Selective Knowledge Transfer in Neurosymbolic AI: COG Packages vs Corpus Learning

This paper presents the first controlled A/B comparison of corpus learning versus pre-packaged knowledge modules (COG packages) in neurosymbolic AI. Using the NuSy Being framework, the authors compare creating a software developer being from a high school foundation via direct corpus learning versus receiving a COG (Curriculum-Oriented Growth) package containing structured Yurtle knowledge. Results show COG transfer achieves identical quality (21.6% competency question coverage) while being approximately 100,000x faster (0.6ms vs 65.7s), validating that knowledge learned by one being is fully transferable to another.