From Scientific Task Automation to the Continuous Evolution of Theory: The Ancestor–Mother–Child Machine Architecture for the Fifth Paradigm of Theoretical Research

How individual research findings enter long-term theoretical accumulation and change the starting points and methods of subsequent research is an organizational problem that must be addressed when artificial intelligence participates in theoretical research. To coordinate the updating of long-term rules, domain knowledge and specific tasks across different timescales, this paper proposes an ancestor–mother–child machine architecture. The ancestor machine provides general research methods and maintains shared rules. The mother machine translates them into domain theories and research procedures, then generates child machines around specific questions. Child machines inherit relevant knowledge and methods to propose and test candidate explanations. After the mother machine reviews their findings, approved changes enter domain theories and research procedures, respectively; issues concerning shared premises are referred to the ancestor machine for review. Revised knowledge and methods inform the next round of generation, allowing each round of research to build on existing understanding and change the basis of subsequent work. The architecture is intended to support a theoretical organism that can absorb new evidence and continuously generate and revise explanations. An illustrative reconstruction of research on memory retention under artificial hibernation explains this process of generation and feedback, providing an organizational framework for the fifth paradigm of theoretical research through human–machine collaboration.