ActPC-Chem: Discrete Active Predictive Coding for Goal-Guided Algorithmic Chemistry as a Potential Cognitive Kernel for Hyperon & PRIMUS-Based AGI

Abstract: We explore a novel paradigm (labeled ActPC-Chem) for biologically inspired, goal-guided AI centered on a form of Discrete Active Predictive Coding (ActPC) operating within an algorithmic chemistry of rewrite rules. ActPC-Chem is envisioned as a foundational "cognitive kernel" for advanced cognitive architectures, such as the OpenCog Hyperon system, incorporating essential elements of the PRIMUS cognitive architecture. The central thesis is that general-intelligence-capable cognitive structures and dynamics can emerge in a system where both data and models are represented as evolving patterns of metagraph rewrite rules, and where prediction errors, intrinsic and extrinsic rewards, and semantic constraints guide the continual reorganization and refinement of these rules. Using a virtual "robot bug" thought experiment, we illustrate how such a system might self-organize to handle challenging tasks involving delayed and context-dependent rewards, integrating causal rule inference (AIRIS) and probabilistic logical abstraction (PLN) to discover and exploit conceptual patterns and causal constraints. Next, we describe how continuous predictive coding neural networks, which excel at handling noisy sensory data and motor control signals, can be coherently merged with the discrete ActPC substrate. Finally, we outline how these ideas might be extended to create a transformer-like architecture that foregoes traditional backpropagation in favor of rule-based transformations guided by ActPC. This layered architecture, supplemented with AIRIS and PLN, promises structured, multi-modal, and logically consistent next-token predictions and narrative sequences.

• The paper introduces a novel ActPC-Chem framework that integrates discrete active predictive coding and algorithmic chemistry for advancing AGI.
• It utilizes dynamic rewrite rules in a self-organizing metagraph, merging subsymbolic pattern recognition with symbolic causal reasoning.
• The architecture is demonstrated through thought experiments and holds promise as a foundational cognitive kernel for Hyperon & PRIMUS-based systems.

Overview of ActPC-Chem Framework

The paper "ActPC-Chem: Discrete Active Predictive Coding for Goal-Guided Algorithmic Chemistry as a Potential Cognitive Kernel for Hyperon & PRIMUS-Based AGI" (2412.16547) presents a conceptual framework for a novel AI architecture termed ActPC-Chem. This architecture integrates discrete Active Predictive Coding (ActPC) within an algorithmic chemistry framework, utilizing rewrite rules to form a unified cognitive kernel for AI systems. The approach aims to combine subsymbolic pattern recognition with symbolic and causal reasoning, facilitating the emergence of general intelligence capabilities. Such a system is hypothesized to seamlessly integrate various PRIMUS cognitive architecture elements, potentially paving the way towards human-level AGI and superintelligence (ASI).

Architecture and Methodology

Discrete Active Predictive Coding

At its core, ActPC-Chem utilizes discrete Active Predictive Coding (ActPC) to foster a predictive-coding-based learning system. ActPC traditionally operates in continuous spaces, offering biologically plausible models through local error signals and Hebbian-like learning rules. By transitioning to a discrete framework, ActPC-Chem adapts these principles to symbolic domains, employing rewrite rules as generative models. The approach combines exploratory (epistemic) signals that maximize prediction errors with goal-oriented (instrumental) signals to minimize prediction errors, forming a balanced reward structure conducive for learning complex tasks.

Algorithmic Chemistry

The algorithmic chemistry aspect of ActPC-Chem refers to the dynamic and self-organizing nature of the metagraph of rewrite rules, reminiscent of a 'digital primordial soup.' This structure supports the continual evolution and transformation of both data and models, guided by prediction errors and reinforcement learning principles. The architecture envisions self-referential systems where rewrite rules can modify themselves and other rules, leading to the spontaneous emergence of new strategies and concepts akin to autopoietic networks.

Integration with Symbolic AI

ActPC-Chem further incorporates symbolic AI methods such as AIRIS for causal reasoning and PLN for probabilistic logical abstraction. These components enhance the system's ability to structure causal logic and probabilistic semantics within the rule-based architecture. By embedding probabilistic interpretations within symbolic logic frameworks, the architecture achieves a coherent integration of diverse cognitive modules, enabling superior cognitive synergy.

Speculative Applications and Implications

Hybrid Architecture: Virtual Bug Example

The paper illustrates ActPC-Chem through thought experiments like the 'virtual robot bug,' demonstrating its potential to perform complex reasoning tasks involving delayed and context-dependent rewards. By integrating causal rule inference with symbolic logical layers, this example underscores the robustness of the architecture in evolving adaptive cognitive processes.

Toward a Transformer-Like Model

One of the speculative propositions is adapting transformer-like architectures through ActPC-Chem. By replacing traditional backpropagation with rule-based transformations, this hybrid model aims to achieve multimodal, logically coherent predictions similar to those of transformers but grounded in a self-organizing kernel that may mitigate common issues such as hallucinations.

Future Directions

The ActPC-Chem framework poses a speculative yet promising approach to developing more adaptive, flexible AI systems. Future work may focus on formalizing discrete natural gradient updates, optimizing rule search and rewriting processes, and extending the framework to real-world robotics and virtual environments like Sophiaverse. As implementation progresses, ActPC-Chem's potential to integrate with Hyperon, PRIMUS, and other advanced cognitive architectures could significantly impact the evolution of AGI towards more general and intelligent forms.

In conclusion, while in its nascent conceptual stage, ActPC-Chem represents a bold vision for the future of AI architectures. Through its synthesis of predictive coding, algorithmic chemistry, and symbolic AI, it sets the groundwork for creating more sophisticated, autonomous, and intelligent systems capable of achieving or exceeding human-level cognitive tasks.