The taxonomy was named in 1950. The thing it describes first existed — in any meaningful sense — around 2012. That 62-year gap is where most AI history gets the story wrong.
1 July 2026 · Case 1 of 5
As you read — hold this question
When does naming a thing become evidence that the thing exists?
62yrs
between Turing naming "artificial intelligence" (1950) and the first system most people would recognise as genuinely intelligent (2012). The label preceded the reality by more than six decades.
What existed before roughly 2012 was computer representations of intelligence — rules encoded by humans, statistical patterns found in data humans pointed systems at. ELIZA substituted keywords into pre-written responses. Deep Blue evaluated 200 million chess positions per second using hand-coded heuristics. These were impressive engineering feats. They were not intelligence, artificial or otherwise. Calling them AI was aspirational branding, not description.
The conventional story
Quiz: ANI / AGI / ASI
What the textbooks say — a smooth arc from 1950 to today
The standard history presents AI as a continuous journey starting with Turing. Every decade has a milestone. Each one is framed as progress toward the same goal.
1950
Turing Test
"AI defined"
1956
Dartmouth Conference
"Solved in 10 years"
1980s
Expert Systems
"AI is here"
1997
Deep Blue
"AI triumphs"
2011
Watson
"Cognitive era"
2022
ChatGPT
"AGI by 2025"
2026
Huang claim
"AGI achieved"
Problem: this timeline implies continuous, coherent progress. It conflates when things were named with when they became capable.
The corrected capability timeline
What AI could actually do — year by year
Large circles mark era-defining capability leaps. Small circles mark granular milestones within each era. The x-axis is non-linear — 1950–2012 compressed on the left (62 years, little happens), 2012–2026 expanded on the right (14 years, everything happens). Hover any dot for detail.
1950–1979 · Conceptual / Rule-based
1980–2011 · Statistical pattern matching
2012–2016 · Narrow deep learning
2017–2021 · Foundation models / Emergence
2022–2026 · Practical multi-domain capability
⬤ = era leap · = granular milestone
The finding — two very different stories
What changes when you correct the timeline
The conventional story
AI has been progressing since the 1950s — a 75-year journey
Each decade has milestones: ELIZA, Deep Blue, Watson, AlphaGo, ChatGPT
We are approaching AGI as the next natural step in a long arc
Jensen Huang's 2026 AGI claim fits within this story of continuous progress
The corrected picture
Pre-2012: computer representations of human rules and statistics — not intelligence
Real capability begins in 2012 (narrow deep learning) and matures in 2022 (multi-domain)
Practical AI has existed for approximately 4 years at scale
Huang's AGI claim arrives just 4 years into AI's practical existence — an extraordinary claim on a very short track record
Applied — Jensen Huang's AGI claim, 2026
Run the taxonomy against the corrected baseline
Huang stated in early 2026 that AI has achieved AGI. Measured against the corrected timeline — and against the ANI/AGI/ASI taxonomy with independent evidence — the claim doesn't hold.
AGI criterion
Verdict
Cross-domain transfer without retraining — does capability in one domain automatically transfer to unrelated domains?
✗ Fails
Human-level reasoning — can the system reason correctly across novel problems at or above the 50th percentile of skilled humans? (DeepMind L2 Competent)
✗ Fails
Genuine autonomy — can it set and pursue goals over extended periods without human-defined task framing?
✗ Fails
Outperforms humans at most economically valuable work (DeepMind L2–L3 range)
~ Partial
High-value performance at scale — commercially deployed at meaningful scale
✓ Passes
The context the taxonomy alone can't show
Against the corrected timeline, Huang's claim is even more striking. Practical AI has existed for roughly four years. The field spent 60+ years calling pattern-matching and tree-search "artificial intelligence." We are now four years into systems that actually deserve the label — and the claim is already AGI. Every major AI benchmark since 2022 has been beaten faster than predicted — but DeepMind's independent measurement (Morris et al.) still places every current LLM at Level 1 Emerging. The claim outpaces the evidence by four capability levels.
Take this away
The taxonomy was named in 1950 but practical AI has existed for roughly four years—which means every confident claim about where we are on the path to AGI is built on an extraordinarily short track record.
What does the corrected capability timeline reveal that the conventional AI history hides?
The conventional story presents AI as a 75-year continuous journey from Turing (1950) to today. The corrected timeline shows that pre-2012 systems—ELIZA, Deep Blue, Watson—were computer representations of human rules and statistics, not intelligence. Genuine narrow capability begins in 2012 with AlexNet. Practical multi-domain capability arrives in 2022 with ChatGPT/GPT-4. Practical AI has existed for approximately four years.
Question 2 of 3
Huang claimed in 2026 that AI has achieved AGI. Against DeepMind’s five criteria, which does current AI fail—and which does it pass?
Current AI fails three criteria: (1) cross-domain transfer without retraining; (2) human-level reasoning at DeepMind’s L2 Competent benchmark; (3) genuine autonomy over extended periods without human-defined task framing. It achieves partial credit on outperforming humans at most economically valuable work, and passes on high-value commercial deployment at scale. DeepMind’s independent assessment (Morris et al.) places every current LLM at Level 1 Emerging—four levels below AGI.
Question 3 of 3
What can the ANI/AGI/ASI taxonomy alone not tell you about Huang’s claim—and what does the corrected timeline add?
The taxonomy shows where current systems sit against the AGI criteria—but it cannot show track record. The corrected timeline adds that dimension: practical AI has existed for roughly four years. The field spent over 60 years calling pattern-matching “artificial intelligence.” We are now four years into systems that actually deserve the label—and the claim is already AGI. The claim outpaces the evidence by four DeepMind capability levels, made on a very short baseline.
Sources
Turing (1950)
Turing, A.M. (1950). Computing machinery and intelligence. Mind, 59(236), 433–460.
Dartmouth (1956)
McCarthy, J., Minsky, M., Rochester, N., & Shannon, C. (1956). A proposal for the Dartmouth summer research project on artificial intelligence.
ANI/AGI/ASI taxonomy
Russell, S., & Norvig, P. (2020). Artificial Intelligence: A Modern Approach (4th ed.). Pearson.
DeepMind levels
Morris et al. (2024). Levels of AGI: Operationalizing Progress on the Path to AGI.arXiv:2311.02462.
Gartner Hype Cycle
Gartner (2025). Hype Cycle for Artificial Intelligence. GenAI: Trough of Disillusionment. Agentic AI: Peak of Inflated Expectations.
Transformer
Vaswani et al. (2017). Attention is all you need. NeurIPS 2017.
Innovare Index
Pricing and capability data from Innovare Index, Edition 006, 1 July 2026.