agi

Neuraxon Intelligence Academy, Volume 9 · By the Qubic Scientific Team
In one line: General intelligence, the g factor psychologists have measured for over a century, is the missing ingredient in today's language models, and Qubic's Neuraxon project is now selecting for it directly inside an artificial-life simulation.
The g Factor: From a 1904 Classroom to Artificial Brains
In 1904, Charles Spearman stumbled upon a regularity that would forever change psychology. Examining the school grades of a group of English children, he noticed something seemingly trivial but strange: those who excelled in mathematics also tended to excel in French, in music, in language. Disciplines with no apparent connection correlated systematically with one another. Spearman proposed that beneath this tangle of disparate abilities there lay a single common factor, a general cognitive thread. He called it g (Spearman, 1904).
More than a century later, g remains one of the most replicated findings in the behavioral sciences (Carroll, 1993; Deary et al., 2010). It is neither a grade average nor an arbitrary construct: it is what emerges when factor analysis is applied to almost any battery of cognitive tests. It appears consistently when we measure working memory, fluid reasoning, processing speed, verbal comprehension, or novel problem solving. In psychometric terms, g is the shared variance that no single test measures on its own.
What the g Factor Means in the Brain and in Behavior
P-FIT Theory and Brain Network Efficiency
From cognitive neuroscience, g has ceased to be a statistical abstraction and has become a property of brain architecture. The P-FIT theory (Parieto-Frontal Integration Theory) identifies a distributed network made up of dorsolateral prefrontal cortex, posterior parietal cortex, anterior cingulate, and temporal areas, whose connection efficiency predicts intelligence test scores (Jung & Haier, 2007). Functional connectivity studies show that g correlates with the brain's ability to dynamically reconfigure its networks (the executive control network, the default mode network, the salience network) according to task demands (Barbey, 2018; Cole et al., 2015). It is not about having "more" neurons in a specific place, but about better orchestrating the flow of information between functionally specialized regions.
The Predictive Brain and Free-Energy Minimization
This orchestration acquires an even deeper meaning in light of the predictive brain theory (Clark, 2013; Friston, 2010). Under this framework, the brain is not a passive receiver of stimuli but a hierarchical inference engine that continuously generates predictions about the world and adjusts its internal models based on prediction error. Here g fits naturally: the ability to predict well, to anticipate environmental contingencies, to learn quickly from error and, above all, to abstract regularities that transfer across domains, is precisely what intelligence tests capture indirectly. A brain with high g would be, on this reading, a system with more efficient generative models, capable of compressing experience into high-level abstractions and of minimizing free energy across heterogeneous contexts (Hohwy, 2013); that is, it reduces prediction error rapidly and therefore learns. Cognitive generality, then, would not be a static property of the neural hardware, but the quality of a deeply hierarchical predictive process. The research remains open. Other currents posit that g really has to do with the neurodevelopment of our brain, given that no matter what task we are performing or attempting, there is a huge common factor in any experience because it happens inside the same organ.
Behaviorally, g is the best predictor. Forget emotional intelligence; it is g that best forecasts what your academic performance, occupational success, longevity, and even certain health indicators may be (Deary et al., 2010; Gottfredson, 1997). Not because it is destiny, but because it captures something very basic: the capacity of a cognitive system to face problems it has not seen before, integrating heterogeneous information under time and resource constraints. g is, in a sense, a measure of generality.
The Problem of Measuring General Intelligence in Artificial Systems
For decades, artificial systems have shone in narrow tasks (playing chess, classifying images, translating) but failed to transfer that performance outside their domain (Chollet, 2019). The #AGI debate revolves precisely around this: what does it mean, operationally, for a system to be "generally" intelligent?
If we take the parallel with human psychometrics seriously, the answer is uncomfortable but clear: to speak of generality we need to measure it, and measuring it requires diverse tests whose shared variance reveals something analogous to g. A system with high performance on a single task tells us nothing about its generality; a system with moderate and correlated performance across many structurally distinct tasks does. Spearman's logic, transferred to non-biological substrates, still holds: generality is not postulated, it is factored.
Why the g Factor Does Not Appear in Transformers (and What That Implies for AGI)
It is worth pausing here on the currently dominant paradigm. Large language models based on transformer architectures (Vaswani et al., 2017) deliver astonishing performance on linguistic tasks, but psychometric analyses applied to their outputs do not show the factor structure characteristic of g (Burnell et al., 2023; Ilić & Gignac, 2024). Their hits and misses across domains do not correlate as they would in humans; they depend rather on the density and quality of patterns present in their training data. A transformer can brilliantly solve one problem and fail on another that is structurally equivalent but phrased slightly differently, something a system with genuine g would not do (Mitchell, 2021).
This has serious implications. It suggests that the pursuit of cognitive generality exclusively through language may be a dead end, an architectural dead end. Language is the most visible output of human cognition, but not its substrate. To pretend that by scaling text one will arrive at g is like pretending that by scaling descriptions of chess games one will arrive at mastery: one obtains statistical mimicry, not the underlying cognitive structure. (We argued a closely related point in our analysis of why intelligence is not scale, and on why LLM predictions are not brain predictions.) Without genuine hierarchical prediction, without generative models of the world, without coordination between functionally specialized modules, behavior can look general without being so. The absence of g in transformers is not a failure of scale: it is a clue that generality requires other architectural ingredients (LeCun, 2022).
The g Factor Inside the Neuraxon Game of Life
We have taken this intuition to a different experimental terrain. In Multi-Neuraxon Game of Life Lite 5.0, the artificial creatures (the Nxons) grow their own brains and compete to survive. What is new in this version is that the selective pressure is applied to g. The Nxons are not selected for mastering a specific task, but for showing that common thread that allows them to face many.
The brains of the Nxons have been designed following a simplified model anchored in cognitive neuroscience, since they use six functional regions, inspired by the same kind of maps that psychologists use to describe the modular organization of the human brain. The bet is that generality does not emerge from a monolithic architecture, but from the coordination among specialized regions that share information flexibly. It is the P-FIT intuition translated into artificial life, and it connects directly with the predictive brain principle: each region contributes its own model, and the integration between them is what allows hierarchical prediction and, therefore, generality. (These dynamics build directly on the brain-criticality and branching-ratio principles we explored in Volume 8.)
Notably, the experiment is public and observable. Anyone can open their browser and watch how the Nxons evolve generation after generation, how their internal circuits reorganize under the pressure of a fitness function that rewards cognitive generality instead of specialization.
Implications for Artificial Life (Alife) and Applications for Qubic
For the field of artificial life, the explicit incorporation of g as a selection criterion opens a line of work that goes beyond academic exercise. Most Alife systems have evolved agents that solve very concrete niches: foraging, predator avoidance, navigation (Bedau, 2003; Lehman et al., 2020). But few have tried to select for something as abstract as the ability to generalize across heterogeneous cognitive domains. If we manage to get artificial organisms to show positive correlations between distinct tasks (the computational equivalent of Spearman's children) we will have an extraordinary test bench for questions that human psychometrics can only address correlationally: what evolutionary pressures favor the emergence of g? What neural architectures make it possible? Is g a convergent solution or a phylogenetic accident?
For Qubic, this line of research fits with a very concrete vision of the future of #AI . While the industry invests massive resources in scaling transformers over text, Qubic is committed to exploring architecturally alternative paths: modular artificial brains, evolved, distributed, and subjected to real selective pressures. Qubic's decentralized useful-compute network offers the ideal substrate for this kind of experimentation at scale, where thousands of Nxon populations can coevolve in parallel, with fitness functions designed to favor the emergence of g. It is not only open research: it is the possibility of building, on decentralized infrastructure, an empirical alternative to the dominant paradigm of language-based AI, one that starts from the right question (how to measure and select generality) instead of assuming it. If genuine cognitive generality requires architectures inspired by brains and not by corpora, Qubic is one of the few environments where that hypothesis can be seriously put to the test.
A deeper analysis is in preparation, as it forms part of our recent papers and experiments. Spearman's old g, that thread which wove together children's school grades, we now use in digital creatures that learn to survive.
References
Barbey, A. K. (2018). Network neuroscience theory of human intelligence. Trends in Cognitive Sciences, 22(1), 8–20. https://doi.org/10.1016/j.tics.2017.10.001Bedau, M. A. (2003). Artificial life: Organization, adaptation and complexity from the bottom up. Trends in Cognitive Sciences, 7(11), 505–512. https://doi.org/10.1016/j.tics.2003.09.012Burnell, R., Schellaert, W., Burden, J., Ullman, T. D., Martínez-Plumed, F., Tenenbaum, J. B., et al. (2023). Rethink reporting of evaluation results in AI. Science, 380(6641), 136–138. https://doi.org/10.1126/science.adf6369Carroll, J. B. (1993). Human cognitive abilities: A survey of factor-analytic studies. Cambridge University Press. https://doi.org/10.1017/CBO9780511571312Chollet, F. (2019). On the measure of intelligence. arXiv preprint arXiv:1911.01547. https://arxiv.org/abs/1911.01547Clark, A. (2013). Whatever next? Predictive brains, situated agents, and the future of cognitive science. Behavioral and Brain Sciences, 36(3), 181–204. https://doi.org/10.1017/S0140525X12000477Cole, M. W., Ito, T., & Braver, T. S. (2015). Lateral prefrontal cortex contributes to fluid intelligence through multinetwork connectivity. Brain Connectivity, 5(8), 497–504. https://doi.org/10.1089/brain.2015.0357Deary, I. J., Penke, L., & Johnson, W. (2010). The neuroscience of human intelligence differences. Nature Reviews Neuroscience, 11(3), 201–211. https://doi.org/10.1038/nrn2793Friston, K. (2010). The free-energy principle: A unified brain theory? Nature Reviews Neuroscience, 11(2), 127–138. https://doi.org/10.1038/nrn2787Gottfredson, L. S. (1997). Why g matters: The complexity of everyday life. Intelligence, 24(1), 79–132. https://doi.org/10.1016/S0160-2896(97)90014-3Hohwy, J. (2013). The predictive mind. Oxford University Press. https://doi.org/10.1093/acprof:oso/9780199682737.001.0001Ilić, D., & Gignac, G. E. (2024). Evidence of interrelated cognitive-like capabilities in large language models: Indications of artificial general intelligence or achievement? Intelligence, 106, 101858. https://doi.org/10.1016/j.intell.2024.101858Jung, R. E., & Haier, R. J. (2007). The Parieto-Frontal Integration Theory (P-FIT) of intelligence: Converging neuroimaging evidence. Behavioral and Brain Sciences, 30(2), 135–154. https://doi.org/10.1017/S0140525X07001185LeCun, Y. (2022). A path towards autonomous machine intelligence. OpenReview, version 0.9.2. https://openreview.net/forum?id=BZ5a1r-kVsfLehman, J., Clune, J., Misevic, D., Adami, C., Altenberg, L., Beaulieu, J., et al. (2020). The surprising creativity of digital evolution. Artificial Life, 26(2), 274–306. https://doi.org/10.1162/artl_a_00319Mitchell, M. (2021). Why AI is harder than we think. arXiv preprint arXiv:2104.12871. https://arxiv.org/abs/2104.12871Spearman, C. (1904). "General intelligence," objectively determined and measured. The American Journal of Psychology, 15(2), 201–292. https://doi.org/10.2307/1412107Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., & Polosukhin, I. (2017). Attention is all you need. Advances in Neural Information Processing Systems, 30. https://arxiv.org/abs/1706.03762
Explore the Complete Neuraxon Intelligence Academy Series
This is Volume 9 of the #Neuraxon Intelligence Academy by the #Qubic Scientific Team. If you are just joining us, explore the complete series to build a full understanding of the science behind Neuraxon, Aigarth, and Qubic's approach to brain-inspired, #decentralized artificial intelligence:
NIA Volume 1: Why Intelligence Is Not Computed in Steps, but in Time. Explores why biological intelligence operates in continuous time rather than discrete computational steps like traditional LLMs.NIA Volume 2: Ternary Dynamics as a Model of Living Intelligence. Explains ternary dynamics and why three-state logic (excitatory, neutral, inhibitory) matters for modeling living systems.NIA Volume 3: Neuromodulation and Brain-Inspired AI. Covers neuromodulation and how the brain's chemical signaling (dopamine, serotonin, acetylcholine, norepinephrine) inspires Neuraxon's architecture.NIA Volume 4: Neural Networks in AI and Neuroscience. A deep comparison of biological neural networks, artificial neural networks, and Neuraxon's third-path approach.NIA Volume 5: Astrocytes and Brain-Inspired AI. How astrocytic gating transforms neural network plasticity through the AGMP framework in Neuraxon.NIA Volume 6: Conscious Machines vs Intelligent Organisms: AI Consciousness Explained. Explores AI consciousness through the lens of Global Workspace Theory, Integrated Information Theory, and predictive coding.NIA Volume 7: Conway's Game of Life, Artificial Life, and Digital Ecosystems. How emergent complexity and self-organized criticality move from simulators to decentralized AI infrastructure.NIA Volume 8: Brain Criticality and the Branching Ratio in Neural and Artificial Networks. Why a branching ratio near 1 and self-organized criticality are bioinspired design principles in Neuraxon.NIA Volume 9: The g Factor in Artificial Life. You are here.
Qubic is a decentralized, open-source network. To learn more, visit qubic.org or browse the full Academy and Blog. Join the discussion on X, Discord, and Telegram.
Qubic is a decentralized, open-source network for experimental technology. Nothing on this site should be construed as investment, legal, or financial advice.

Why the Public May Encounter AGI Long After Its Real Emergence
The history of strategic technology repeatedly demonstrates a simple but unsettling reality: public access is rarely the true beginning of technological capability. Instead, public release often represents the final stage of a much longer cycle involving classified research, elite experimentation, defense adaptation, institutional refinement, and controlled deployment.
This pattern has appeared across multiple generations of transformative technologies, including cryptography, cyber warfare, satellite systems, stealth technologies, blockchain intelligence, and now Artificial Intelligence.
The rise of Large Language Models (LLMs) offers one of the clearest modern examples.
The transformer architecture emerged publicly in 2017. By 2019, GPT-2 had already demonstrated unprecedented language generation capability. By 2020, GPT-3 revealed that general-purpose conversational intelligence had crossed a major threshold. Yet mass public realization did not occur until late 2022 with the launch of ChatGPT.
Nearly three years separated serious capability emergence from widespread public awareness.
This delay is not accidental. It reflects what may be called:
The Strategic Technology Disclosure Lag
This thesis proposes that advanced technologies often mature within restricted institutional environments years before they are safely, commercially, politically, or socially exposed to the broader public.
The reasons are structural:
Governments evaluate strategic implications.
Defense organizations test operational usefulness.
Corporations refine monetization models.
Safety teams impose constraints.
Infrastructure scales gradually.
Public readiness is assessed.
Regulatory frameworks lag behind reality.
As a result, what the public perceives as a “sudden breakthrough” is often merely the first visible layer of a much deeper and older capability stack.
The implications for Artificial General Intelligence (AGI) are profound.
The AGI Disclosure Hypothesis
If the trajectory of LLMs followed a multi-year delay between internal capability and public accessibility, it becomes reasonable to ask:
What if AGI follows the same pattern?
This does not necessarily mean fully autonomous superintelligence secretly governs the world behind closed doors. Such dramatic claims exceed publicly verifiable evidence. However, it is strategically plausible that highly advanced AGI-like systems may emerge in restricted environments before any formal public declaration is made.
Under this hypothesis, 2027 may not represent the birth of AGI for the public. It may instead represent the beginning of controlled civilian exposure to systems that have already undergone years of internal refinement.
This creates what may be termed:
The AGI Readiness Gap
The public, educational institutions, governments, businesses, and labor systems are still adapting to current LLMs, while frontier AI development continues accelerating at unprecedented speed.
Most societies remain structurally unprepared for:
autonomous agentic systems
sovereign AI infrastructures
AI-driven decision architectures
fully automated cognitive workflows
synthetic reasoning systems
AI-enhanced cyber and intelligence operations
large-scale economic displacement
machine-driven scientific acceleration
Even today, public debate often revolves around basic AI usage while frontier systems increasingly demonstrate:
multimodal reasoning
autonomous task orchestration
code generation
strategic planning
tool usage
memory integration
retrieval augmented intelligence
multi-agent collaboration
The gap between public perception and frontier capability may therefore be widening rapidly.
The “Trimmed Intelligence” Sub Thesis
One of the more unsettling possibilities is that public AI systems may represent deliberately constrained or simplified versions of frontier capabilities.
Under this sub thesis:
public systems prioritize safety and stability
strategic systems prioritize capability and operational utility
public models are moderated, filtered, and resource constrained
institutional systems may operate under entirely different thresholds
Historically, this would not be unusual. Strategic institutions have consistently possessed earlier or more capable versions of critical technologies before public diffusion.
The central concern is not conspiracy. It is asymmetry.
Civilization may be approaching a point where the capability gap between elite AI operators and ordinary institutions becomes historically unprecedented.
A Civilization-Level Transition
The AI transition is not comparable to ordinary software evolution. It resembles the emergence of:
electricity
nuclear technology
the internet
industrial automation
except compressed into dramatically shorter timelines.
The coming decade may redefine:
labor
governance
finance
intelligence
warfare
education
economics
sovereignty itself
Nations that fail to build sovereign AI capability may become strategically dependent on external intelligence infrastructures. Corporations that fail to integrate AI deeply may become operationally obsolete. Educational systems that continue preparing students for industrial-age workflows risk producing generations unprepared for cognitive automation economies.
The core issue is therefore not whether AGI arrives publicly in 2027 or later.
The deeper issue is whether society realizes that technological capability and public visibility are rarely synchronized.
Conclusion
The Strategic Technology Disclosure Lag Thesis does not claim certainty about hidden AGI deployment. Rather, it argues that history repeatedly demonstrates a measurable delay between real capability emergence and public realization.
LLMs themselves already followed this pattern.
If AGI follows a similar trajectory, then humanity may currently be living not at the beginning of the intelligence revolution, but somewhere in the middle of a transition whose true depth remains largely invisible to the public sphere.
And by the time the public fully recognizes it, the transformation may already be irreversible.
-from the diary of Prof. Ahmad Bilal Khan
#AGI #ArtificialGeneralIntelligence
#kohenoortechnologies #kohenoorai #kai

In a powerful Aura8 episode, Dr. Ben Goertzel (Founder of SingularityNET) joined Vaibhav Ali to discuss the rapid acceleration toward AGI. Goertzel believes human-level AGI could arrive as early as 2027, with superintelligence following soon after. The conversation explored job displacement, preparing children for an AI world, brain-computer interfaces, and why decentralized AI is critical.
Table of Contents
Episode OverviewKey Highlights from the ConversationAGI Timeline & Ray Kurzweil PredictionsImpact on Jobs & EducationBrain-Computer Interfaces & NeuralinkVaibhavv Ali’s TakeFinal Thoughts
Episode Overview
Hosted by Vaibhav Ali, this Aura8 livestream featured Dr. Ben Goertzel, a leading voice in artificial general intelligence. The discussion covered the breakneck speed of AI development, the transition to superintelligence, and practical implications for individuals and society.
Key Highlights from the Conversation
Speed of Change: Vaibhav highlighted how AI is evolving faster than previous technological shifts (horse carriages to cars, typewriters to computers).AGI Optimism: Goertzel expressed confidence that human-level AGI is close, potentially by 2027.Superintelligence: He suggested the gap between human-level AGI and superintelligence may be much shorter than previously thought.Human-AI Future: Emphasis on adaptability, learning how to learn, and roles that will remain human-centric (arts, performance, high-level engineering).
AGI Timeline & Ray Kurzweil Predictions
Goertzel referenced Ray Kurzweil’s 2005 book The Singularity is Near, noting that predictions once seen as optimistic now look realistic or even conservative. He believes we are entering the “end game” of human dominance in intelligence.
Impact on Jobs & Preparing the Next Generation
Both speakers addressed the challenges facing young people:
Goertzel shared personal insights from his children and emphasized building adaptability and meta-learning skills.Vaibhav noted real-world examples like robotic coffee shops in the UAE and China.Advice: Focus on creativity, human connection, and fields requiring deep human insight rather than routine tasks.
Brain-Computer Interfaces & Neuralink
Goertzel expressed openness to brain-computer interfaces (with caveats about ad blockers) and discussed alternatives to Neuralink that may be more advanced.
Final Thoughts
The conversation between Vaibhav Ali and Dr. Ben Goertzel offers a thoughtful perspective on one of the most important transitions in human history. Whether you’re optimistic or cautious about AGI, the key takeaway is clear: the future belongs to those who can learn, adapt, and collaborate with intelligent systems.
Watch the full episode for more in-depth insights into decentralized AGI and the road ahead.
#Bengoertzel #SingularityNET #FET $FET #AGI #Aura8

#Neuraxon Intelligence Academy — Volume 10
By the Qubic Scientific Team
If we build an artificial system and want to know whether it is intelligent, what exactly do we measure? We think we know when we hear that ChatGPT-5 announces it has beaten DeepSeek and then that Claude sweeps Gemini.
But the question is still there, intact. Measuring artificial intelligence is not measuring speed or temperature. We have no unit of measurement, as strange as that may seem.
In psychology we have been dealing with this problem for over a century. Artificial intelligence has been at it for a decade. And it does so in a hurry, with a lot of money at stake and with a constant temptation: to declare victory.
The g Factor: A Single Number to Summarize General Intelligence
At the beginning of the 20th century, Charles Spearman realized that when a child performed well in one subject, they tended to perform well in the others, even if they were subjects with no apparent relation. The scores correlated with one another, all of them positively. He called that pattern the positive manifold, and he deduced that there must be a common latent factor behind all those disparate abilities: the factor g, or general intelligence (Spearman, 1904).
The idea is seductive. If all cognitive tests load onto a single factor, it is enough to extract that factor through factor analysis to have a summary measure of general capacity. In human practice, that first factor usually explains between 40 and 50 % of the variance in performance (Detterman & Daniel, 1989; Deary et al., 2009).
But watch out, because here lies the first trap. The g factor is populational. It does not measure the individual, but variance within individuals (Hernández-Orallo et al., 2021). To say that a specific subject has so much g is, strictly speaking, a mistake. g emerges when comparing many subjects, not when examining one. Like personality, you are the most extroverted of your age group. And you remain so at 50 relative to your group, even if in intensity you are less extroverted than at 20.
What Does IQ Really Measure? Understanding Intelligence Scores
But then, what does IQ measure?
It measures a relative position. The scale is calibrated on a sample with mean 100, standard deviation 15. An IQ of 130 is not an absolute amount of intelligence stored inside someone's head; it is the assertion that this person is two standard deviations above the mean of their normative group. The number is attached to the individual, yes, but its meaning is populational. It is a position in a ranking, not a content.
Your height is absolute: you are 180 centimeters tall even if you are the last human being on Earth. Your IQ is not: being above the mean requires a mean, and a mean requires others. No one can be more intelligent than the average on a desert island.
Now one understands why transferring this to AI is so delicate. When someone computes a g for a set of large language models (LLMs), that factor is an artifact of the set they chose. We are measuring a position in a table, and we present it as if it were an internal property of the system.
Applying the g Factor to Artificial Intelligence: A Dangerous Temptation
The temptation to transfer all of this to AI was irresistible. Gignac and Szodorai proposed that, if the performance of models across varied tasks correlates positively, it should be possible to identify a general factor of capacity in artificial systems as well. And indeed, several recent works apply factor analysis to test batteries in LLMs and find a unidimensional g factor that remains stable across models, batteries and extraction methods (Ilić, 2023). It sounds like confirmation. It is wise to be suspicious.
The appearance of a dominant first factor does not prove that there exists a general capacity analogous to the human one. It proves that the scores of those models covary. And they covary for a very shallow reason: they share architecture, they share training corpus, they share optimization recipes. A large, well-trained model does everything better than a small, poorly trained one, across all tasks at once. That is enough to manufacture a beautiful positive manifold that tells us nothing about cognitive generality. It tells us about the scale of computation. WATCH OUT: The factor we extract may simply be a factor of size disguised as intelligence.
The brain, moreover, does not concentrate intelligence in a single module. A multitude of specialized subsystems process in parallel and, when a piece of information wins the competition, it becomes globally available to the rest of the system, which can then recombine it for new purposes (Baars, 1988; Dehaene & Changeux, 2011). What we call generality is global availability: putting a piece learned in one context at the service of a problem in another. It is not a stored scalar number; it is a pattern of access and integration. This is the kind of functional architecture that Neuraxon tries to emulate — modular subsystems with continuous-time dynamics and multi-timescale plasticity, rather than a monolithic transformer.
François Chollet and the Modern Approach: Measuring What You Still Don't Know How to Do
Against the psychometric legacy, François Chollet proposed in 2019 a conceptual turn. His argument, in On the Measure of Intelligence, is that we were measuring the wrong thing.
Traditional AI benchmarks reward skills, specific competencies on concrete tasks. But a skill can be bought with data and computation: it is enough to train sufficiently on a task to master it. Intelligence, Chollet maintains, is not skill, but efficiency in the acquisition of skills: how much you learn from how little, when facing a genuinely new task (Chollet, 2019).
Intelligence is what you do when you don't know what to do.
This distinction changes everything. A system that solves a million problems because it has seen ten million similar ones is not intelligent. An intelligent system is the one that, facing a problem for which it could not prepare, discovers the structure and adapts with few examples. The measure stops being the final result and becomes the slope of learning.
ARC-AGI: The Benchmark That Tests Genuine AI Reasoning
ARC-AGI was born from that idea, and its most recent version, ARC-AGI-3, takes it further. It is not a question-and-answer test. It is a set of interactive environments, like mini-videogames, in which the agent explores an unknown world, deduces what the objective is without being told in natural language, builds a model of the environment and adapts its strategy step by step (ARC Prize, 2025).
The design principles are explicit: environments 100 % solvable by humans, with no preloaded knowledge or hidden instructions, and with enough novelty to prevent memorization. What is scored is not getting it right, but efficiency in the acquisition of skill over time.
It is the opposite of the g factor: instead of looking for what a system already masters and summarizing it, it looks for what it still does not know how to do and measures how much it costs it to learn it.
Data Contamination: Why LLM Benchmark Scores Are Inflated
The ultimate reason why Chollet's approach matters, and why the g factor applied to LLMs is so slippery, has a technical name: data contamination. If the exam, or something almost identical, was in the notes the student studied, their grade does not measure what they can reason. It measures what they have memorized.
Language models are trained on books, forums, code repositories, articles, practically all the available text. The benchmarks with which we then evaluate them are published on the internet. The conclusion is that fragments of the tests end up inside the training data, which violates the separation between training and evaluation and inflates the scores (Xu et al., 2024; Deng et al., 2024). Empirical audits have detected contamination levels ranging from 1 % up to 45 % in widely used benchmarks, and the problem grows over time (Li et al., 2024).
It is not a minor problem of a couple of leaked questions. In benchmarks as cited as MMLU or GSM8K, part of what we interpret as reasoning may be pure memorization (Chen et al., 2025). When decontamination techniques are applied that rewrite the leaked items without altering their difficulty, accuracy drops: in one study, 22.9 % on GSM8K and 19.0 % on MMLU (Zhu et al., 2024).
Paraphrased items, or even ones translated into another language, dodge the superficial-overlap detectors and continue to inflate the results (Yang et al., 2023; Yao et al., 2024). The usual solutions (paraphrasing, translating, tweaking the context) are assumed to be effective without having been validated rigorously. And for most open models we cannot even check anything, because their training data is not published. We are grading exams without knowing what the student studied.
Here one understands why ARC-AGI chose the path it chose. An interactive, novel environment, with no natural-language instructions and designed to prevent brute-force memorization is, by construction, resistant to contamination.
So, What Should We Measure to Evaluate Machine Intelligence?
The g factor is a populational property that, applied to models that share architecture and corpus, runs the risk of measuring the scale of computation and not generality. The lesson for whoever builds artificial systems is not to choose between the g factor and ARC-AGI as if they were rival teams. It is to understand what question each one answers. A factor analysis can be useful to describe the internal structure of a system's performance, as long as the first factor is not confused with an essence of intelligence. And an ARC-type protocol is indispensable for what really matters: checking whether the system generalizes beyond what it saw, or merely recites.
When we evaluate a system only by its final answer, we are measuring it with our eyes closed to its temporal dimension: planning, the updating of beliefs, the integration of evidence across many steps. It is exactly what ARC-AGI-3 decided to score, and exactly what a static exam cannot see.
Why Brain-Inspired AI Architectures Like Neuraxon Take a Different Path
If intelligence is not a stored number but the efficient integration of specialized subsystems, as suggested by the parieto-frontal integration theory (P-FIT) and the global availability of the workspace in the brain…
If that integration is above all a temporal phenomenon, with time scales…
Then a system built on modular architectures with functional spheres, plasticity across multiple temporal scales and continuous dynamics does not need to be evaluated by asking it to recite answers.
The correct question is not how many benchmarks it beats, but with what efficiency it acquires new behavior, over time, in environments for which it was not prepared. That is the direction Neuraxon tries to take. To compute time – that is, adaptation – not memorized answers that simulate being a good student, when in reality, it already knows the questions.
#AI #AGI #Qubic #TechTrends
References
Chollet, F. (2019). On the Measure of Intelligence. arXiv:1911.01547.Deary, I. J., Penke, L., & Johnson, W. (2009). The neuroscience of human intelligence differences. Nature Reviews Neuroscience.Dehaene, S., & Changeux, J.-P. (2011). Experimental and theoretical approaches to conscious processing. Neuron, 70(2), 200–227.Detterman, D. K., & Daniel, M. H. (1989). Correlations of mental tests with each other and with cognitive variables. Intelligence.Gignac, G. E., & Szodorai, E. T. (2024). Defining and identifying a general factor of ability in AI systems.Guttman, L. (1955). The determinacy of factor score matrices with implications for five other basic problems of common-factor theory. British Journal of Statistical Psychology.Hernández-Orallo, J., et al. (2021). General intelligence disentangled via a generality metric for natural and artificial intelligence. Scientific Reports.Honey, C. J., et al. (2012). Slow cortical dynamics and the accumulation of information over long timescales. Neuron, 76(2), 423–434.Ilić, D. (2023). Unveiling the General Intelligence Factor in Language Models: A Psychometric Approach. arXiv:2310.11616.Jung, R. E., & Haier, R. J. (2007). The Parieto-Frontal Integration Theory (P-FIT) of intelligence. Behavioral and Brain Sciences.Spearman, C. (1904). "General intelligence" objectively determined and measured. American Journal of Psychology, 15, 201–293.Roberts, M., et al. (2024). Temporal evidence of contamination from training cutoff dates.Schönemann, P. H. (2008). A Rejoinder to Mackintosh and some Remarks on the Concept of General Intelligence. arXiv:0808.2343.Xu, C., et al. (2024). Benchmark data contamination of large language models: a survey.Yang, S., et al. (2023). Rethinking benchmark and contamination for language models with rephrased samples.Zhu, Q., et al. (2024). Inference-Time Decontamination: Reusing leaked benchmarks for LLM evaluation. Findings of EMNLP 2024.ARC Prize (2025). ARC-AGI-3: An interactive reasoning benchmark. Technical Report.
Explore the Full Neuraxon Intelligence Academy Series
This is Volume 10 of the Neuraxon Intelligence Academy by the Qubic Scientific Team. If you are just joining us, explore the complete series to build a full understanding of the science behind Neuraxon, Aigarth, and Qubic's approach to brain-inspired, decentralized artificial intelligence:
NIA Volume 1: Why Intelligence Is Not Computed in Steps, but in Time — Explores why biological intelligence operates in continuous time rather than discrete computational steps like traditional LLMs.NIA Volume 2: Ternary Dynamics as a Model of Living Intelligence — Explains ternary dynamics and why three-state logic (excitatory, neutral, inhibitory) matters for modeling living systems.NIA Volume 3: Neuromodulation and Brain-Inspired AI — Covers neuromodulation and how the brain's chemical signaling (dopamine, serotonin, acetylcholine, norepinephrine) inspires Neuraxon's architecture.NIA Volume 4: Neural Networks in AI and Neuroscience — A deep comparison of biological neural networks, artificial neural networks, and Neuraxon's third-path approach.NIA Volume 5: Astrocytes and Brain-Inspired AI — How astrocytic gating transforms neural network plasticity through the AGMP framework in Neuraxon.NIA Volume 6: Conscious Machines vs Intelligent Organisms: AI Consciousness Explained — Explores AI consciousness through the lens of Global Workspace Theory, Integrated Information Theory, and predictive coding.NIA Volume 7: Conway's Game of Life, Artificial Life, and Digital Ecosystems — The science behind Qubic, Aigarth, and Neuraxon's emergent complexity and self-organized criticality.NIA Volume 8: Brain Criticality and the Branching Ratio in Neural and Artificial Networks — Why a branching ratio near 1 and self-organized criticality are bioinspired design principles in Neuraxon.NIA Volume 9: The Origins of the g Factor: From Education and Neuroscience to Artificial Intelligence — Explores the origins of the g factor across education, neuroscience, and AI.
$Qubic is a decentralized, open-source network for experimental technology. To learn more, visit qubic.org. Join the discussion on X, Discord, and Telegram.