Bringing Back Rule Induction to Fluid Intelligence Research An Initial Validation of the ARC-AGI Benchmark in Humans

Jul 16, 2026

Two competing perspectives on fluid intelligence (gf) measures propose constraints by either working memory capacity or the ability to induce novel relations. The ARC-AGI benchmark largely requires rule induction and was proposed as a gf measure for humans and artificial systems, yet its psychometric properties in human samples were unexamined. In a first study with 100 participants, ARC-AGI items showed good psychometric properties and substantially correlated with figural fluid intelligence (ρ=.63), with only weak links to figural originality. Findings support initial validity of ARC-AGI for human gf.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages49

Reading Time2 h 3 min

Abstract

Definition of Fluid Intelligence

Fluid Intelligence as Novel Information Processing

Validation of ARC-AGI in Humans

Psychometric Properties and Correlations

Implications and Future Research

Bringing Back Rule Induction to Fluid Intelligence Research? An Initial Validation of the ARC-AGI Benchmark in Humans Jasmin Thelen 1, Oliver Wilhelm 1 1 Institute of Psychology and Education, Ulm University, Germany This article is a preprint. Please note that it has not (yet) undergone scientific peer-review. Author Note Jasmin Thelen [https://orcid.org/0000-0002-3468-6894](https://orcid.org/0000-0002-3468-6894) Oliver Wilhelm [https://orcid.org/0000-0001-7980-1166](https://orcid.org/0000-0001-7980-1166) We did not receive any financial support for the research, authorship, and/or publication of this article. We confirm that the work conforms to Standard 8 of the American Psychological Association's Ethical Principles of Psychologists and Code of Conduct. All data and code are available at [https://osf.io/b2ugc](https://osf.io/b2ugc). Correspondence concerning this article should be addressed to Jasmin Thelen, Institute of Psychology and Education, Ulm University, Albert-Einstein-Allee 47, 89081 Ulm, Germany, Email: [j](jasmin.thelen@uni-ulm.de)[asmin.thelen@uni-ulm.de](jasmin.thelen@uni-ulm.de) Abstract Two competing perspectives on fluid intelligence (gf) measures propose that performance is primarily constrained either by working memory capacity or by the ability to induce novel relations. The first perspective is currently dominant in measurement, as evident from the use of a limited set of recurring rules, whereas the second perspective is reflected in many definitions but rarely present in measurement. The ARC-AGI benchmark predominantly requires rule induction and was proposed as a measure of gf for both humans and artificial systems. However, its psychometric properties have not yet been examined inhuman samples. We therefore investigated the psychometric characteristics and nomological network of ARC-AGI in a first study with 100 participants. A compilation of ARC-AGI items showed good psychometric properties and correlated substantially with figural fluid intelligence as measured by a figural reasoning test (ρ = .63) . Associations with figural originality were weak. These findings provide initial support for the validity of ARC-AGI as a measure of human fluid intelligence. Future research should include more rule induction tasks as well as additional multivariate covariates. This study is unusual by studying a task in humans that was initially designed for machines. We suggest systematically embedding AI benchmarks into the nomological network of human cognitive abilities to enable more systematic evaluation and interdisciplinary cooperation. Words: 213 Keywords: fluid intelligence, AI benchmarks, induction, structural equation modeling Bringing Back Rule Induction to Fluid Intelligence Research? An Initial Validation of the ARC-AGI Benchmark in Humans Fluid intelligence (gf) is an indispensable and prototypical part of human cognitive abilities. It is the cognitive ability most strongly associated with general intelligence (Carroll, 1993) and consistently predicts academic achievement, job performance, and health outcomes (Gottfredson & Deary, 2004; Kuncel et al., 2001, 2004; Sackett et al., 2022; Watrin et al., 2022) . Many definitions emphasize novel information processing as essential for gf (W. J. Schneider & McGrew, 2018), yet established gf tests typically rely on a relatively small set of rules that are repeatedly used across items (e.g., Becker et al., 2016; Carpenter et al., 1990; Schroeders & Walter, 2026) . This raises the question whether current measures include novel information processing to a sufficient degree. Recently, gf gained considerable relevance outside of psychology: it has been incorporated into the evaluation of artificial intelligence (AI) (Hendrycks et al., 2025; Qu et al., 2024; Song et al., 2024) . Presumably the most prominent example is the Abstraction and Reasoning Corpus for Artificial General Intelligence (ARC-AGI, Chollet, 2019; Chollet, K

FAQs

What two perspectives on fluid intelligence are contrasted in the paper?

The paper contrasts gf being primarily constrained either by working memory capacity or by the ability to induce novel relations.

What is ARC-AGI and why is it relevant to measuring human fluid intelligence?

ARC-AGI is an abstraction and reasoning benchmark designed to require rule finding, proposed as a gf measure for both humans and artificial systems.

What did the initial human study find about ARC-AGI’s validity?

With 100 participants, ARC-AGI items showed good psychometric properties and substantially correlated with figural fluid intelligence (ρ=.63), while associations with figural originality were weak.

Jul 16, 2026

Upload by Evangeline