Self-editing meta-learning in Emergent Models

Prove whether, given sufficient memory and computation time and with reward supplied as an input, some Emergent Models can implement both a prediction subroutine and a reward-driven update subroutine that modifies a designated portion of the latent program, and characterize the maximum self-editing plasticity and its robustness implications.

Background

The paper proposes state retention as a mechanism for carrying information between successive predictions. In a latent-universal model, the retained terminal state can serve as the next program, potentially allowing the system to modify its own latent computation in response to rewards. The authors explicitly label the existence of such prediction-and-update mechanisms as a conjecture and state that the proposal remains speculative.

References

In latent-universal settings, universality implies that the substrate can in principle represent any algorithm, while the latent nature of the program makes that algorithm part of the mutable state itself, suggesting the following conjecture: given sufficient memory and computation time, and providing the reward as an input, there exist some EMs that implement internally: (i) a prediction subroutine, which maps the current input to an output, and (ii) an update subroutine, which uses reward signal to modify a designated sub-portion of the program. In this view, the model could not only execute a prediction but also a feedback-driven self-improvement procedure acting on its own latent program, realizing a form of meta-learning \citep{vanschoren2018metalearningsurvey}. However, this remains speculative and requires further analysis, especially to determine the maximum degree of self-editing plasticity a substrate can support and the implications for robustness.

— Emergent Models: Intelligence from Tiny Substrates  (2608.14019 - Bocchese et al., 14 Aug 2026) in Section 2.5, “Sequential Operation, State Retention and Meta Learning”