Providing Self-Aware Systems with Reflexivity
This addresses the challenge of building more autonomous and adaptive AI systems, though it appears incremental as it builds on existing theories of consciousness and program interpretation.
The paper tackles the problem of creating self-aware systems by introducing computational reflexion, a mechanism for programs to inspect their own code during computation, and provides a formal definition and proof-of-concept implementation.
We propose a new type of self-aware systems inspired by ideas from higher-order theories of consciousness. First, we discussed the crucial distinction between introspection and reflexion. Then, we focus on computational reflexion as a mechanism by which a computer program can inspect its own code at every stage of the computation. Finally, we provide a formal definition and a proof-of-concept implementation of computational reflexion, viewed as an enriched form of program interpretation and a way to dynamically "augment" a computational process.