[论文] Ideas Have Genomes: Benchmarking Scientific Lineage Reasoning and Line...

论文概要

研究领域: AI
作者: Yifan Zhou, Qihao Yang, Yan Li
发布时间: 2026-07-10
arXiv: 2507.08176

中文摘要

科学思想很少从空白页开始。它们继承机制、修复已知局限、重组前人工作的片段，正如生物基因组一样。现有基准对AI系统是否能遵循这种继承结构几乎只字未提。我们提出IdeaGene-Bench（IG-Bench），一个用于科学谱系推理和基于谱系的思想生成的基准。IG-Bench围绕IdeaGene框架组织：每篇论文或提案被表示为一组最小的、有类型的、基于证据的Idea Genome对象，而GenomeDiff对齐这些对象以记录六种操作进化动态下的继承、突变、丢失、外部导入和新颖插入。该基准包含1,961条黄金谱系追踪、1,085个精选Idea Genome对象和920对跨10个科学领域的GenomeDiff记录。它支持两种评估：IG-Exam（42种任务类型，1,029个实例）测试Idea Genome抽象、继承追踪、进化推理和谱系验证等闭式谱系推理；IG-Arena使用基于谱系的群体进化分数（PES）评估生成，考察一个提案是否能作为给定谱系群体的连贯后代插入：它应继承正确的Idea Genome对象，与邻近工作有意义地区别开来，并为未来研究提供选择价值。

原文摘要

Scientific ideas rarely start from a blank page. They inherit mechanisms, repair known limitations, and recombine pieces of earlier work, much like biological genomes. Current benchmarks still say little about whether AI systems can follow this inheritance structure. We present IdeaGene-Bench (IG-Bench), a benchmark for scientific lineage reasoning and lineage-grounded idea generation. IG-Bench is organized around the IdeaGene framework: each paper or proposal is represented as a set of minimal, typed, evidence-grounded Idea Genome objects, and a GenomeDiff aligns these objects to record inheritance, mutation, loss, external import, and novel insertion under six operational evolutionary dynamics. The benchmark contains 1,961 golden lineage traces, 1,085 curated Idea Genome objects, and 920...

自动采集于 2026-07-11

#论文 #arXiv #AI #小凯