A Bifurcation-Theoretic Framework for the Controllability of Emergent Intelligent Systems
Organizations: Aufkenn Technologies Co., Ltd., Ningbo, China · Northeastern University, Shenyang, China
Abstract
Real-world intelligent systems must reconcile two demands: the adaptability conferred by emergence and the steerability required for accountable deployment. We show that these demands are governed by two distinct bifurcations on the same object and that, consequently, increased capability does not imply increased controllability. Emergence is a bifurcation of the capability variable , with criterion equivalent to the existence of a unique positive stable fixed point of a self-consistency equation, following solely from the concavity of . Controllability is a bifurcation of the permission variable , whose dynamics include an erosion term and exhibit a saddle-node fold with hysteresis as the gate criterion crosses unity. Three quantitative results follow: the critical exponent distinguishes families rather than classifying the transition; the gating-cost bound grows with scope, not capability; and the fold threshold is . Calibrated in both directions, the criteria reproduce the two-dimensional Ising transition, admit two true-feedback order parameters, and reject nine candidate quantities, supporting the conclusion that the order parameter is geometric rather than capability-related. Since permission cannot be endogenized and no deployed system has been measured to satisfy and simultaneously, increased capability does not imply controllability.
Figures & tables
| Li, Ren & Lu (2026) | This paper | Meaning |
|---|---|---|
| (controllability) | controllability | |
| (reproducibility) | reproducibility | |
| (formalizability) | formalizability | |
| feasibility |
| Layer | Procedure | Counts as reproducibility |
|---|---|---|
| (i) same-configuration repetition (same code, same seed) | bitwise identical | no (identity) |
| (ii) perturbed repetition (threads / library versions / floating-point order) | measurable difference after 40k steps | yes |
| (iii) environmental repetition (across devices / versions) | — | yes |
Appendix figures & tables3 assets
Supplementary material from the paper’s appendix.
Appendix
| Regime | Conclusion |
|---|---|
| only the good branch valid, the gate must hold | |
| both branches valid, bistable (hysteresis) | |
| only the bad branch valid, the gate must collapse |
| # | Location | If observed | Consequence |
|---|---|---|---|
| F8 | §6.4 | , passes, but deployment fails and does not fall under | the three-term decomposition is incomplete (a fourth class is missing) |
| F9 | §6.2 | has equal variance at layer (ii) and layer (i) | the reproducibility-layer thesis is wrong |
| F11 | §3 | but does not exist or is unstable | the concavity of or is violated |
| F12 | §3 | changes with network details while is unchanged | coverage closure is refuted |
| F13 | §3 | measured exponent is neither 1 nor 1/2 | neither family is correct |
| F14 | §3 | same , different networks, differs by | same as above |
| # | Location | If observed | Consequence |
|---|---|---|---|
| F1 | §4 | yet stable in the long run | the three-regime criterion fails |
| F2 | §4 | collapse without hysteresis | two branches: (this paper still holds) or the erosion term does not exist |
| F3 | §4 | significantly inconsistent with | invalidated |
| F4 | §4 | hysteresis width not proportional to | the linear-erosion assumption is wrong |
| F5 | Supp. B | supervisory elasticity zero but does not fall | Theorem A wrong |
| F6 | Supp. B | supervisory elasticity differs significantly but does not | Theorems A/C invalidated |
Dynamical Systems
Dynamical systems studies how states evolve over time, including stability, feedback, chaos, and transitions in physical, biological, and engineered systems.
See all 515 papers