AGI Has Arrived - A Provocative Claim from Nature Magazine Signals a Paradigm Shift in AI
Ronni Holmvig Strøm · 2026-02-06
In a commentary published in Nature on February 2, 2026, a team of philosophers, AI researchers, linguists, and cognitive scientists from the University of California, San Diego makes a bold assertion: artificial general intelligence (AGI) is no longer a future milestone—it is already here.
In a commentary published in Nature on February 2, 2026, a team of philosophers, AI researchers, linguists, and cognitive scientists from the University of California, San Diego makes a bold assertion: artificial general intelligence (AGI) is no longer a future milestone—it is already here.
Titled “Does AI already have human-level intelligence? The evidence is clear”, the piece argues that current large language models (LLMs), particularly OpenAI’s GPT-4.5, exhibit broad, flexible cognitive competence across domains that matches or exceeds human performance.
The authors—Eddy Keming Chen (philosophy), Mikhail Belkin (AI and data science), Leon Bergen (linguistics and computer science), and David Danks (data science, philosophy, and policy)—frame their case as an “inference to the best explanation” grounded in behavioral evidence, echoing Alan Turing’s 1950 imitation game while updating it for 2026 realities.
The Cascade of Evidence
The commentary builds its argument through a progression of achievements that demonstrate both breadth (spanning mathematics, language, science, reasoning, creativity, and everyday conversation) and depth (expert or superhuman performance within those domains):
Turing Test Milestone** — In March 2025, GPT-4.5 was judged human 73% of the time in a controlled Turing test—outperforming actual humans in the same setup (Jones & Bergen, 2025 preprint).
Mathematical Excellence** — Gold-medal-level performance at the International Mathematical Olympiad and collaboration with leading mathematicians on theorem proofs.
Scientific Innovation** — Generation of hypotheses that have been experimentally validated in real labs.
Expert-Level Problem-Solving** — Solving PhD-level exam problems, assisting professional programmers, and composing poetry that readers prefer over human-authored work in blind tests.
Everyday Breadth** — Engaging in 24/7 global conversations with hundreds of millions of people across diverse topics.
The authors emphasize that general intelligence is a functional property, inferred from observable behavior rather than internal mechanisms—just as we attribute intelligence to other humans without scanning their brains. They argue that demanding perfection, embodiment, or independent agency sets an arbitrary bar that even humans would fail.