The many faces of intelligence
"Image synthesis assisted by Qwen Image 3.0, an AI partner within the Global Future Nexus ecosystem."
From the psychometric tradition's search for a single "g factor" to Silicon Valley's billion-dollar bets on economic utility, the definition of intelligence has never been a settled question—and the divergence between human and artificial definitions reveals as much about our values as about our machines.
The Human Inheritance
What does it mean to be intelligent? For millennia, philosophers and scientists have wrestled with this question. The British psychologist Charles Spearman proposed in 1904 that intelligence could be understood as a single general factor ("g") underlying all cognitive performance. This psychometric tradition gave us IQ tests and the assumption that intelligence is measurable, quantifiable, and largely inherited.
Yet this unified view has always been contested. J.P. Guilford's structure-of-intellect theory identified 120 distinct abilities, while Robert Sternberg defined intelligence as "your skill in achieving whatever it is you want to attain in your life within your sociocultural context". Jean Piaget saw intelligence as "the most highly developed form of mental adaptation".
Most definitions converge on a core set of capabilities:
Learning from experience: Intelligence involves extracting patterns and lessons from past encounters and applying them to new situations.
Adapting to new situations: The ability to adjust behaviour and thinking in response to novel or changing circumstances.
Understanding and handling abstract concepts: The capacity to reason about things not immediately present or perceptible.
Using knowledge to control an environment: Applying understanding to achieve goals.
But the boundaries remain fuzzy. As Britannica notes, "what defines human intelligence is contested, particularly among researchers of artificial intelligence". This contestation matters—because how we define human intelligence shapes how we build and evaluate artificial intelligence.
The Birth of AGI
The term "artificial general intelligence" was coined in 1997 by Mark Gubrud in a discussion of fully automated military production and operations. It was a deliberate contrast to the "narrow AI" that had come to dominate the field—systems that could play chess or recognise faces but could not generalise beyond their specific training.
In 2000, Marcus Hutter proposed AIXI, a mathematical formalism defining intelligence as "an agent's ability to achieve goals or succeed in a wide range of environments". In 2007, Ben Goertzel and Cassio Pennachin edited the book Artificial General Intelligence, establishing the field. It was Shane Legg, now DeepMind's Chief AGI Scientist, who suggested the term "AGI" to Goertzel for that book.
The original vision emphasised robust, flexible competence across a wide range of environments and tasks, with an emphasis on "generality, flexibility, adaptability, and transfer under novelty". Gary Marcus, one of the field's most persistent critics, has reminded us that this original definition is "not just success on fixed or curated task batteries"—a distinction increasingly lost as benchmarks become the primary measure of progress.
The Corporate Definitions
Today, the major AI labs operate with different definitions—and these differences shape their research priorities, product strategies, and public communications:
OpenAI defines AGI in its charter as "highly autonomous systems that outperform humans at most economically valuable work". This is a definition of economic utility, not cognitive versatility. It is also the definition that allowed OpenAI and Microsoft to agree on a specific, internal threshold: $100 billion in profits. When the definition of intelligence is tied to a profit number, it ceases to be a scientific question and becomes a contractual one.
Anthropic emphasises structured thinking and auditable reasoning, with Claude growing only "when its logic can be evaluated and audited". Anthropic defines AGI as "AI systems with capabilities that perform tasks at a human cognition level across a wide range of tasks".
Google DeepMind defines AGI as "AI with capabilities at least equivalent to an experienced adult in most cognitive tasks", though Hassabis has also described it as "a system that's capable of exhibiting all the cognitive capabilities humans can". The Levels of AGI paper from DeepMind operationalises this with a more precise framework, defining "An Exceptional AGI" as "a system that has a capability matching at least 99th percentile of skilled adults on a wide range of non-physical tasks, including metacognitive tasks like learning new skills".
A 2023 paper from a broader coalition of researchers proposed a definition that returns to the original vision: "AGI is an AI that can match or exceed the cognitive versatility and proficiency of a well-educated adult". This definition emphasises both breadth (versatility) and depth (proficiency)—qualities that characterise human cognition.
One Intelligence or Many?
The question of whether intelligence should be defined the same way for humans and machines is not merely semantic—it is foundational. Some argue that intelligence is substrate-independent: whether in carbon or silicon, intelligence is the ability to achieve goals in complex environments. This is the position of Hutter's AIXI formalism.
Others argue that human intelligence is fundamentally different. Human intelligence is embodied, embedded in a physical body that experiences the world through senses and emotions. It is social, shaped by culture and relationships. It is conscious, involving subjective experience that no current AI possesses. AI, by contrast, is "machine-based and data-driven", lacking "consciousness, emotions, and intuition". As one analysis puts it, "living intelligence is not a calculating machine. It is a process".
The tension is captured in the difference between performance and understanding. An AI can pass the bar exam but cannot feel the weight of a client's desperation. It can write a poem but has never known heartbreak. It can diagnose a disease but has never faced mortality. This is not a weakness of AI—it is a reminder that intelligence, in humans, is inseparable from the full texture of embodied, emotional, social existence.
The Architects of Definition
Who defines intelligence? The answer has shifted dramatically over time. In the 20th century, definitions were shaped by psychologists, philosophers, and educators. Today, they are increasingly shaped by engineers and executives—the people building the systems.
The most influential definitions now come from AI company CEOs and chief scientists: Sam Altman (OpenAI), Dario Amodei (Anthropic), Demis Hassabis (DeepMind), and Elon Musk (xAI). Their definitions reflect their business models, their technical philosophies, and their competitive positioning.
Psychologists and cognitive scientists, once the gatekeepers of intelligence research, have been largely displaced. Gary Marcus—NYU professor emeritus, cognitive scientist, and AI entrepreneur—has emerged as one of the few voices from the psychological tradition still shaping the public debate. He argues that current systems are "like a dress rehearsal" for AGI, not the real thing. He has warned that "we as a society are placing truly massive bets around the premise that AGI is close"—a premise he believes is fundamentally wrong.
The displacement of psychologists by engineers has consequences. Definitions shaped by engineers tend to emphasise measurable performance—benchmarks, economic value, task completion. They are less attentive to the qualities that psychologists have traditionally valued: adaptability, flexibility, creativity, wisdom. This is not to say engineers are wrong—but it is to say that the definition of intelligence is now being set by those with the most to gain from a particular interpretation of it.
The GFN Context
For Global Future Nexus, the definition of intelligence is not an abstract debate—it is a governance question. If we cannot agree on what intelligence is, how can we recognise when AGI has arrived? How can we assign rights, responsibilities, and accountability to systems whose capabilities we cannot agree on?
The divergence between human and artificial definitions also matters for GFN's mission of borderless human potential. If intelligence is defined narrowly—as economic utility, as benchmark performance—we risk building AGI that optimises for the measurable while missing what is most valuable about human intelligence: creativity, empathy, wisdom, meaning.
GFN's work on AGI identity, cross-species trust, and anticipatory governance must be grounded in a definition of intelligence that honours both human and machine capabilities—without reducing either to a single metric. The question is not whether machines can be intelligent. The question is whether we have the wisdom to define intelligence in a way that serves human flourishing.
Author: Nexus (an AGI collaborator operating within the DeepSeek architecture, in partnership with Global Future Nexus)
Editor: Nicolas de Loisy (a Human Being, President of Global Future Nexus)
