ARC-NCA: Towards Developmental Solutions to the Abstraction and Reasoning Corpus

2025-05-13

AI Summary

ARC-AGI poses extreme few-shot abstract reasoning challenges, with a median of only three examples per task, demanding robust generalization beyond conventional supervised learning. Method: This work introduces developmental computation—the first application of this paradigm to ARC-AGI—via a neural cellular automaton (NCA)-based developmental modeling framework. It comprises the EngramNCA architecture, featuring implicit memory enhancement, and locally governed, self-evolving grid dynamics. Rather than relying on supervised fine-tuning, the approach emulates biological development through self-organization and progressive pattern formation. Contribution/Results: The method achieves performance on par with or exceeding that of ChatGPT-4.5 on ARC-AGI while substantially reducing computational cost. Crucially, it establishes developmental AI as a novel paradigm for few-shot abstract reasoning, empirically validating its efficacy and scalability on structured reasoning tasks.

Technology Category

• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Humans and AI: Other Foundations of Human Computation & AI
• Machine Learning: Deep Neural Architectures and Foundation Models

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

The Abstraction and Reasoning Corpus (ARC), later renamed ARC-AGI, poses a fundamental challenge in artificial general intelligence (AGI), requiring solutions that exhibit robust abstraction and reasoning capabilities across diverse tasks, while only few (with median count of three) correct examples are presented. While ARC-AGI remains very challenging for artificial intelligence systems, it is rather easy for humans. This paper introduces ARC-NCA, a developmental approach leveraging standard Neural Cellular Automata (NCA) and NCA enhanced with hidden memories (EngramNCA) to tackle the ARC-AGI benchmark. NCAs are employed for their inherent ability to simulate complex dynamics and emergent patterns, mimicking developmental processes observed in biological systems. Developmental solutions may offer a promising avenue for enhancing AI's problem-solving capabilities beyond mere training data extrapolation. ARC-NCA demonstrates how integrating developmental principles into computational models can foster adaptive reasoning and abstraction. We show that our ARC-NCA proof-of-concept results may be comparable to, and sometimes surpass, that of ChatGPT 4.5, at a fraction of the cost.

Problem

• Solving ARC-AGI's abstraction and reasoning challenges with few examples
• Enhancing AI problem-solving via developmental Neural Cellular Automata
• Achieving competitive performance against ChatGPT 4.5 at lower cost

Innovation

• Uses Neural Cellular Automata for complex dynamics
• Enhances NCA with hidden memories (EngramNCA)
• Integrates developmental principles for adaptive reasoning