Developing an Understanding of Neural Networks through Threshold Logic: A Problem-Based Learning Approach

Abstract

This paper offers a theoretical and historical reconstruction of threshold logic as a foundational model for understanding neural computation. Originally developed in the 1960s and largely forgotten in contemporary AI education, threshold logic provides a structurally transparent, spatially intuitive, and cognitively resonant framework for interpreting decision functions in artificial neurons. Rather than merely proposing a pedagogical technique, we argue for the epistemological value of reintroducing this model in the age of generative AI, where black-box abstractions increasingly dominate educational practices. Grounded in logic design, control theory, and cognitive models, threshold logic is revisited here not simply as a teaching aid, but as a epistemic bridge between symbolic and sub-symbolic paradigms. We show how its geometric representations — such as planes intersecting unit cubes — allow learners to engage with neural functions as intelligible structures rather than opaque algorithms. Drawing from problem-based learning and constructionist pedagogy, we illustrate how this approach can scaffold conceptual understanding across diverse learner populations. The originality of this work lies in recovering and recontextualizing a nearly abandoned approach, positioning threshold logic as both a cognitive anchor and a historical alternative to dominant code-centered instruction. While empirical evaluation remains a task for future work, the proposed framework offers a robust theoretical foundation for rethinking neural network education within an AI-native academic landscape.