Using analogy to overcome brittleness in ai systems

摘要: One of the most important aspects human reasoning is our ability to robustly adapt new situations, tasks, and domains. Current AI systems exhibit brittleness when faced with situations This work explores how structure mapping models analogical processing allow for robust reuse domain knowledge. focuses on two methods existing knowledge in novel The first method, model formulation, applies analogy task formulation. Model formulation process moving from a scenario or system description formal vocabulary abstractions causal that can be used effectively problem-solving. Analogical uses prior examples determine which abstractions, assumptions, quantities, equations, are applicable within same domain. By employing examples, range an extendable by adding additional example-specific models. The robustness this method learning evaluated series experiments domains, everyday physical sketches textbook physics second transfer via analogy, task-level cross-domain learning. DTA helps overcome allowing abstract knowledge, case equation schemas control transferred learns between entities relations understood domain, through comparisons structures explanations. Then, using mapping, theory inferred extended theories themselves. across variety domains (e.g., mechanics, electricity heat flow). Successful analogies result persistent mappings, support incremental target multiple analogies.