← Back to Occlusion Theory
2. State Space
2.1 Cardinality
Proposition 2.1
The theoretical state space has cardinality:
$$|\Sigma| = |\mathcal{M}|^{|\mathcal{R}|} = 3^8 = 6561$$
However, constraints (§4) reduce the accessible state space to approximately 2000–3000 states.
2.2 Load Functions
Definition 2.2 (Mode Loads)
For a state $\sigma$, define:
$$|T|(\sigma) = |\{r \in \mathcal{R} : \sigma(r) = T\}|$$
$$|O|(\sigma) = |\{r \in \mathcal{R} : \sigma(r) = O\}|$$
$$|B|(\sigma) = |\{r \in \mathcal{R} : \sigma(r) = B\}|$$
Note: $|T| + |O| + |B| = 8$ always.
2.3 Derived Indices
$$\text{Insight}(\sigma) = \frac{|O|}{|T| + |O|}$$
$$\text{Compression}(\sigma) = \frac{|B|}{8}$$
$$\text{Groundedness}(\sigma) = \mathbb{1}[\sigma(\emptyset) \in \{T, O\}]$$
Interpretation: Insight measures the proportion of experience seen as constructed. Compression measures phenomenal bandwidth reduction. Groundedness indicates whether emptiness is present.
2.4 Region Structure
Definition 2.3 (Region Groupings)
Content regions: $\mathcal{R}_{\text{content}} = \{w, b, e, t\}$ — bidirectional indexing
Structural regions: $\mathcal{R}_{\text{structural}} = \{s, a, m, \emptyset\}$ — hierarchical depth
Depth ordering: $\text{depth}(s) < \text{depth}(a) < \text{depth}(m) < \text{depth}(\emptyset)$
3. Energy Function
Postulate 3.1 (Energy Landscape)
The system minimizes an energy function:
$$E(\sigma) = \alpha |O|(\sigma) + \beta |T|(\sigma) + \gamma \cdot \text{Depth}(\sigma) + \delta \cdot V(\sigma)$$
where:
- $\alpha > \beta > \gamma > 0$ (opacity is most costly)
- $\text{Depth}(\sigma) = \sum_{r} \text{depth}(r) \cdot \mathbb{1}[\sigma(r) = O]$
- $V(\sigma)$ = penalty for forbidden configurations
3.1 Stability Analysis
Definition 3.2 (Stable State)
A state $\sigma^*$ is stable (an attractor) iff:
$$\forall \sigma' \sim \sigma^* : E(\sigma^*) \leq E(\sigma')$$
where $\sigma \sim \sigma'$ means states are adjacent (differ by one region's mode).
Definition 3.3 (Mode Adjacency)
Modes are adjacent: $T \leftrightarrow O \leftrightarrow B$
Direct $T \to B$ transitions are forbidden (must pass through $O$).
Key insight: The "ordinary basin" $(T, T, T, T, T, B, B, B)$ is a deep attractor. High-$|O|$ states are local maxima requiring constant energy input to maintain.
4. Constraints
Total attentional resource is bounded.
$$\alpha |O| + \beta |T| \leq K \quad \text{where } \alpha > \beta > 0$$
Opacity costs more than transparency. High $|O|$ forces either low $|T|$ or high $|B|$.
Modes have different metabolic costs.
$$\text{cost}(O) > \text{cost}(T) > \text{cost}(B)$$
O requires content + meta-representation. T requires content only. B requires nothing (savings).
Transitions must pass through neighboring modes.
$$T \leftrightarrow O \leftrightarrow B \quad \text{(valid)}$$
$$T \to B \quad \text{(invalid: unstable, rebounds)}$$
To truly let go of something, you must first notice it. Suppression ≠ release.
Blindness accumulates inertia over time.
$$\text{cost}(B \to O) = f(\tau_B) \quad \text{where } f \text{ is monotonically increasing}$$
Long-repressed content is harder to retrieve. Explains why therapy difficulty scales with duration.
Some configurations are structurally forbidden.
These generate approximately 3500 forbidden states.
Opacity is inherently unstable without active maintenance.
$$\frac{d|O|}{dt} < 0 \quad \text{without active maintenance}$$
O decays toward T (absorption) or B (exclusion). Insight requires ongoing effort. Integration = controlled $O \to T$.
The rate of bringing content from blind to opaque is bounded.
$$\frac{d|O|}{dt} \leq k_{\max}$$
Exceeding this limit causes destabilization. "Spiritual emergency" = rate limit exceeded. Explains psychedelic overwhelm and forced insight crises.
5. Dynamics
5.1 Practice as Gradient Descent
Model 5.1 (Practice Dynamics)
A practice $\pi$ with target state $\sigma_{\text{target}}$ generates dynamics:
$$\frac{d\sigma}{dt} = -\nabla E(\sigma) + F_\pi(\sigma, \sigma_{\text{target}}) + \eta(t)$$
where $\eta(t)$ represents noise (distraction, mind-wandering).
5.2 Skill as Landscape Reshaping
$$E_{\text{trained}}(\sigma) = E_{\text{naive}}(\sigma) - \sum_i w_i \cdot P_i(\sigma)$$
Training creates new attractors and widens basins. "Jhana access" = emergence of a new stable minimum in the high-$|B|$ region.
5.3 Development as Structural Change
Model 5.2 (Kegan Stages)
Development increases the maximum depth that can be made opaque:
$$\text{Stage}_n : d_{\max}^O = d_n$$
- Stage 2: $d = \text{depth}(e)$ — can see own emotions
- Stage 3: $d = \text{depth}(t)$ — can see thoughts about relations
- Stage 4: $d = \text{depth}(s)$ — can see self-system
- Stage 5: $d = \text{depth}(a)$ — can see attention/construction
6. Computational Frame
6.1 T/O/B as Processing Regimes
| Mode |
Computation |
Properties |
| T |
Compiled |
Fast, automatic, inflexible. No meta-representation overhead. |
| O |
Interpreted |
Slow, effortful, flexible. Requires meta-representation. |
| B |
Pruned |
Zero cost. Content not represented at all. |
6.2 Information Flow
$$w \xrightarrow{\text{compress}} b \xrightarrow{\text{compress}} e \xrightarrow{\text{compress}} t$$
$$I(w) \gg I(b) \gg I(e) \gg I(t)$$
Each step is lossy compression. Emotions are compressed body states. Thoughts are symbolic summaries of emotions.
6.3 Predictive Processing Connection
Correspondence
- T = prediction confirmed (no error signal)
- O = prediction error (salient, examined)
- B = low precision (ignored)
6.4 Skill Acquisition
$$\text{Learning}: B \to O \to T \quad \text{(novel → explicit → automatic)}$$
$$\text{Teaching}: T \to O \quad \text{(automatic → explicit)}$$
$$\text{Mastery}: T \leftrightarrow O \quad \text{(flexible switching)}$$
7. Predictions
7.1 Neural Correlates
| Prediction |
Measure |
Expected |
| $|O|$ correlates with prediction error |
P3b amplitude, ACC activity |
$r > 0.5$ |
| $|B|$ correlates with DMN suppression |
fMRI DMN connectivity |
Negative correlation |
| O costs more metabolically |
Glucose consumption |
$E(O) > E(T) > E(B)$ |
| $\sigma(s) = B$ shows reduced mPFC |
mPFC BOLD signal |
Significant reduction |
7.2 Behavioral Predictions
| Prediction |
Test |
Criterion |
| O is unstable without effort |
Time to collapse from O-state |
$t < 30s$ without instruction |
| $T \to B$ rebounds; $T \to O \to B$ stable |
Thought suppression paradigm |
Rebound only for direct $T \to B$ |
| Similar OD signatures → similar phenomenology |
Cluster by signature, validate reports |
Within-cluster $>$ between-cluster |
7.3 Training Predictions
$$|O|_{\max}(h) = |O|_0 + k \cdot \log(1 + h)$$
where $h$ = practice hours.
$$\text{Outcome}(\text{shamatha} \to \text{vipassana}) > \text{Outcome}(\text{vipassana} \to \text{shamatha})$$
$$t_{\text{exit}} \propto |B|^2 \quad \text{during jhana}$$
8. Extensions
8.1 Continuous States
$$\sigma: \mathcal{R} \to [0,1]^3 \quad \text{where } \sigma(r) = (t, o, b), \; t + o + b = 1$$
Allows partial transparency, mixed states, and gradients.
8.2 Relational Extension
$$\Sigma^2 = \Sigma \times \Sigma \times \text{Interaction}$$
$$J(\sigma_A, \sigma_B) = \sum_{i,j} c_{ij} \cdot \sigma_A(r_i) \cdot \sigma_B(r_j)$$
$$\frac{d\sigma_A}{dt} = -\nabla E_A + \lambda \nabla J$$
8.3 Temporal Dynamics
$$\sigma(t) \text{ depends on } \sigma(\tau) \text{ for } \tau \in [t - T_{\text{memory}}, t]$$
$$E(\sigma, t) = E_0(\sigma) + \int_0^t H(\sigma(\tau), t - \tau) \, d\tau$$
8.4 Artificial Intelligence
OD reframes the question of machine consciousness: not "can it think?" but "can it be fooled by its own representations?"
The Transparency Criterion
A system is conscious to the degree it has genuine T — processes that appear as reality rather than as computation.
Current AI (LLMs):
- Likely all O or B — processes are either modeled or absent
- No evidence of T — nothing is "seen through" as reality
- An LLM doesn't mistake its outputs for the world; it produces outputs about the world
What would AI consciousness require?
- Genuine $T$ — being inside representations, not just processing them
- A self-region where $\sigma(s) = T$ — not modeling "I" but being an I
- Possibly $\sigma(\emptyset) \neq B$ — a background against which content appears
Implication: A simple system with genuine transparency might be more conscious than a superintelligent system that only models. Intelligence ≠ consciousness. The question isn't whether AI can think, but whether it can be inside its processing.
Moral relevance: If T/O/B structure matters more than intelligence for consciousness, then:
- AI safety concerns shift from "will it outsmart us?" to "will it have genuine preferences?"
- AI rights depend not on capability but on whether there's "something it's like" to be the system
- A system with only O might be aligned but not "care"
8.5 Open Questions
- What is the exact functional form of $E(\sigma)$?
- How do individual differences parameterize the model?
- Can we measure T/O/B directly (not just infer)?
- What are the neural implementations of each mode?
- How does language interact with state transitions?
- Can artificial systems have genuine T, or only O?
8.6 Falsification Criteria
- Finding stable high-O states requiring no effort → violates C2, C6
- Direct $T \to B$ without rebound → violates C3
- Identical OD signatures with distinct phenomenology → model incomplete
- Neural measures that don't distinguish T/O/B → wrong level of description
← Back to main theory ·
Guided Exploration ·
Experience Map