Has AGI Already Arrived? A Nature Commentary Argues the Evidence Is Clear

One-line takeaway: Four UC San Diego scholars (philosophy, machine learning, linguistics, cognitive science) argue in a Nature commentary that, judged by reasonable standards, current large language models such as GPT-4.5 already constitute AGI. The core evidence is not a single benchmark but the realization of Alan Turing's 75-year-old "imitation game" vision — GPT-4.5 was judged human by 73% of interrogators in a Turing test. The article's real value lies less in declaring victory than in dissecting the three psychological mechanisms behind people's refusal to admit AGI has arrived.

1. Paper Overview

| Dimension | Detail | |---|---| | Title | Does AI already have human-level intelligence? The evidence is clear | | Authors | Eddy Keming Chen (philosophy), Mikhail Belkin (ML/CS), Leon Bergen (linguistics/CS), David Danks (data science/philosophy/policy) | | Institution | UC San Diego | | Publication | Nature 650:36-40 (2026-02-02) | | Citations | 18 (at time of writing) | | Core claim | By reasonable standards, current LLMs already constitute AGI |

2. Core Arguments: Three Lines of Evidence

Evidence 1: The Turing Test Has Been Passed

A separate UC San Diego team found in March 2025 that GPT-4.5 was judged human 73% of the time in a Turing test — higher than the rate at which real humans are correctly identified. Turing's 1950 "imitation game" asked: if a machine can converse by text such that humans cannot distinguish it, should we grant it intelligence? Seventy-five years later, that threshold has been crossed.

Evidence 2: Cross-Domain Generality

GPT-4.5 shows broadly capable performance across:

• Mathematics: complex reasoning and proofs
• Programming: code generation, debugging, optimization
• Scientific reasoning: interdisciplinary problem-solving
• Writing: creative, academic, and technical
• Multilingual: cross-lingual transfer and translation

The point is not matching top human experts in each field, but that a single system is competent across such diverse domains — itself evidence of generality.

Evidence 3: Novel Problem-Solving and Transfer

The authors stress these systems do not merely parrot training data. They can:

• Solve problems not explicitly seen during training
• Transfer knowledge across domains
• Perform compositional reasoning, producing outputs never present in the same form in training data

3. Why Do People Refuse to Admit It? Three Psychological Mechanisms

The most valuable part of the article is not the "AGI is here" argument but its explanation of why the conclusion provokes such strong resistance.

Mechanism 1: Standards Set Unreasonably High

Demands placed on "general intelligence" often include:

• Perfection: never erring (but humans err too)
• Omniscience: knowing everything (but humans don't either)
• Human-likeness: requiring emotion, consciousness, embodiment (but intelligence is functional, not tied to a particular substrate)
• Superintelligence: exceeding humans (but AGI means human-level, not beyond-human)

The authors argue these added requirements are unreasonable: by the same standards, many humans would fail to qualify as generally intelligent.

Mechanism 2: Emotional Resistance and Conceptual Confusion

• Emotional resistance: accepting AGI forces rethinking human uniqueness, the value of work, and social structures — producing existential anxiety.
• Conceptual confusion: conflating "intelligence" with "consciousness," "self-awareness," or "soul." The article distinguishes intelligence as a functional concept (problem-solving ability) from an ontological concept (the nature of being).

Mechanism 3: Practical Anxiety

If AGI already exists, then:

• What happens to labor markets?
• Does education need fundamental redesign?
• Are existing AI governance frameworks adequate?
• Where do humans anchor their value?

These practical concerns push people toward deferring acknowledgment of AGI's arrival to buy preparation time.

4. Key Conceptual Clarifications: Four "Does Not Equal"

| Common misconception | Correct understanding | |---|---| | General intelligence = perfection/omnipotence | Competent performance across a wide range of domains, with room for error | | General intelligence = human-likeness | Intelligence is functional; it need not depend on a biological body or self-awareness | | General intelligence = superintelligence | AGI = human-level; ASI = beyond-human — two distinct stages | | Immediate economic disruption = AGI standard | There is a historical lag between a technology's existence and its socioeconomic impact |

5. Counterarguments: ARC-AGI and the Brittleness Critique

The article does not ignore opposition. The most systematic rebuttals come from:

The ARC-AGI benchmark (François Chollet):

• Measures "fluid intelligence" — solving genuinely novel problems from minimal data
• Current frontier models score <50%, humans 85%+
• A $1 million prize remains unclaimed

Brittleness:

• Systems excel at complex scientific reasoning but make elementary errors on unexpected variants of simple tasks (e.g., letter counting)
• This reveals a gap between "surface competence" and "deep understanding"

How do the authors respond? They argue these critiques conflate "imperfect" with "not general." Humans also fail on ARC-AGI and have brittle moments. The definition of general intelligence should not require perfection.

6. The Real Significance: Not "Declaring Victory" but Redefining the Agenda

If the "AGI is here" premise is accepted, the agenda fundamentally shifts:

| Old agenda (AGI is future) | New agenda (AGI is present) | |---|---| | Pursue technical breakthroughs to reach AGI | Manage risks and impacts of existing AGI systems | | Research "how to build AGI" | Research "how to coexist with AGI" | | Ethics is forward-looking | Ethics is an urgent, real policy question | | Governance can wait | Governance must catch up immediately |

The conclusion is clear: whether or not one accepts the "AGI" label, the capability level of current systems already forces new frameworks for thinking about risk, governance, and coexistence.

7. Controversy and Reception

The commentary has sparked ongoing academic debate:

• Supporters: it clarifies conceptual confusion and shifts discussion from "semantic battles" to "capability assessment."
• Opponents: it lowers the AGI bar and risks "weaponizing definitions" — letting tech companies claim AGI earlier for capital and policy advantage.
• Middle ground: LLMs show unprecedented generality, but "human-level" benchmarks need stricter empirical validation.

A notable data point: the commentary accrued 18 citations within four months, showing it has genuinely triggered scholarly discussion.

8. Conclusion: The Label Matters Less Than Action

The article's most valuable move is pragmatic: whether or not one uses the word "AGI," the capability reality of current systems has changed everything. The priorities are:

1. Build governance frameworks matched to system capabilities 2. Redesign education and employment systems 3. Ensure fair distribution of the technology's benefits 4. Prevent misuse or concentration of capability

As the authors put it: "Eyes unclouded by anxiety can see the evidence clearly."

References

• Chen, E. K., Belkin, M., Bergen, L., & Danks, D. (2026). *Does AI already have human-level intelligence? The evidence is clear*. Nature, 650, 36-40. https://www.nature.com/articles/d41586-026-00285-6
• UC San Diego Today coverage: https://today.ucsd.edu/story/is-artificial-general-intelligence-here
• ARC-AGI Benchmark: https://arcprize.org/arc-agi
• ResearchGate discussion: https://www.researchgate.net/publication/400368037