Our Team

Joseph Dung -- Primary researcher and software developer

Tuan Phong -- Student Researcher

Project details

----

This proposal presents an innovative approach to concept generation within OpenCog Hyperonby combining paraconsistent Formal Concept Analysis (FCA) with Large Language Models(LLMs) for creativity evaluation. The research builds on established concept blendingtechniques while extending them through information-theoretic criteria and neural-symbolicintegration. By implementing these methods in MeTTa, we aim to create a system thatautonomously generates novel, logically coherent concepts with practical applications inmaterials science, drug discovery, and sustainability. Our approach satisfies the RFPrequirements through robust implementation within Hyperon's Distributed Atomspace,comprehensive evaluation frameworks, and clear demonstration of practical use cases,supported by a team with extensive experience in symbolic reasoning and AI systems mesh of  concept formation and creative reasoning. The OpenCog Hyperon frameworkrepresents a significant advancement in this direction, providing a foundation for integratingsymbolic reasoning with machine learning approaches
. Within this ecosystem, the MeTTaprogramming language serves as a flexible environment for implementing cognitive processes,including concept creation algorithms.

Concept formation in intelligent systems has historically followed two primary approaches: formalmethods like Formal Concept Analysis (FCA) that provide logical rigor, and statistical methods that offer flexibility under uncertainty. OpenCog Classic previously implemented basic FCA tosupport concept creation tasks, but recent developments in paraconsistent and fuzzy logicsoffer opportunities to extend these capabilities to handle contradictions and uncertainty.

Meanwhile, Large Language Models (LLMs) have demonstrated remarkable capabilities inunderstanding and generating natural language, but they often struggle with maintaining logicalconsistency or generating truly novel concepts. The integration of LLMs with formal reasoningsystems represents a promising direction for advancing concept creation capabilities.

This proposal addresses the specific need identified in the RFP to advance "the exploration ofconcept creation methods, such as formal concept analysis and concept blending, and theirintegration into the Hyperon framework"
. By combining paraconsistent FCA, information-theoretic concept blending, and LLM-driven reinforment exploration and evaluation, we aim to create a system that autonomously generates and validates novel concepts while maintaining logical coherence through a form of introspectective game play that resolves itself from extrospective Knowledge Graph structures standing also as evaluating referees in the endless conceltual  re-combination game..              

Links and references

Fauconnier, G., & Turner, M. (2002). The Way We Think: Conceptual Blending and
the Mind's Hidden Complexities. Basic Books.
Goertzel, B. (2014). Artificial General Intelligence: Concept, State of the Art, and
Future Prospects. Journal of Artificial General Intelligence, 5(1), 1-46.
Gärdenfors, P. (2000). Conceptual Spaces: The Geometry of Thought. MIT Press.
Belnap, N. D. (1977). A Useful Four-Valued Logic. In Modern Uses of Multiple-Valued
Logic (pp. 5-37). Springer.