Schematic Processing as a Framework for Learning and Creativity in CBR and CC

Kat Agres, Geraint A. WigginsIccbr (workshops) 2015

Abstract

Traditional Case-Based Reasoning (CBR) and Computational Creativity (CC) systems often rely on matching specific, veridical instances to solve problems or generate new content. However, human cognition suggests that memory is not a literal recording but a reconstructive process organized into generalized structures called schemata. This paper proposes a theoretical framework that integrates schematic processing into AI systems, arguing that moving from specific cases to generalized mental frameworks is essential for achieving human-like learning and sophisticated combinatorial creativity.

Introduction

In the fields of Case-Based Reasoning (CBR) and Computational Creativity (CC), the ability to leverage past experiences to solve new problems or create novel artifacts is fundamental. Yet, a significant gap remains between how AI systems handle data and how the human mind processes information. While many AI systems focus on "veridical" memory—the storage and retrieval of exact instances—human memory is inherently malleable and structured by abstraction. This paper argues that for AI to evolve, it must adopt "schematic processing," a concept from cognitive psychology that allows for the generalization of experiences into flexible mental structures.

The Problem: The Limits of Instance-Matching

Traditional CBR systems typically operate by finding the closest match to a new problem within a database of specific past cases. While effective for routine tasks, this approach struggles with high-level creativity. In CC, the challenge is often "combinatorial creativity"—the ability to blend disparate concepts in a way that is both novel and contextually appropriate. Without abstracted structures, systems are limited to literal interpolations, failing to capture the underlying "gist" or style that characterizes human creative output.

The Schematic Processing Framework

To address these limitations, the authors propose a framework rooted in several key psychological principles:

1. From Episodic to Semantic Memory

Schematic processing represents the transition from episodic memory (specific, time-bound events) to semantic memory (generalized rules and knowledge). By abstracting specific cases into schemata, an AI system can move beyond what happened in a single instance to understand the underlying patterns that govern a domain.

2. Assimilation and Accommodation

Drawing on Jean Piaget’s developmental theory, the framework suggests that AI systems should interact with new data through two processes:

  • Assimilation: Fitting new information into existing schemata when the data aligns with established patterns.
  • Accommodation: Modifying the internal structure of a schema when new data is sufficiently "surprising" or contradictory, allowing the system to learn and evolve its understanding of the world.

Music as a Model for Schematic Creativity

The authors utilize music cognition as a primary domain to illustrate these concepts. They reference the work of Robert Gjerdingen on musical schemata, such as the "gap-fill" pattern. In this schema, a large melodic leap (the gap) creates a cognitive expectation for a subsequent sequence that "fills" that gap.

In a CBR context, rather than matching a new melody to one specific song, a schema-based system would match it against a "generalized case." This allows the system to generate music that adheres to the style and structural expectations of a genre, enabling more effective extrapolation and interpolation during the creative process.

Key Insights for AI Development

  • Memory as Reconstruction: AI should prioritize "gist" representations over raw data storage. This mimics the human ability to understand the semantic meaning of an event even when specific details are forgotten.
  • The "Generalized Case": By organizing cases into regions of a conceptual space (schemata), systems can achieve greater efficiency. This acts as a form of cognitive "caching," allowing for faster retrieval and more flexible application of knowledge.
  • Self-Evaluation and Novelty: A schema-based system can judge its own output. If a generated artifact conforms perfectly to a common schema, the system can identify it as "familiar" or "cliché." If it requires a slight, intentional modification of a schema, the system can recognize the output as "novel and appropriate"—the hallmark of creativity.

Conclusion and Future Directions

This paper provides a qualitative roadmap for the future of artificial creativity, establishing schemata as the essential building blocks for exploratory and combinatorial creativity. While the framework is currently theoretical, it sets the stage for future implementations in automatic music composition and conceptual blending.

Future work should focus on developing quantitative benchmarks to measure the performance of schema-based CBR systems against traditional models. Additionally, exploring how AI can achieve "bisociation"—the creative blending of two unrelated schemata—remains a vital frontier for achieving high-level artificial intelligence.

Summary updated

BibTeX
@inproceedings{agres2015schematic,
  author = {Kat Agres and Geraint A. Wiggins},
  title = {{Schematic Processing as a Framework for Learning and Creativity in CBR and CC}},
  booktitle = {Iccbr (workshops)},
  year = {2015},
}
Figure