本体论
对抗制
计算机科学
分类
公制(单位)
分类学(生物学)
人工智能
约束(计算机辅助设计)
协议(科学)
概念化
存水弯(水管)
建设性的
机器学习
语义学(计算机科学)
印为红字的
括号(现象学)
意会
数据科学
原型
参数统计
数据挖掘
理论计算机科学
标识
DOI:10.5281/zenodo.18904509
摘要
Open-ended problems such as institutional design lack enumerable solution spaces and cheap automatic evaluators. World-diversity search moves variation upstream of the answer. Target-blind Builders construct textual worlds with altered causal rules; purpose-blind Solvers address the same target under each rule set; and compilers translate the resulting mechanisms into a common target-domain schema. Building on this previously released operator, this paper introduces ontology-governed intervention search. A Taxonomist curates a persistent typed ontology of complete interventions, judges structural equivalence, and can turn rejection into targeted commissioning: its diagnosis specifies which causal relation the next proposal must change, while accepted interventions and revised categories persist as control state for later Explorers. Together, the operators provide the expansion stage of a proposed expansion-before-optimization pipeline: diversify causal premises, then make the resulting candidate population cumulative and steerable. The aim is that an open problem acquires a map of its mechanism space, not a list of answers; this is untested. Across eight workflows on one retirement-design problem, the released system produced 196 candidate records; its world stage comprised 25 world-and-solve branches. In an unblinded post-hoc census, the world's premise removed or eased a difficulty of the target problem in every branch, clearly in 17 and partly in eight, and all 25 Solvers still built a multi-component design beyond it. Seven complete rejection–redirection–mechanism-change chains appeared across 75 planned graph positions. These artifacts document automated fan-out, premise-conditioned generation, and graph-conditioned redirection, but do not establish external novelty, utility, causal performance, cross-domain generality, or large-scale efficiency.
科研通智能强力驱动
Strongly Powered by AbleSci AI