Effect of thermodynamic coupling on abductive performance

Determine whether introducing thermodynamic coupling—so that a predictive system’s internal cost or uncertainty signal increases with epistemic error—improves abductive performance as causal demand and world-model complexity increase.

Background

The paper proposes that a predictive architecture may require a physical mechanism coupling epistemic error to computational cost in order to revise erroneous rules and generate genuinely new axioms. It suggests evaluating such a coupled architecture on tasks involving increasing causal difficulty and world-model complexity, but the proposed experiments are not conducted in the paper. The unresolved issue is whether thermodynamic coupling would improve abduction itself, rather than merely improve uncertainty calibration.

References

Section~8 turned this proposal into an empirical question: whether introducing such coupling would improve abductive performance under increasing causal and model complexity remains to be tested.

LLMs Don't Pay for the Jump  (2608.14397 - Balani et al., 14 Aug 2026) in Conclusion, Section 8