Beneficial and harmful explanatory machine learning

Abstract Given the recent successes of Deep Learning in AI there has been increased interest role and need for explanations machine learned theories. A distinct notion this context is that Michie’s definition ultra-strong learning (USML). USML demonstrated by a measurable increase human performance task following provision to symbolic theory performance. paper demonstrates beneficial effect logic classification task, yet no existing work our knowledge examined potential harmfulness machine’s involvement comprehension during learning. This investigates explanatory effects simple two person games proposes framework identifying based on Cognitive Science literature. The approach involves cognitive window consisting quantifiable bounds it supported empirical evidence collected from trials. Our quantitative qualitative results indicate aided which satisfies achieved significantly higher than self Results also demonstrate fails satisfy leads worse unaided