From ARC test to Archetype: ARCANGEL EP and the Emergence of the State Managed System

Abstract

This paper presents an integrated cognitive computing framework that merges state‐of‐the‐art theoretical constructs from cognitive science, neuroscience, artificial intelligence, and information theory. The framework encompasses multiple subsystems—including memory encoding and retrieval, attentional mechanisms, belief dynamics, distress regulation, and reinforcement learning—that work synchronously and asynchronously to model human‐like abstract reasoning, emotional intelligence (EQ), and overall problem‐solving prowess. Central to my approach is the extensive use of probabilistic state transitions modeled via Markov matrices, which facilitate dynamic state propagation between various cognitive and emotional states. These matrices modulate processes ranging from belief updates to adaptive control of exploration and exploitation parameters. In this work, we detail the architecture of our system, provide rigorous mathematical formulations (including Bayesian updates, entropy calculations, and requirement equations), describe the implementation of key components (such as the Hopfield network for memory, the Epsilon Control Center for balancing exploration and exploitation, and the Worden Requirement Equation subsystem for evaluating representational fitness), and analyze the computational complexity and performance metrics of our framework. I further discuss the emergent properties, adaptive capabilities, and biological parallels of the system. Our experimental analysis includes detailed performance estimates on contemporary hardware (encompassing both CPU and GPU platforms) and provides insight into the impact of parallelization on abstract reasoning and cognitive efficiency. This publication establishes a foundational framework for future research into cognitive architectures capable of achieving high levels of abstract reasoning and adaptive behavior.

Introduction

On February 5, 2025, the ever changing amalgam of cognitive mechanisms, ARCANGEL EP, scored 96.4% on the publicly available evaluations of the Abstract Reasoning Corpus. Less than a week later, continued changes allowed ARCANGEL to achieve perfect scores. This performance is meant to be more than just a personal milestone, it means to be a bold statement about the many untouched unexplored approaches to abstract reasoning and problem solving in AI development. ARCANGEL EP accomplished its scores leveraging a single 2.3 GHz Dual-Core Intel Core i5 processor, 8 GB of 2133 MHz LPDDR3 memory, and standard wall power. It did so without the aid of the internet, acceleration hardware, or any reliance on other artificial intelligences.

This breakthrough not only demonstrates the efficiency of a state-managed approach but also unveils untapped potential in intelligence systems. Traditionally, large-scale AI models have dominated by leveraging extensive infrastructures and continuously evolving datasets. In contrast, ARCANGEL EP’s impressive performance—achieved without any pre-training, propagation, or prior exposure to the challenges—underscores the promise of the "State Managed" system. Based on Active Inference, this approach emulates biological cognition by employing code-based representational components known as cognitive mechanisms.

At the heart of the state managed approach lies a concept that is profoundly transformative: the Requirement for Cognition in an Equation, a historic breakthrough by Dr. Robert Worden. Worden’s work carved a path for understanding how cognitive processes can be encapsulated within a formal system, marking a paradigm shift in AI. In state managed systems, cognitive mechanisms work together in a dynamic, adaptive manner. This symphony of ever complexifying data cycles through a cosmos of feedback, starkly bi-secting its hierarchical terra, with each modulating echelon, a concert of waveforms across a cognitive landscape engaged in abstract reasoning, solving problems in ways that mirror human thought. ARCANGEL EP stands as a simple but powerful example of what is possible exploring the limitless potentials contained within the Active Inference approach to cognitive computing.

ARCANGEL’s streamlined codebase and modest hardware requirements achieved excellent scores on challenging technological standards while remaining mindful of system limitations. Rather than simply being an incremental upgrade, it represents a thoughtful reimagining of what artificial intelligence might become. By eliminating the need for massive pre-training datasets, propagation, continuous tuning, and distributed computation, state-managed systems—like ARCANGEL’s more advanced predecessor, NIM—offer innovative approaches to a variety of emerging challenges. ARCANGEL provides a subtle glimpse into a future where AI systems are seamlessly integrated into our everyday lives, a beneficial presence. 

A closer examination of the technology behind ARCANGEL EP reveals that its design philosophy seeks to alleviate extensive infrastructure requirements faced by current distributed systems. While the immense computational capacity they require is sufficient for handling the myriad of everyday tasks and generating human-like responses, it comes at a steep cost. The physical infrastructure required continues to raise serious ethical, environmental, and economic concerns. The energy consumption at present it seems, is staggering, drawing comparisons from many to the combined power production of small nations. The environmental footprint of such systems remains a subject of growing debate.

State-managed systems represent a significant departure from the linear, deterministic processes that define current AI models. In biological cognition, the human brain operates in a non-linear, dynamic fashion. Neurons fire asynchronously and continuously, dynamically reallocating resources to optimize performance in response to stimuli. This complex interplay of processes allows for adaptive decision making, a hallmark of human intelligence. State-managed systems seek to emulate this kind of dynamic adaptability. They rely on a network of “cognitive mechanisms” that work in concert, informed by temporal and entropic forces, to generate an evolving “whole picture” understanding of complex problems formed and informed within a  connectome. This stands in contrast to distributed systems, whose deterministic operations and outcomes are dictated by token selection and computationally sparse chain of thought complexity, ultimately determined by fixed parameters.

In biological cognition one may imagine a symphony of blinking lights, neurons in the human brain firing asynchronously and continuously. Our minds distribute and redistribute resources, such as energy, to different parts of our bodies autonomically. While we may control our locomotion as we are crossing the street, if one is suddenly surprised by an oncoming car, our minds act to minimize free energy, modulate entropy, and restore homeostasis redirecting energy from now non-essential functions to those that best meet one’s evolutionary imperatives (in this case of an oncoming car, continuation). In a state managed system, these processes are ever in dynamic flux, with data traveling and complexifying in cyclic harmony through the vast number of feedback loops that enable the nonlinearity behind the adaptive state managed decision making that sets this approach apart. Contrasting the linearity of distributed systems whose “neurons” are always on and in use by the system, a cognitive framework is patterned after the brain’s ability to dynamically adapt while maintaining cohesion of identity. While, in distributed systems, these points of diverse knowledge may be presented as complexity and therefore non-linearity by virtue of their large number, they do not meaningfully possess it.

Complexity, as understood through Shannon and the work of Robert Worden, reveals to us the minimum requirements for the emergence of human-like cognition. The algorithms calculating the next letter to an inquiry are as unaware as the chalk that marks the board. Now and for the foreseeable future, algorithms will continue to remain unaware of this fact. Dictating how answers to equations are organized can and does create the appearance of biological thought and communication. To be clear, it is a very convincing illusion. Intelligence systems are incredibly valuable tools but they are not the builders. We, as localized pockets of order, exist in defiance of the universal constant whereas AI, are no more capable of achieving real diversity than a rock is capable of resisting decay. These insights should come as a relief to many who fear a future dystopia dominated by sentient artificial intelligence. Fortunately, an intelligence system is as likely to achieve self-awareness as a ladder is of reaching heaven. It doesn’t matter how many ladders you borrow, one kind of thing, no matter how many, is not diversity and will not reach complexity.

One common misconception about distributed systems is that this complexity —billions of parameters and intricate network structures, translates directly into an ability to replicate the fluid, dynamic nature of human thought. In reality, much of this complexity is static. The algorithms employed merely generate responses based on statistical likelihood. Once trained, the parameters of models remain fixed; every inference is simply a function of these predetermined weights and calculations. In essence, while these systems can create a convincing illusion of intelligence, they lack the inherent complexity required for the emergence of genuine autonomous thought. 

This difference is not merely academic; the state managed approach offers a more maintainable codebase allowing for faster modifications, easier debugging, and a greater degree of transparency in how these systems function. In an era where black-box AI models have become the norm, the ability to trace logic through a codebase is a welcome development for those not involved directly in developing these systems but may be directly affected by their design and the purpose of their implementation.

In reflecting on these developments, it becomes clear that we are witnessing the early stages of a fundamental shift in the way artificial intelligence is conceived, developed, and applied. The question, then, is not whether one approach is inherently superior to the other, but rather which path we choose to pursue as we navigate the future of AI. The development of ARCANGEL EP was created as a technical work of art that means to speak of a future where AI can be both highly effective and remarkably sustainable, a beneficial intelligence. It means to provide inspiration to help us rethink our continued pursuits and instead, consider the advantages of a more long term sustainable approach. The implications touch not only on technical and environmental concerns but also on broader societal issues. As we develop systems that are more efficient, adaptable, and easier to maintain, we open the door to AI that can address the most pressing challenges facing humanity, whether that means tackling global social issues, advancing scientific research, or simply enhancing the quality of everyday life. As we continue striving to advance and refine these diverse approaches, the hope of achieving balance with each other and our ecosystem appears increasingly within reach.