Approximation as the Path to the Goal
We model approximation as a cycle of action, verification, and method revision under constraints on knowledge, time, and resources. An executor uses a model to choose an action, checks the result, execution trace, and consequences against the goal, and retains corrections that pass verification. We define a skill as a method that reproducibly delivers the required result with specified accuracy, within a defined domain of applicability, and at an acceptable cost. Programs can develop such methods through variation, selection, and inheritance. An Actor Graph represents the current state, goal, method, and verification rules and records each step as a transaction. Nested cycles pass goals, constraints, and budgets to inner cycles and return results, costs, and unresolved discrepancies to outer cycles. Under a uniform contraction condition, we derive an upper bound on asymptotic deviation from a changing goal as a function of goal drift and execution error. The cycle ends when the executor reaches the goal, exhausts its resources, or cannot justify the cost of continuing. The executor retains versions of the goal, method, and verification procedure, execution history, and outstanding obligations to support reproducibility and allow another executor to resume the cycle.
Authors
- Alexander Vityaz (ORCID: https://orcid.org/0009-0006-0489-7881)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-17
- DOI
- https://doi.org/10.5281/zenodo.24310399
- Primary Topic
- AI-based Problem Solving and Planning
- Type
- article
- Field-Weighted Citation Impact
- 0.00