The complete static-to-agentive construction is \[\boxed{ \begin{aligned} & (\Omega,\mathfrak D,\mathfrak S,\Sigma,\mathcal D_{\mathrm{form}},\mathcal L) \\[2mm] &\Downarrow \qquad (\mathcal D,\widehat V,\widehat E,\widehat H,\widehat P,\widehat B) \\[2mm] &\Downarrow \qquad (G,\{X_i\},\{T_e\}) \\[2mm] &\Downarrow \qquad X_i=C_i\oplus R_i \\[2mm] & (\{C_i\},\{H_e\},\mathcal A,\{\rho_i\}) \\[2mm] &\Downarrow \qquad \Gamma(H)=\varprojlim C_i \\[2mm] &\Downarrow \qquad L_H=D_H^*D_H \\[2mm] &\Downarrow \qquad \Gamma_\varepsilon(H) = \mathop{\mathrm{ran}}\,\mathbf 1_{[0,\varepsilon]}(L_H) \\[2mm] &\Downarrow \qquad W_t^{\mathrm{str}} \\[2mm] &\xrightarrow{\text{alignment}} Z_t=\mathbb P(Y_t^+\in\cdot\mid\mathbf H_t) \\[2mm] &\Downarrow \qquad \operatorname{SRW}_\Theta(W_t;\mathcal D) \\[2mm] &\Downarrow \qquad (W_t,G_t,\widehat Q_t,M_t,B_t) \\[2mm] &\xrightarrow{\text{goal and value use}} (A_t,R_t) \\[2mm] &\xrightarrow{\text{causal repair}} e_t\leq e_{\max} \\[2mm] &\Downarrow \qquad \mathbf W = [\mathfrak R]_{\text{gauge}+\text{causal equivalence}}. \end{aligned} }\] Compactly: \[\boxed{ \begin{aligned} \text{scaffold and history} &\to \text{learned organization and boundary} \\ &\to \text{local core--transfer realization} \\ &\to \text{compatible sections and global descent} \\ &\to \text{causal operation and persistent identity} \\ &\to \text{cue-controlled goal retention and counterfactual action} \\ &\to \text{monitor-mediated repair and bounded maintenance deficit} \\ &\to \text{decomposition- and specification-robust identified certification}. \end{aligned} }\] The final two arrows are optional extensions rather than consequences of the earlier ones. A relational workspace can exist without being an agent, and an agentive controller can fail to maintain the workspace that realizes it.
Machine-Native ReasoningReading edition
91. Unified formulation
Reading tools
Reading library
Text settings
Saved in this browser for reading chapters. Browser zoom also works.
Loading chapter…