Hegelian Dialectics for AI Concept Formation and Evolution

Human concept learning has long been modeled as a search for distinguishing features between category members and non-members. This inherently involves comparing and contrasting instances – a notion resonant with dialectical reasoning. In Hegel’s method, a thesis (an initial idea or concept) encounters an opposing antithesis, and their conflict is resolved in a higher-level synthesis. Successive dialectical cycles yield progressively more comprehensive and coherent understanding. Inspired by this, we explore how to endow AI systems with a dialectical loop for concept formation: starting from raw existence (data instances), detecting change (conflicting evidence), forming an abstraction (concept), and then, if warranted, moving to meta-abstraction (the abstraction of abstractions).