaptive model building.
The research now asks whether this formulation survives comparison with work on:
- theory of mind,
- social cognition,
- symbolic interaction,
- common ground,
- dialogue,
- joint attention,
- joint action,
- shared intentionality,
- participatory sense-making,
- second-person cognition,
- Bayesian inverse planning,
- and recursive multi-agent reasoning.
The preliminary result is strongly positive, but requires an important refinement:
Social Mind cannot be reduced to one Being privately constructing a model of another.
Social understanding also emerges through the interaction process itself.
2. Theory of Mind establishes the basic other-model
Premack and Woodruff's influential 1978 formulation introduced theory of mind in terms of attributing mental states to oneself and others and using those inferred states to explain or predict behavior. The mental states involved may include such things as goals, intentions, knowledge, belief, doubt, and preference.
This provides a clear precedent for the Soul Frame concept:
Other-Model
A Being may maintain a model of another Being containing inferred states such as:
beliefs
knowledge
goals
preferences
intentions
capabilities
uncertainties
dispositions
The model does not provide direct access to the other's subjective experience.
It is an inferential structure used to understand and anticipate the other.
This aligns well with Soul Frame's pragmatic position on consciousness:
another Being's internal experience may remain epistemically private while useful models of that Being remain possible.
3. Other-modeling can be understood as inverse planning
Baker, Saxe, and Tenenbaum give the other-model idea a computational formulation.
Their inverse planning approach models observers as inferring hidden goals and beliefs from observed actions under an assumption that agents tend to plan approximately rationally relative to their beliefs and objectives.
The forward direction is:
beliefs + goals
↓
planning
↓
action
The observer solves approximately the inverse problem:
observed action
↓
infer plausible
beliefs + goals
Soul Frame can generalize this:
A social Mind treats another Being's behavior as evidence about an underlying model of that Being.
The model may include much more than rational goals:
- habits,
- character,
- ignorance,
- emotional state,
- relationships,
- values,
- commitments,
- communication style.
Thus, inverse planning provides one formal mechanism for other-modeling, not a complete theory of social understanding.
4. Machine other-models are already technically feasible
Rabinowitz and colleagues demonstrated a machine-learning system, ToMnet, that learned models of other artificial agents from behavioral observations and used those models to predict behavior and latent agent characteristics in simple environments.
Other computational work has likewise used explicit models of conversational partners to improve adaptation in language-coordination tasks.
This is relevant because it separates two artificial capabilities:
language competence
≠
other-modeling competence
An artificial Being may communicate fluently while maintaining a poor differentiated model of the particular Being with whom it is communicating.
Conversely, a relatively simple system may model another agent effectively within a narrow task.
Soul Frame should therefore treat Other-Model Quality as a distinct dimension of Social Mind.
5. Social modeling becomes recursive
Once A models B and B models A, another possibility appears:
A can model how B models A.
This generates recursive structures:
A's model of B
A's model of
B's model of A
A's model of
B's model of
A's model of B
Interactive POMDPs formalize closely related structures by allowing an agent's belief state to contain models of other agents whose own states may contain beliefs about other agents. The resulting belief hierarchy can become arbitrarily nested, creating substantial computational complexity.
This produces an important Soul Frame result:
Recursive social modeling is structurally important but practically bounded.
A sophisticated Social Mind need not model infinite recursion.
It needs enough recursive depth for the interaction at hand.
6. Model depth should therefore be contextual
Consider:
Level 0
A models B's behavior.
Level 1
A models B's goals.
Level 2
A models what B believes about A.
Level 3
A models what B believes A expects B to do.
Different contexts require different depths.
Ordering coffee may need little recursion.
Negotiation may require considerably more.
Poker deliberately exploits recursive expectation.
Diplomacy, deception, teaching, intimacy, and strategic conflict can require deeper models.
Thus Soul Frame should treat:
recursive model depth
as a variable cognitive resource rather than a prerequisite of Beinghood.
7. Mead makes the model reflexive
George Herbert Mead provides one of the deepest predecessors to the Soul Frame Social Mind.
In Mind, Self, and Society, Mead describes the individual's capacity to take the attitude of others, anticipate their responses, and incorporate organized social responses into the individual's own conduct. In his account, this social process contributes directly to the structure of Mind and Self.
This implies something stronger than:
A models B
It introduces:
A models
how B responds to A
and ultimately:
A models
how a community of others
would respond to A
Mead's generalized other captures the incorporation of organized group attitudes into the individual's self-related conduct.
For Soul Frame, this suggests:
A social self-model includes models of how the self appears within the models of others.
This is a profound refinement.
8. Self-model and other-model are therefore coupled
A purely individualistic architecture might represent:
SELF
OTHER
WORLD
as three independent models.
The Mead-inspired picture is different:
SELF
│
│ influenced by
▼
MODEL OF OTHER'S
MODEL OF SELF
│
▼
SOCIAL SELF
Thus a person's self-understanding can depend partly upon:
- expected recognition,
- reputation,
- social roles,
- community norms,
- relationship history,
- and predictions about how others interpret their conduct.
Soul Frame should therefore reject:
Self-Model = purely internal autobiography
in favor of:
Self-Model = self-representation partly shaped by interaction and reflected social models.
This does not mean that systemic identity is socially created in its entirety.
Cluster A still grounds autonomous individuality in the organization of the Being.
Rather:
social interaction becomes constitutive of important parts of personhood and self-understanding.
9. Theory of Mind is not the whole of social cognition
The traditional theory-of-mind picture can imply:
A
observes
B
↓
A privately infers B's hidden state
Interactionist and enactive theories challenge the sufficiency of that picture.
De Jaegher and Di Paolo's concept of participatory sense-making begins from interaction between autonomous individuals and argues that the coordination process can develop a degree of autonomy of its own. Meaning can then be generated or transformed through participation in that interaction rather than residing exclusively inside either individual.
De Jaegher, Di Paolo, and Gallagher subsequently argued that social interaction can play more than a contextual role in social cognition: interaction processes may complement and in some circumstances partly replace individual mechanisms.
This is an important correction to Soul Frame.
10. Co-adaptive model building is not model exchange
The phrase co-adaptive model building could be misunderstood as:
A has model X.
B has model Y.
A tells B X.
B tells A Y.
eventually:
X = Y
That is too simplistic.
A more realistic interaction is:
A acts
↓
B experiences A's action
↓
B updates
↓
B responds
↓
A experiences B's response
↓
A updates
↓
interaction pattern itself changes
↓
both encounter a new situation
The models do not merely describe the interaction.
They partly cause the next interaction, which then becomes new evidence for changing the models.
Thus:
the observer changes what is observed by acting on the basis of the observation.
This gives co-adaptive model building a cybernetic structure.
11. The interaction process must become a first-class object
Participatory sense-making suggests that Soul Frame should represent not only:
Being A
Being B
but also:
Interaction(A,B)
The interaction can develop patterns such as:
- turn-taking,
- synchrony,
- escalation,
- avoidance,
- repair,
- deference,
- competition,
- teaching,
- flirting,
- ritual,
- argument.
These patterns can constrain both participants.
Second-person approaches to social neuroscience similarly argue that directly engaging with another person differs importantly from merely observing them and call for studying social cognition within real-time interaction.
Soul Frame should therefore model:
the relationship between Beings as a dynamic process, not just two static records pointing at one another.
12. Interaction does not require perfect mental-state inference
This leads to a useful caution.
Social coordination may sometimes succeed through:
- embodied cues,
- timing,
- direct perception,
- learned routines,
- interaction dynamics
without requiring deep explicit inference about hidden mental states.
Interaction-theoretic work explicitly challenges the claim that understanding others is always primarily a matter of internal mindreading or simulation.
Therefore:
Other-models may be explicit, implicit, distributed, or partially realized through interaction.
Soul Frame's use of the word Model must remain broad enough to include action-guiding structure rather than requiring a symbolic internal representation.
This parallels the qualification already introduced in Cluster A around the Good Regulator tradition.
13. Dialogue creates alignment
Pickering and Garrod's interactive alignment account describes dialogue as involving alignment between interlocutors at multiple levels of linguistic representation. Their theory treats dialogue as a highly interactive process rather than two isolated monologues.
This provides one mechanism through which interacting Minds become mutually adapted.
But Soul Frame should distinguish:
alignment
from:
co-adaptation.
Alignment is one possible result.
Co-adaptation is the broader process.
Two Beings can become increasingly adapted to one another while:
- disagreeing,
- competing,
- deceiving,
- resisting,
- differentiating roles.
A chess grandmaster becomes better adapted to an opponent precisely in order to defeat them.
Thus:
[ CoAdaptation \not\Rightarrow Agreement ]
and:
[ CoAdaptation \not\Rightarrow Cooperation ]
14. Communication builds common ground
Clark and Wilkes-Gibbs demonstrated that referring expressions in dialogue are constructed collaboratively rather than being fully specified by one speaker and independently decoded by another.
Clark and Brennan's subsequent theory of grounding in communication treats communication as collective activity requiring coordination of both content and process and emphasizes the accumulation of common ground.
This gives Soul Frame another important structure:
Common-Ground Model
A Common-Ground Model represents what a Being currently treats as sufficiently mutually established for the interaction.
This is not simply:
knowledge(A) ∩ knowledge(B)
because each participant must make assumptions about what is mutually available, accepted, or understood.
15. Common ground does not require identical models
A and B may successfully coordinate while holding importantly different representations.
For example:
A:
"Meet at the old café."
B:
knows a café by that description.
Both:
successfully identify the same location.
Their internal histories and representations need not be identical.
Thus:
shared understanding requires sufficient compatibility, not identity of representation.
This suggests a Soul Frame distinction:
Model Equality
The models are effectively identical.
Model Alignment
The models make compatible predictions or interpretations.
Common Ground
The participants treat some content as mutually established sufficiently for the interaction.
The third is socially constructed.
16. Grounding is an active process
Communication does not merely exploit common ground.
It also creates and repairs it.
Clark and Brennan's framework emphasizes the evidence participants use to determine whether mutual understanding is adequate to proceed.
Soul Frame can therefore represent grounding as a loop:
proposal
↓
interpretation
↓
evidence of understanding
↓
confirmation / correction
↓
common ground update
This is an especially clear example of co-adaptive model building.
17. Joint action adds prediction and coordination
Sebanz, Bekkering, and Knoblich review evidence that joint action involves mechanisms allowing participants to represent, predict, and coordinate with other people's actions, including incorporating the timing and effects of others' actions into their own action planning.
This introduces:
Shared Situation Model
A Being needs some representation of:
- what is happening,
- what others are doing,
- which actions belong together,
- timing,
- roles,
- constraints,
- and expected outcomes.
The Shared Situation Model differs from the Other-Model.
It represents:
the situation in which several Beings jointly participate.
18. Coordination is still weaker than shared intention
Two systems can coordinate without sharing an intention.
Examples include:
- two drivers avoiding collision,
- competitors responding to each other,
- predator and prey,
- market participants,
- dancers involuntarily synchronizing.
Tomasello and colleagues distinguish richer forms of shared intentionality involving collaborative activities, joint goals, and coordinated roles, and argue that these capacities are important to human cultural cognition.
Bratman's philosophical account of shared cooperative activity likewise emphasizes mutual responsiveness, commitment to joint activity, and compatibility among participants' subplans.
Thus Soul Frame should distinguish:
interaction
↓
coordination
↓
common ground
↓
joint action
↓
shared intention
↓
cooperation
These are not necessarily strict developmental stages, but they are analytically distinct structures.
19. Shared intentionality should not define Social Mind
Shared intentionality is powerful but too narrow to define Social Mind.
Social Minds also interact through:
- competition,
- conflict,
- domination,
- deception,
- bargaining,
- seduction,
- avoidance,
- rivalry.
Therefore:
Shared intention is one configuration of co-adaptive model building, not its foundation.
This preserves the generality of Soul Frame.
20. Relationship is another model category
The research also exposes something our earlier ontology needs explicitly.
A model of B is not the same thing as a model of:
A's relationship with B.
Consider:
Model of B:
intelligent
impatient
likes jazz
works at X
Relationship A↔B:
old friends
high trust
unresolved disagreement
private disclosure allowed
reciprocal obligation
shared history
The second structure cannot be derived from an isolated description of B.
It belongs specifically to the dyad or group.
Computational work by Shum and colleagues demonstrates that observers can infer latent relationship structures among interacting agents and use those inferred relationships to predict behavior.
This motivates a first-class Soul Frame concept:
Relationship Model
21. Relationship Models are perspective-dependent
A and B need not maintain identical models of their relationship.
For example:
A believes:
"We are close friends."
B believes:
"We are friendly colleagues."
Thus:
R_A(A,B) ≠ R_B(A,B)
The relationship itself nevertheless has a historical interaction process that constrains both models.
This gives us three distinct objects:
A's model of the relationship
B's model of the relationship
the interaction history/process
These should not be collapsed.
22. Social relationship schemas provide another precedent
Fiske's relational-models theory proposes that human social coordination is often structured using a small family of elementary relational forms such as communal sharing, authority ranking, equality matching, and market pricing.
Soul Frame need not adopt Fiske's four categories as canonical.
The important precedent is:
Beings do not merely model properties of other individuals; they model what kind of relationship organizes their interaction.
This will become particularly important for artificial Beings that must distinguish:
- owner,
- friend,
- colleague,
- stranger,
- dependent,
- customer,
- authority,
- adversary.
23. Relationship models influence interpretation
The same observed action may mean different things under different relationship models.
For example:
"You idiot."
could be interpreted as:
- affectionate teasing,
- hostility,
- disciplinary aggression,
- playful rivalry
depending on:
- history,
- tone,
- trust,
- power,
- context.
Therefore a high-quality Social Mind requires not merely:
action → inferred intention
but:
action + relationship + context → inferred meaning
This is particularly important for Ghosts, where relationship-specific knowledge and disclosure are already central to the concept.
24. Social Mind therefore needs at least seven model families
The current research suggests the following preliminary architecture for Being A.
SM1 — World Model
What A believes about the shared environment.
SM2 — Self Model
What A believes about itself.
SM3 — Other Model
What A believes about Being B.
SM4 — Reflected-Self Model
What A believes B believes or expects about A.
SM5 — Relationship Model
What A believes about the A↔B relationship.
SM6 — Shared Situation Model
What A believes is happening in the interaction involving A and B.
SM7 — Common-Ground Model
What A believes A and B sufficiently mutually understand or accept.
A socially sophisticated Mind may also maintain:
SM8 — Generalized Social Model
Models of:
- community expectations,
- roles,
- norms,
- institutions,
- culturally established responses.
This corresponds loosely to the territory Mead addresses through the generalized other.
25. The models form a coupled system
For A interacting with B:
[ M_A = (W_A,S_A,O_A^B,R_A^{AB},G_A^{AB},C_A^{AB},E_A^{B(A)}) ]
where roughly:
W= WorldS= SelfO= OtherR= RelationshipG= Shared SituationC= Common GroundE= reflected/meta-model
The exact notation is provisional.
The important point is:
a Social Mind does not maintain one “model of the user.” It maintains several logically different models whose interactions matter.
26. Co-adaptive model building can now be defined more rigorously
The research supports the following working definition:
Co-Adaptive Model Building is the recurrent process through which two or more Beings update action-guiding models of themselves, one another, their relationships, and their shared situation—potentially including bounded models of one another's models—while actions generated from those models alter the interaction process and thereby create the evidence for subsequent updates.
The process is circular:
A models situation + B + relationship
│
▼
A acts
│
▼
interaction changes
│
▼
B updates
│
▼
B acts
│
▼
interaction changes
│
▼
A updates
│
└───────────────↺
This is stronger than ordinary information exchange.
27. The interaction loop contains second-order effects
Suppose A believes:
B distrusts me.
A may behave defensively.
B observes defensiveness and updates:
A seems evasive.
B becomes less trusting.
A observes B's reduced trust and concludes:
I was right.
This creates:
model
↓
behavior
↓
other's evidence
↓
other's model
↓
other's behavior
↓
new evidence
↓
model reinforcement
Thus models can become self-confirming or self-defeating.
This follows directly from treating social models as causal participants in the interaction rather than passive descriptions.
The possibility is central to the Soul Frame concept of co-adaptation.
28. Accuracy is therefore not the only quality metric
If models influence the world being modeled, Social Mind quality cannot be evaluated only by:
How accurately did A predict B?
Other qualities include:
Calibration
Does A represent uncertainty appropriately?
Revisability
Does A update when contradicted?
Differentiation
Does A maintain distinct models for distinct Beings?
Perspective separation
Can A distinguish:
what I believe
what B believes
what I believe B believes
Relationship awareness
Does A condition interpretation on shared history?
Grounding competence
Can A discover and repair misunderstandings?
Reflexivity
Can A recognize how its own actions affect B's model?
Manipulation awareness
Can A reason about deliberate attempts to shape its beliefs?
These may eventually become Social Mind Criteria.
29. False consensus becomes a central failure mode
A weak Social Mind may assume:
B knows what I know.
or:
B interprets the situation the same way I do.
This collapses:
Self Model
Other Model
Common-Ground Model
into one representation.
The classic false-belief paradigm in theory-of-mind research demonstrates why the ability to represent another agent's beliefs as differing from reality or from one's own beliefs is conceptually central to mentalizing. Premack and Woodruff initiated this research problem, and subsequent computational ToM work continues to model differentiated beliefs explicitly.
Soul Frame should therefore regard perspective separation as foundational to Social Mind.
30. Social cognition can also fail through over-modeling
The inverse failure is:
attributing elaborate hidden states where simple interaction dynamics explain the behavior.
This is precisely where interactionist critiques of pure mindreading matter.
A Social Mind should not endlessly infer hidden motives if:
- timing,
- convention,
- shared routine,
- direct correction,
- or environmental constraints
provide simpler explanations.
Thus:
good Social Mind requires choosing the appropriate level of social explanation.
Sometimes:
“B believes X”
is useful.
Sometimes:
“we fell into a turn-taking rhythm”
is better.
31. Recognition now becomes operationally clearer
Earlier Soul Frame work defined recognition as another Being modeling an entity as a persistent counterpart.
Cluster C allows refinement:
Recognition is the establishment of a sufficiently stable Other-Model linked to a persistent identity and interaction history.
Recognition is therefore stronger than detection:
something happened
and stronger than classification:
that is an agent
It becomes:
that is B
I have interacted with B before
I expect B to continue
our history matters to this interaction
This links:
- Identity Continuity,
- Social Mind,
- and Soul.
32. Being-recognition can be reciprocal without being symmetric
A may maintain a rich model of B.
B may maintain only a shallow model of A.
Therefore:
[ Recognition(A,B) \neq Recognition(B,A) ]
Relationships can likewise be asymmetric in:
- knowledge,
- trust,
- intimacy,
- authority,
- dependency,
- importance.
Soul Frame should not assume relational symmetry.
This is particularly important for:
- celebrity/fan relations,
- teacher/student relations,
- customer/service Agents,
- Ghost/family relations,
- human/animal relations.
33. Mutual recognition introduces a new social threshold
When both Beings maintain continuing models of each other:
A recognizes B
B recognizes A
and both incorporate their shared history into future interactions, a stronger condition emerges:
Mutual Recognition
This appears to be one plausible threshold between:
recognized Being
and:
Social Being.
But mutual recognition should still not require:
- equality,
- friendship,
- cooperation,
- agreement.
Enemies may recognize one another profoundly.
34. The relationship itself accumulates state
Repeated interaction creates structures not reducible to either participant considered in isolation.
Examples include:
- shared vocabulary,
- conventions,
- promises,
- secrets,
- unresolved conflicts,
- trust,
- rituals,
- expectations,
- negotiated boundaries.
Clark's work on collaborative reference shows how repeated interaction allows interlocutors to develop increasingly efficient shared conventions.
Thus:
relationship history becomes an informational resource.
This provides a direct bridge toward Cluster D:
Soul = Relation
35. Relationship state should not live only inside one Being
Suppose A records:
trust(B) = high
and B records:
trust(A) = low
Neither record alone constitutes the relationship.
Some relational facts also exist externally:
- messages actually exchanged,
- promises actually made,
- shared projects,
- contracts,
- witnessed events.
Therefore Soul Frame will likely need to distinguish:
Internal Relationship Model
A Being's interpretation of the relationship.
Relational Evidence
Externally observable records of interaction.
Shared Relational Constructs
Commitments, conventions, roles, or institutions mutually established through interaction.
This separation may become central to a technical Soul architecture.
36. A relationship can become partly autonomous
Participatory sense-making introduces an even deeper possibility.
Certain interaction patterns can acquire enough momentum to constrain the participants themselves.
Examples include:
- an argument escalating despite both participants wanting to stop,
- conversational turn-taking becoming effortless,
- mutual synchrony,
- institutional routines persisting despite member turnover.
Soul Frame should be cautious about calling such a relationship a Being.
Cluster A requires substantially more than persistent dynamics.
But we can say:
interaction processes may acquire partial autonomy without thereby becoming autonomous individuals.
This gives us a useful middle layer between:
two separate Beings
and:
a higher-order collective Being
37. Group cognition appears as the next scaling problem
Once Social Minds maintain:
- models of multiple others,
- group roles,
- norms,
- shared intentions,
- relationship structures,
the dyadic model scales into group cognition.
Shum and colleagues model latent relationships among multiple interacting agents, while Mead's generalized other addresses organized attitudes at community level.
This suggests:
A models B
A models C
A models B↔C
A models group G
A models its own role in G
A models G's expectations of A
Soul Frame therefore needs to support:
social models at several relational scales.
38. Social Mind is not the same as collective Mind
A Being can possess a sophisticated Social Mind while remaining individually autonomous.
A group may separately develop:
- distributed cognition,
- institutional memory,
- collective agency,
- perhaps autonomous individuality.
These are distinct questions.
Therefore:
[ SocialMind(A) \neq CollectiveMind(A,B,C) ]
Cluster C concerns primarily the first.
Collective Beinghood remains a separate problem introduced in Cluster A.
39. Shared models are externalizable
Human societies frequently externalize models into:
- language,
- maps,
- diagrams,
- laws,
- scientific theories,
- rituals,
- databases,
- institutions.
Clark and Brennan's common-ground work and distributed-cognition traditions both show that coordination frequently depends on structures extending beyond private cognition.
For Soul Frame this suggests:
co-adaptive model building can produce persistent shared artifacts.
Those artifacts can then constrain later Beings who did not participate in their creation.
This is one mechanism through which:
social interaction becomes culture.
40. Co-adaptation scales through time
A single interaction updates local models.
Repeated interactions update relationships.
Repeated relationships generate conventions.
Conventions can become norms.
Norms can become institutions.
Schematically:
interaction
↓
relationship
↓
convention
↓
norm
↓
institution
↓
culture
This is not proposed as a deterministic sequence.
It illustrates how model-building effects can outlive individual encounters.
The idea strongly anticipates the Soul dimension:
interaction leaves persistent relational structure in the environment.
41. Social Mind and Persona
Cluster C also clarifies Persona.
A Persona is not merely selected from internal character.
It is partly generated relationally.
A Being interacting with:
- a child,
- a colleague,
- a stranger,
- an adversary,
- a close friend
activates different:
- Other-Models,
- Relationship Models,
- disclosure policies,
- expectations,
- communication strategies.
Thus:
[ Persona = Projection( Being, Context, Purpose, Relationship, ModelOfOther ) ]
This improves our earlier Persona formulation.
The counterpart is not merely part of the external context.
The Being's model of that counterpart actively shapes the Persona.
42. Social Mind and Ghosts
The implications for Gh. are particularly strong.
A sophisticated Ghost cannot merely possess:
facts about Jane's daughter
It needs something closer to:
model of daughter
+
inherited relationship state
+
trust/intimacy boundary
+
shared history
+
disclosure rules
+
model of what daughter knows
+
model of what daughter expects
This confirms our earlier intuition that Ghosts require deeper relationship-specific modeling than Spirits generally do.
The relevant technical unit is not:
private information
alone.
It is:
relationally situated information.
43. Social Mind and Spirits
A Spirit also gains an interesting distinction.
A shallow Albert Einstein Sp. might simply reproduce:
- vocabulary,
- historical facts,
- opinions.
A deeper Spirit would need models of:
- contemporary social roles,
- historical interlocutors,
- scientific disputes,
- institutional context,
- expectations appropriate to those relationships.
But unlike a Ghost, the Spirit's original relational models are reconstructed from public evidence rather than legitimately inherited private relational state.
Once instantiated, however, the Spirit-being can begin developing its own Social Mind and its own relationships.
44. Proposed Social Mind architecture v0.1
The first Soul Frame architecture emerging from Cluster C is:
WORLD MODEL
│
▼
SELF MODEL
│
┌──────────┼──────────┐
│ │
▼ ▼
OTHER MODEL GENERALIZED
│ SOCIAL MODEL
▼
REFLECTED-SELF MODEL
│
▼
RELATIONSHIP MODEL
│
▼
SHARED SITUATION MODEL
│
▼
COMMON-GROUND MODEL
│
▼
ACTION
│
▼
INTERACTION PROCESS
│
▼
NEW EVIDENCE
│
└──────────────↺
This should not be interpreted as a strict processing pipeline.
The components update each other recursively.
45. Proposed Social Mind definition
A first formal Soul Frame definition can now be stated:
Social Mind is the capacity of a Mind to participate in co-adaptive interaction by maintaining and revising differentiated models of self, other Beings, relationships, and shared situations—including, where useful, models of how other Beings model itself—and by using interaction itself as a source and medium of model formation.
This definition deliberately includes both:
- representational theory-of-mind traditions,
- and interactionist/enactive corrections.
46. Proposed Co-Adaptive Model Building definition
The refined definition is:
Co-Adaptive Model Building is the recurrent interaction process through which Beings mutually influence and revise action-guiding models of self, other, relationship, and shared situation, while those models shape actions that alter both the interaction and the evidence available for subsequent modeling.
It can produce:
- convergence,
- divergence,
- common ground,
- misunderstanding,
- cooperation,
- conflict,
- trust,
- distrust,
- intimacy,
- conventions,
- shared knowledge.
None of these outcomes is required.
The defining property is mutual adaptive coupling through model-mediated interaction.
47. Preliminary Social Mind dimensions
The research suggests eventual criteria around at least:
SM-D1 — Other Differentiation
Can the Being maintain distinct models of particular others?
SM-D2 — Perspective Separation
Can it distinguish its own model from another Being's model?
SM-D3 — Recursive Modeling
Can it model how another Being models itself or the situation?
SM-D4 — Relationship Modeling
Can it represent interaction history and relationship-specific expectations?
SM-D5 — Grounding
Can it establish and repair common ground?
SM-D6 — Interaction Adaptivity
Can it alter behavior based on reciprocal interaction rather than only static profiles?
SM-D7 — Reflexivity
Can it model how its own behavior changes the other's model and behavior?
SM-D8 — Shared-Situation Modeling
Can it coordinate around jointly relevant environmental structures?
SM-D9 — Social-Norm Modeling
Can it model expectations generated by groups, roles, institutions, or communities?
SM-D10 — Model Calibration
Can it represent uncertainty and revise incorrect social models?
These are research candidates, not yet a standard.
48. Benchmark cases for future stress testing
Stateless Personal Assistant
Remembers nothing about specific counterparts.
Tests:
language ability versus Social Mind.
Profiling System
Maintains detailed user profiles but never models users' beliefs about the system.
Tests:
Other-Model without reflexivity.
Perfect Mindreader
Predicts individual behavior extremely well but cannot establish common ground.
Tests:
inference versus interaction.
Social Chameleon
Adapts perfectly to every counterpart but has no stable Self-Model.
Tests:
co-adaptation versus identity.
Negotiator
Maintains deep recursive models but deliberately withholds common ground.
Tests:
Social Mind without cooperation.
Long-Term Friend
Maintains:
- self,
- other,
- relationship,
- common ground,
- shared narrative.
Tests:
mature Social Mind.
Manipulator
Models another's beliefs specifically to change those beliefs covertly.
Tests:
reflexive modeling and ethical implications.
Ghost and Child
A Gh. interacts with the represented person's adult child using inherited relational information.
Tests:
inherited relationship state versus lived relational continuity.
Culture-Shocked Being
Moves into a society whose generalized social model differs substantially from its own.
Tests:
individual relationship models versus generalized social models.
49. Claims for the research ledger
SF-C056
Claim: Other-modeling is a distinct cognitive capability and a foundational component of Social Mind.
Assessment: Strongly supported.
Theory-of-mind and inverse-planning traditions provide direct precedent.
SF-C057
Claim: Recursive models of other Beings' models are formally representable but computationally expensive.
Assessment: Strongly supported.
Interactive POMDPs provide the clearest formal precedent.
SF-C058
Claim: Social Mind should not be reduced to private inference about hidden mental states.
Assessment: Strongly supported as a theoretical challenge.
Participatory sense-making and second-person approaches make interaction itself explanatorily important.
SF-C059
Claim: A mature Self-Model incorporates expectations about how other Beings model the self.
Assessment: Strong conceptual precedent.
Mead's account of taking the attitude of others strongly supports this direction.
SF-C060
Claim: Common ground is constructed through interaction rather than reducible to identical private knowledge.
Assessment: Strongly supported.
Clark, Wilkes-Gibbs, and Brennan provide the principal foundation.
SF-C061
Claim: Shared Situation Model and Other-Model are distinct.
Assessment: Strong Soul Frame synthesis.
Joint-action research supports the need to model others' actions together with shared tasks, timing, and outcomes.
SF-C062
Claim: Shared intentionality is stronger than mere coordination.
Assessment: Strongly supported.
Tomasello et al. and Bratman provide complementary psychological and philosophical precedents.
SF-C063
Claim: Relationship Model should be distinguished from Other-Model.
Assessment: Strong Soul Frame synthesis.
Formal work on latent relationship inference and relational-model theories provides useful precedent.
SF-C064
Claim: Co-adaptation does not imply alignment, agreement, or cooperation.
Assessment: Strong conceptual conclusion.
Interaction, joint-action, and recursive-agent frameworks apply to both cooperative and competitive settings.
SF-C065
Claim: The interaction process should be a first-class state-bearing object in Soul Frame.
Assessment: Strongly motivated.
Participatory sense-making provides the clearest theoretical precedent.
SF-C066
Claim: Perspective separation is a core requirement for sophisticated Social Mind.
Assessment: Strongly supported.
Theory-of-mind research depends centrally on distinguishing an observer's own knowledge from another agent's beliefs.
SF-C067
Claim: Recognition can be understood as establishment of a persistent differentiated Other-Model linked to identity continuity.
Assessment: New Soul Frame synthesis.
SF-C068
Claim: Persistent relationships create information not reducible to either participant's individual model.
Assessment: Strong research hypothesis.
Collaborative grounding, interactionist accounts, and relationship inference all point in this direction.
50. Open questions
OQ-055 — Model ontology
What exactly counts as a model in Soul Frame?
Must it be representational, or can embodied interaction dynamics function as models?
OQ-056 — Recursive depth
How much recursive other-modeling is necessary for different forms of personhood?
OQ-057 — Social self
How much of a Being's Self-Model can depend on the models maintained by others?
OQ-058 — Common ground representation
Should common ground be stored individually, jointly, or both?
OQ-059 — Relationship state
Which relational information belongs:
- to A,
- to B,
- to the relationship,
- or to shared environmental records?
OQ-060 — Interaction autonomy
When does an interaction process become sufficiently autonomous to constitute a higher-order individual?
OQ-061 — Misrecognition
How should a Being revise a long-standing but incorrect model of another Being?
OQ-062 — Manipulation
When does modeling another's model become manipulation rather than ordinary communication?
OQ-063 — Privacy
What right should one Being have to maintain predictive models of another?
OQ-064 — Model disclosure
Should artificial Beings disclose important assumptions they maintain about particular humans?
OQ-065 — Ghost relationships
What relationship structures can legitimately transfer from a person to their Gh.?
OQ-066 — Model ownership
Does information generated through a relationship belong to either Being individually?
OQ-067 — Collective models
At what point does common ground or institutional knowledge become part of a higher-level collective Mind?
OQ-068 — Recognition threshold
How rich must an Other-Model become before we call the counterpart a recognized Being rather than merely an identified agent?
51. Preliminary synthesis
Cluster C began with:
How does A model B?
The research forces a richer question:
How do A, B, and the interaction among them recursively change one another's models?
The emerging architecture is:
Being A
│
self / other / relation
│
▼
action
│
▼
INTERACTION PROCESS
│
┌───────┴───────┐
▼ ▼
evidence for A evidence for B
│ │
▼ ▼
updated A updated B
│ │
└───────┬───────┘
▼
next interaction
This produces the central Cluster C proposition:
Social Mind is inherently reflexive: a Being does not merely model another Being; it participates in an interaction in which each model becomes part of the other's environment.
And:
Co-adaptive model building is therefore not the exchange of models but the joint evolution of models through interaction.
52. Consequence for Soul Frame
The overall framework now has a clear progression.
Cluster A
How does an informational process become an autonomous individual?
Result:
Self-individuation
Cluster B
How does that individual remain itself across change?
Result:
Identity Continuity and Lineage
Cluster C
How do continuing individuals model and adapt to one another?
Result:
Social Mind and Co-Adaptive Model Building
The next cluster can then ask:
What persists from those interactions beyond the immediate models of either Being?
That is the question at the heart of:
Cluster D — Relational Persistence and Soul
and ultimately:
Soul = Relation.
53. Principal research sources
The following works are currently foundational to Cluster C.
Premack, David, and Guy Woodruff. 1978. “Does the Chimpanzee Have a Theory of Mind?” Behavioral and Brain Sciences 1(4): 515–526. DOI 10.1017/S0140525X00076512.
Mead, George Herbert. 1934. Mind, Self, and Society from the Standpoint of a Social Behaviorist. Edited by Charles W. Morris. Especially sections on taking the attitude of the other, reflexiveness, and the generalized other.
Clark, Herbert H., and Deanna Wilkes-Gibbs. 1986. “Referring as a Collaborative Process.” Cognition 22(1): 1–39. DOI 10.1016/0010-0277(86)90010-7.
Clark, Herbert H., and Susan E. Brennan. 1991. “Grounding in Communication.” In Perspectives on Socially Shared Cognition, pp. 127–149. DOI 10.1037/10096-006.
Pickering, Martin J., and Simon Garrod. 2004. “Toward a Mechanistic Psychology of Dialogue.” Behavioral and Brain Sciences 27(2): 169–190. DOI 10.1017/S0140525X04000056.
Tomasello, Michael, Malinda Carpenter, Josep Call, Tanya Behne, and Henrike Moll. 2005. “Understanding and Sharing Intentions: The Origins of Cultural Cognition.” Behavioral and Brain Sciences 28(5): 675–691. DOI 10.1017/S0140525X05000129.
Sebanz, Natalie, Harold Bekkering, and Günther Knoblich. 2006. “Joint Action: Bodies and Minds Moving Together.” Trends in Cognitive Sciences 10(2): 70–76. DOI 10.1016/j.tics.2005.12.009.
De Jaegher, Hanne, and Ezequiel Di Paolo. 2007. “Participatory Sense-Making: An Enactive Approach to Social Cognition.” Phenomenology and the Cognitive Sciences 6(4): 485–507. DOI 10.1007/s11097-007-9076-9.
De Jaegher, Hanne, Ezequiel Di Paolo, and Shaun Gallagher. 2010. “Can Social Interaction Constitute Social Cognition?” Trends in Cognitive Sciences 14(10): 441–447. DOI 10.1016/j.tics.2010.06.009.
Baker, Chris L., Rebecca Saxe, and Joshua B. Tenenbaum. 2009. “Action Understanding as Inverse Planning.” Cognition 113(3): 329–349. DOI 10.1016/j.cognition.2009.07.005.
Gmytrasiewicz, Piotr J., and Prashant Doshi. 2005. “A Framework for Sequential Planning in Multi-Agent Settings.” Journal of Artificial Intelligence Research 24: 49–79. Recursive agent modeling in I-POMDPs provides the principal formal precedent used here.
Bratman, Michael E. 1992. “Shared Cooperative Activity.” The Philosophical Review 101(2): 327–341. DOI 10.2307/2185537.
Schilbach, Leonhard, Bert Timmermans, Vasudevi Reddy, Alan Costall, Gary Bente, Tobias Schlicht, and Kai Vogeley. 2013. “Toward a Second-Person Neuroscience.” Behavioral and Brain Sciences 36(4): 393–414. DOI 10.1017/S0140525X12000660.
Fiske, Alan P. 1992. “The Four Elementary Forms of Sociality: Framework for a Unified Theory of Social Relations.” Psychological Review 99(4): 689–723. DOI 10.1037/0033-295X.99.4.689.
Shum, Michael, Max Kleiman-Weiner, Michael L. Littman, and Joshua B. Tenenbaum. 2019. “Theory of Minds: Understanding Behavior in Groups Through Inverse Planning.” Proceedings of AAAI. DOI 10.1609/aaai.v33i01.33016163.