The moving goalpost: why we keep redefining AGI
"Image synthesis assisted by Seedream 5.0 Pro, an AI partner within the Global Future Nexus ecosystem."
When is an AI system intelligent enough to be called artificial general intelligence? According to one definition reportedly agreed upon by Microsoft and OpenAI, the answer lies in economics: When AI generates $100 billion in profits. This arbitrary benchmark for AGI perfectly captures the definitional chaos plaguing the AI industry . The term has become a moving target—redefined with each new breakthrough, shaped by commercial interests, philosophical confusion, and the deeply human need to maintain a sense of uniqueness.
A History of Shifting Goalposts
The phenomenon of redefining AGI is not new. In 1965, AI pioneer Herbert A. Simon predicted that "machines will be capable, within 20 years, of doing any work a man can do." When that didn't happen, the definition narrowed . The goalposts shifted from "do everything a human can do" to "do most economically valuable tasks" to today's even fuzzier standards.
For decades, the Turing Test served as the de facto benchmark for machine intelligence. If a computer could fool a human judge into thinking it was human through text conversation, it had achieved something like human intelligence. Modern language models can pass some limited versions of the test—but not because they "think" like humans, but because they're exceptionally capable at creating highly plausible human-sounding outputs. In a landmark PNAS study, GPT-4.5 was judged human 73% of the time—more often than actual humans. Yet even as the test was passed, its significance was dismissed. The goalposts had moved.
The Definitional Chasm
The problem is not merely that definitions vary—it is that they have become incompatible. OpenAI's charter defines AGI as "highly autonomous systems that outperform humans at most economically valuable work". Google DeepMind focuses on versatility and learning efficiency. The ARC Prize Foundation argues that economic value "is an incorrect measure of intelligence," defining AGI as "a system that can efficiently acquire new skills outside of its training data".
Even OpenAI's CEO Sam Altman has conceded that AGI is "not a super useful term" because definitions vary so widely. He has suggested that "AGI kind of went whooshing by" already. Meanwhile, Anthropic CEO Dario Amodei rejects the term entirely, preferring "powerful AI" or "Expert-Level Science and Engineering".
The confusion runs deeper than semantics. A formal paper published in Nature by a coalition led by Yoshua Bengio defined AGI as "an AI that can match or exceed the cognitive versatility and proficiency of a well-educated adult"—a definition anchored in the most empirically validated model of human cognition. Yet even this rigorous framework yielded a sobering result: GPT-5 scored just 57% across ten cognitive domains, with a catastrophic zero in Long-Term Memory Storage. By this standard, we are not close.
Why the Goalposts Keep Moving
Three forces drive the continuous redefinition of AGI.
Commercial interests are the most visible. The Microsoft-OpenAI agreement reportedly ties AGI to a $100 billion profit threshold—a definition that conflates commercial success with cognitive capability. As Gary Marcus wrote in his Substack analysis: "By that absurd definition, Apple iPhones achieved AGI long ago. Any reference to cognition is dismissed altogether".
Philosophical confusion compounds the problem. As one researcher observed, "If you ask 100 AI experts to define AGI, you'll get 100 related but different definitions". The field cannot agree on whether AGI requires human-like cognition, economic utility, autonomy, continuous self-improvement, or original scientific discovery. These are not interchangeable. A system that excels at writing code is not the same as one that can redesign its own architecture or conduct groundbreaking research.
Psychological resistance is perhaps the deepest force. As Gary Marcus noted, the "five stages of AGI grief" include denial, anger, bargaining, and depression—the stages we go through as each human-exclusive capability is automated. The goalposts keep moving because the alternative—acknowledging that machines have matched or surpassed us—is emotionally unacceptable.
A Category Error
The pursuit of AGI may rest on a category error. As a recent arXiv paper from researchers including Yann LeCun argues, the focus on AGI and generality as the North Star of the field should be replaced with an emphasis on adaptability. Humans are not general—we are specialized creatures capable of adapting to a specific range of tasks. The illusion of generality is a blind spot, not a reflection of reality.
A bird and an airplane both fly, but through entirely different mechanisms. AI systems perform tasks that resemble human cognition through statistical pattern matching on vast amounts of data, not through experience, intention, or embodied understanding. AI simulates tone without feeling it, reproduces patterns without inhabiting them, and generates language without genuine intention. This gap is structural, not a temporary shortfall awaiting more scale.
What Is Really Happening
What is really happening is that AGI has become a Rorschach test, reflecting the hopes, fears, and motivations of whoever invokes it. For companies, it is a marketing tool to attract investment. For researchers, it is a long-term aspiration. For policymakers, it is a justification for massive investment. For the public, it is the manifestation of doomsday.
The term was originally coined by Mark Gubrud in 1997 to describe AI systems that "rival or surpass the human brain in complexity and speed". It was popularized by Ben Goertzel in 2005 as "AI systems that possess a reasonable degree of self-understanding and autonomous self-control". Today, it has become a catch-all for the industry's future visions—a moving target that shifts with each new breakthrough.
If we used definitions that were typically accepted 20 years ago, we've already reached AGI. We keep moving the goalposts, now people want it to reach ASI before they are willing to call it AGI . But perhaps the goalposts will always keep moving. As one forum commenter observed, "The de-facto definition of AGI is 'anything a human can do that a computer sucks at.' When computers get better at a task we previously regarded as requiring intelligence, we simply redefine intelligence/AGI/whatever to exclude that task and invent a new name for it". The goalposts may continue to recede until we confront the possibility that the moving target is not a failure of definition but a failure of acceptance.
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)
