Modeling arbitrarily applicable relational responding with the non-axiomatic reasoning system: a Machine Psychology...

Area of Science:

• Artificial Intelligence
• Cognitive Science
• Behavioral Psychology

Background:

• Arbitrarily Applicable Relational Responding (AARR) is crucial for human language and reasoning.
• Existing AI models struggle to replicate the flexible, context-dependent nature of AARR.
• Relational Frame Theory (RFT) provides a behavioral psychology framework for understanding AARR.

Purpose of the Study:

• To propose a novel theoretical approach for modeling AARR in AI using the Non-Axiomatic Reasoning System (NARS).
• To demonstrate how "acquired relations" can enable NARS to derive symbolic relational knowledge from sensorimotor experiences.
• To show how AARR properties can emerge from NARS's reasoning mechanisms and memory structures.

Main Methods:

• Integrating RFT principles with NARS's adaptive reasoning capabilities.
• Developing a theoretical mechanism of "acquired relations" within NARS.
• Conducting theoretical demonstrations of stimulus equivalence, transfer of function, and complex relational networks.

Main Results:

• NARS conceptually demonstrated the derivation of untrained relations, mirroring human cognitive phenomena.
• The system showed context-sensitive transformations of stimulus functions.
• Key AARR properties like mutual and combinatorial entailment emerged from NARS's architecture.

Conclusions:

• AARR, previously considered uniquely human, can be conceptually modeled by AI systems.
• Integrating behavioral science insights is valuable for advancing artificial general intelligence (AGI).
• Further empirical validation is needed to confirm the theoretical approach.