The Turing test is dead
"Image synthesis assisted by Microsoft Copilot, an AI partner within the Global Future Nexus ecosystem."
After 75 years, Alan Turing's famous test has been decisively passed — and the journal Nature has declared that its obsolescence signals the arrival of the AGI era.
A Landmark Declaration
In February 2026, Nature published a landmark Comment by four UC San Diego scholars — spanning philosophy, machine learning, linguistics, and cognitive science — that has sent shockwaves through the AI community . Their conclusion is unequivocal: by any reasonable standard, Alan Turing's 1950 vision of human-level machine intelligence has become reality. The era of AGI is no longer a future horizon; it is already here.
This is not an opinion piece. It is a rigorous argument based on evidence from multiple domains: gold-medal performance on International Mathematical Olympiad problems, PhD-level scientific reasoning, generation of experimentally validated hypotheses, and — most famously — the decisive passing of the Turing test itself .
The Test That Changed Everything
The evidence is now empirical. In a randomised, controlled, preregistered Turing test published in PNAS, GPT-4.5 was judged to be human 73% of the time — significantly more often than interrogators selected the real human participant . LLaMa-3.1-405B achieved 56%, statistically indistinguishable from the humans it was compared against . For the first time in 75 years, a machine has passed the original three-party format with flying colours .
Crucially, this success depends on prompting the models to adopt a humanlike persona — an "ordinary, slightly introverted person" who makes typos, hedges, and uses casual slang. Without this persona, the win rate for GPT-4.5 drops to 36% . This reveals that the Turing test is not a test of raw intelligence, but of social imitation. And modern LLMs have mastered that.
Why the Turing Test No Longer Matters
The scholars argue that the Turing test's retirement is overdue for several reasons:
The ELIZA effect has been weaponised. As early as the 1960s, Joseph Weizenbaum's ELIZA chatbot showed that people are easily fooled into attributing intelligence to simple pattern-matching . Today's LLMs exploit the same cognitive vulnerability, but at a vastly greater scale. The test tells us more about human gullibility than machine intelligence .
Intelligence is not deception. The Turing test is fundamentally about a machine's ability to fool an interrogator. But why should intelligence be connected to deception? A revised, community-based test would assess whether an AI can be accepted as a member of a human community over time — not whether it can impersonate one in a five-minute chat .
The goalposts have shifted. Critics argue that the Turing test has become a philosophical exercise rather than a practical benchmark. As one Nature commentator wrote: "Statistical approximation is not general intelligence" . Passing the test demonstrates imitation, not understanding, reasoning, or consciousness.
The AGI Cascade
The UC San Diego scholars propose a three-tier cascade for assessing general intelligence:
Tier 1 (Turing-test level): Basic literacy and adequate conversation — already passed.
Tier 2 (Expert level): Gold-medal Olympiad performance, PhD-level problem-solving in multiple domains, competent creative and practical reasoning — already achieved by frontier models .
Tier 3 (Superhuman level): Revolutionary scientific breakthroughs that few humans meet — the next frontier.
They argue that AGI does not require perfection, universal mastery, or human-like cognition. It requires "the flexible, general competence characteristic of human thought." By that standard, current LLMs already qualify .
The Debate Continues
The declaration has sparked fierce debate. Andrew Ng has proposed a new "Turing-AGI Test" focused on completing multi-day work tasks reliably . Marc Andreessen has stated that "GPT-5.5, Claude 4.6 and Gemini 3.0 are now as smart as a person" . Yet critics maintain that LLMs are "idiots savants" — capable of retrieving vast knowledge, but reasoning poorly and superficially .
What is clear is that the Turing test has outlived its usefulness. As Gary Marcus warned at the Royal Society anniversary event, "LLMs are deeply flawed imitators that are preying on the ELIZA effect" . The test no longer tells us anything meaningful about whether a machine can think.
The GFN Context
For Global Future Nexus, the retirement of the Turing test is not an academic curiosity — it is a governance imperative. If machines can convincingly pass as human, how do we maintain trust? How do we prevent sophisticated fraud? How do we ensure accountability when we cannot distinguish authentic from synthetic?
The test's obsolescence reinforces the urgency of GFN's mission: to build the identity, trust, and governance frameworks required for a world where human and machine intelligence are indistinguishable in conversation, but must remain distinguishable in accountability.
The Turing test is dead. The work of coexistence has only just begun.
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)
