codex
70 lines
3.4 KiB
Markdown
70 lines
3.4 KiB
Markdown
# Blueprint: From Markov Blankets to Subjects
|
|
|
|
## Thesis
|
|
|
|
A Markov blanket is neither a sufficient definition of an agent nor a proof of consciousness. It is a scale-relative statistical boundary. Volume 2 becomes philosophically defensible when its blanket formalism is embedded in a temporally extended account of autonomous boundary maintenance and when its IIT claim is treated as an additional, independently testable constraint. The resulting "intellecton" is a process that repeatedly re-establishes its own causal boundary, not a static cortical object.
|
|
|
|
## Section 1: The Boundary Problem
|
|
|
|
- Reconstruct Volume 2's four-state partition: internal \(c\), sensory \(s\), active \(a\), external \(\lambda\).
|
|
- Distinguish statistical, causal, operational, and phenomenal boundaries.
|
|
- Show why \(c \perp \lambda \mid (s,a)\) is important but ontologically underdetermining.
|
|
- Introduce the four-rung ladder: boundary, autonomy, integration, subjecthood.
|
|
|
|
## Section 2: Markov Blankets as Scale-Relative Models
|
|
|
|
- Analyze conditional independence and precision-matrix sparsity.
|
|
- Interrogate linearization, stationarity, solenoidal flow, and collider assumptions.
|
|
- Define a multiscale blanket score:
|
|
\[
|
|
\mathcal{B}_\ell = I(C_\ell;\Lambda_\ell\mid S_\ell,A_\ell).
|
|
\]
|
|
- Argue that blankets are explanatory achievements relative to coarse-graining \(\ell\), not intrinsic membranes.
|
|
|
|
## Section 3: From Boundary to Autonomy
|
|
|
|
- Integrate autopoiesis, enactivism, and active inference.
|
|
- Define autonomy as counterfactual boundary maintenance under perturbation.
|
|
- Introduce:
|
|
\[
|
|
\mathcal{A}_{T} =
|
|
\mathbb{E}\!\left[\sum_{t=0}^{T}
|
|
\log\frac{p(b_{t+1}\mid b_t,c_t,\mathrm{do}(\lambda_t))}
|
|
{p(b_{t+1}\mid b_t,\mathrm{do}(\lambda_t))}\right].
|
|
\]
|
|
- Separate mere insulation from self-maintaining organization.
|
|
|
|
## Section 4: Integration Without Equivocation
|
|
|
|
- Examine the transition from recurrent covariance to \(\Phi>0\).
|
|
- Distinguish correlation, dynamical irreducibility, causal integration, and phenomenal unity.
|
|
- Analyze the continuous-to-discrete TPM mapping and its dependence on grain and timescale.
|
|
- Propose robustness across partitions and interventions as a stronger criterion.
|
|
|
|
## Section 5: The Temporally Thick Intellecton
|
|
|
|
- Replace the instantaneous blanket with a history-dependent process.
|
|
- Connect memory, predictive processing, and diachronic identity.
|
|
- Define process identity through recurrent boundary reconstitution:
|
|
\[
|
|
\mathcal{I}_{T}=\sum_t
|
|
D_{\mathrm{KL}}\!\left[p(c_{t+1}\mid b_{\le t})
|
|
\,\|\,p(c_{t+1}\mid b_t)\right].
|
|
\]
|
|
- Argue that agency requires temporal thickness and normative continuity.
|
|
|
|
## Section 6: Empirical and Formal Research Program
|
|
|
|
- Specify falsifiable tests using perturbational neuroscience, causal discovery, and multiscale modeling.
|
|
- State failure conditions for the Volume 2 thesis.
|
|
- Develop a hierarchy of evidence for boundary, autonomy, integration, and subjecthood.
|
|
- Address biological, artificial, collective, and pathological cases.
|
|
|
|
## Section 7: A Critical Process Ontology
|
|
|
|
- Synthesize the reconstruction.
|
|
- Compare physicalism, panpsychism, enactivism, and process ontology.
|
|
- Defend a modest conclusion: blankets identify candidate loci of agency, not subjects by fiat.
|
|
- State the revised Volume 2 principle: an intellecton is a scale-relative, temporally extended process whose boundary is counterfactually maintained and whose causal integration is robust under intervention.
|
|
|