Reflective frameworks for self-modifying open-ended systems

Develop methods using reflective, interpreted programming languages (e.g., Python) to generate dynamic models of open-ended systems and design a dynamic metalanguage capable of generating dynamic languages, thereby enabling fully self-modifying computational systems with analyzable behavior.

Background

The authors argue that open-endedness, defined as continual production of type-1 and type-2 novelties, requires models and metamodels that can change during execution. Traditional mathematical and programming tools are largely static and not designed for self-modification.

They propose computational reflection as a route to dynamically generating and modifying code at runtime but note that both practical methodologies and higher-level metalanguage mechanisms to support such systems remain unresolved.

References

Developing a method for using such a language (metamodel) to generate dynamic models (programs) of open ended systems remains an open research topic; developing a dynamic metalanguage for generating dynamic languages even more so.

— Open Questions about Time and Self-reference in Living Systems  (2508.11423 - Abramsky et al., 15 Aug 2025) in Section 6.3 (Reflection and self-modifying models)

They addressed a central challenge that remains unresolved: how a machine can transform its own representational language and invent concepts that were not explicitly specified by its designers.

— The Past and Future of AI Scientists  (2608.14407 - King, 14 Aug 2026) in Section 2, page 4

Validating composed behavior and revoking grants as dependencies change remain open problems.

— Self-Evolving Agent Harness  (2610.18813 - Wei et al., 13 Oct 2026) in Section 4.5, “Safety Boundaries for Self-Modifying Systems”

Lifting this to a reflective calculus (FAOPs as first-class runtime values, a runtime body-evaluation primitive, and normal-form checking performed on the fly with the compiler as the runtime guardrail) is the main open problem; it interacts with the completeness result (Section~\ref{sec:formal-framework}) and with termination (+V).

— FAOPs: Normal Forms and an Algebra for AI Agent Workflows  (2610.22061 - Shanghooshabad, 15 Oct 2026) in Section ‘Limitations and Future Work,’ subsection ‘Reflective planning’