From Self-Regulation of Research Activities to Theory Creation: Persistent Self-Models and Recursive Improvement in the Three-Generation Machine Architecture

Sustained theoretical research requires revising explanations in light of new evidence and assessing whether the available materials and research capabilities can support those revisions. This paper examines consciousness mechanisms in research systems from a functional perspective and proposes a framework connecting self-explanation, self-prediction and self-regulation through a persistent self-model. The object of study is a system comprising models, state storage, tools and research procedures. The ancestor machine studies shared rules; the mother machine studies domain theories and procedures; and child machines gather evidence and conduct tests around specific questions. Each layer inherits research identity, goals, grounds for judgment and capability limits, comparing expectations recorded before action with execution evidence to adjust research questions and construct theoretical candidates. Results separately inform revisions to object-level theory, self-understanding and research procedures. Independently evaluated new procedures may participate in forming their next versions. Using selective survival of synaptic connections as background, we construct a continuous case involving revised capability judgments, question changes and candidate reconstruction, and propose comparative designs for state interventions, inheritance across rounds and procedure modification. The framework specifies how self-representation can inform research planning and subgoal regulation; these functions and their contribution to theory creation still require implementation and testing.