Turing Test Explained: Has AI Passed It in 2026?
Quick answer: The Turing test, introduced by Alan Turing in 1950, asks whether a machine can imitate human conversation well enough that a judge chatting by text cannot tell it from a person. Some AI systems have now passed controlled versions: in research by Cameron Jones and Benjamin Bergen, published in PNAS in May 2026, GPT-4.5 with a humanlike persona was judged to be the human in 73% of five-minute conversations, and GPT-5 reached 59% in 15-minute conversations. These results show convincing imitation, not proof of humanlike understanding or consciousness.
Facts checked November 22, 2026.
“Can machines think?” Alan Turing opened his 1950 paper “Computing Machinery and Intelligence” with that question, then replaced it with a test you could actually run. For 75 years the Turing test was a distant goal. In 2025, researchers reported the first rigorous evidence that an AI passed it. Here is what the test is, how it is scored, and what passing does and does not mean.
The Imitation Game
Turing’s setup has three participants: a human judge, a human witness and a machine. The judge chats with both by text, without seeing them, and must decide which is the human. If judges cannot do better than chance, the machine has done well. Turing predicted that by about the year 2000, computers would play so well that an average judge would have no more than a 70% chance of making the right identification after five minutes of questioning. That was a forecast, not an official passing threshold; modern studies instead compare the machine with chance and with real humans.
The Math of Passing: What Is “Chance”?
In the three-party version, the judge picks one of two witnesses as the human. A judge who guesses randomly picks the machine 50% of the time. So:
| Machine “win rate” (judged human) | Interpretation |
|---|---|
| Well below 50% | Judges can spot the machine |
| About 50% | Judges cannot tell the difference |
| Well above 50% | The machine seems more human than the humans |
Whether a result is “well” above or below 50% is a statistics question. As an illustration, with 100 independent judgments the standard error of a proportion near 50% is \(\sqrt{0.5 \times 0.5 / 100} = 0.05\), so a 73% win rate would sit \(\frac{0.73 – 0.50}{0.05} \approx 4.6\) standard errors above chance, very unlikely to be luck. The real study described below analyzed 1,023 games in its main experiments and used pre-registered statistical tests. (For more on margins of error, see how to read AI benchmarks.)
The Research: GPT-4.5 and GPT-5 in Turing Tests
Cameron Jones and Benjamin Bergen of UC San Diego posted their study in March 2025, and it was published in the Proceedings of the National Academy of Sciences (PNAS) in May 2026. Participants chatted simultaneously with a real person and an AI, then chose which was human. The main studies used five-minute conversations; a replication with 15-minute conversations tested GPT-5.
| System | Conversation length | Judged to be the human |
|---|---|---|
| GPT-4.5 (with a humanlike persona prompt) | 5 minutes | 73% |
| LLaMa-3.1-405B (same prompt) | 5 minutes | 56% |
| GPT-4o (baseline) | 5 minutes | 21% |
| ELIZA (1966 chatbot) | 5 minutes | 23% |
| GPT-5 (with persona prompt) | 15 minutes | 59.3% |
GPT-4.5 was chosen as the human significantly more often than the real human participants were; LLaMa-3.1 was statistically indistinguishable from the humans. The authors call it the first empirical evidence that an artificial system passes a standard three-party Turing test. Longer conversations did not stop persona-prompted models from passing. Note the role of the persona prompt: without instructions to act like a young, introverted person fluent in internet slang, the models did much worse. Results depend on the model version and setup, so they do not mean every chatbot conversation would fool you.
Earlier Claims
In 2014, a chatbot called Eugene Goostman, posing as a 13-year-old Ukrainian boy, was reported to have convinced 33% of judges at an event in London, a claim that remains disputed. Critics argued that the persona excused errors and that five-minute chats with a few judges were too weak a test. The annual Loebner Prize ran restricted Turing tests from 1991 to 2019 without any program convincingly passing.
Does Passing Mean AI Can Think?
- Turing’s own view: he proposed the test to replace the vague question “Can machines think?” with one that can be measured.
- The Chinese Room: in 1980, philosopher John Searle argued that a system could manipulate symbols perfectly without understanding them.
- Imitation vs ability: a model can pass a five-minute chat yet still make basic factual or math mistakes. Researchers increasingly measure specific skills instead, with benchmarks for math, coding and reasoning.
The practical lesson is social, not philosophical: if AI can pass as human in short text conversations, scams, fake accounts and bots become harder to spot.
The Turing Test in Everyday Life: CAPTCHAs
CAPTCHA stands for “Completely Automated Public Turing test to tell Computers and Humans Apart.” Websites use these reverse Turing tests, where a computer judges whether you are human. As AI gets better at reading distorted text and images, CAPTCHAs have shifted toward tracking behavior instead of puzzles.
Common Misconceptions
“The Turing test has one official set of rules.” Turing described it loosely; results depend on duration, judges, witnesses and prompts.
“Passing means AI is conscious.” The test measures conversational imitation, not inner experience.
“Chatbots passed decades ago.” Earlier claims used weak setups; the Jones and Bergen study, peer-reviewed in 2026, was the first rigorous, pre-registered three-party result.
“A 73% score means AI is 73% human.” It means judges picked the AI as the human in 73% of games, 23 percentage points above the 50% expected by chance.
Watch: The Turing Test Explained
If you learn better by watching, these videos are worth your time.
Check Your Understanding
Five quick questions to see how much of this has stuck. Pick an answer and press Submit; you will see a short explanation either way.
Answer each question and press Submit to see your result right away.
Frequently Asked Questions
What is the Turing test?
A test proposed by Alan Turing in 1950: a human judge chats by text with a human and a machine and tries to tell which is which. A machine passes if judges cannot reliably identify it.
Has AI passed the Turing test?
In controlled studies, yes. Research by Cameron Jones and Benjamin Bergen, published in PNAS in May 2026, found GPT-4.5 with a humanlike persona was judged human 73% of the time in five-minute three-party tests, and GPT-5 reached 59% in 15-minute tests.
Has ChatGPT passed the Turing test?
Specific versions with special persona prompts did in research settings: GPT-4.5 and GPT-5. Results depend on the model, the prompt and the test setup, and the baseline GPT-4o scored only 21%.
Does passing the Turing test mean AI is intelligent?
It shows AI can imitate human conversation convincingly. Whether that amounts to thinking or understanding is debated, for example in John Searle’s Chinese Room argument.
What did Alan Turing predict?
That by about 2000, an average judge would have no more than a 70% chance of identifying the machine correctly after five minutes of questioning.
Is a CAPTCHA a Turing test?
Yes, a reverse one: a computer tries to tell whether you are a human or a bot.
Keep Exploring
- Why Does AI Hallucinate? Math, Examples and Fixes
- Can AI Do Math? What to Trust in 2026
- How to Read AI Benchmarks: 7 Statistical Checks
Sources and Further Reading
- Turing, Computing Machinery and Intelligence (Mind, 1950)
- Jones and Bergen, Large language models pass a standard three-party Turing test (PNAS, May 2026)
- UC San Diego: AI can seem more human than real humans in a classic Turing test (May 19, 2026)
- Jones and Bergen, Large Language Models Pass the Turing Test (arXiv 2503.23674, Mar 2025)
- Stanford Encyclopedia of Philosophy: The Turing Test
- Stanford Encyclopedia of Philosophy: The Chinese Room Argument
Related to This Article
More math articles
- Algebra 1 Worksheets: 64 Free PDFs with Answers
- Interpreting election returns
- National Identity and Sectional Culture: Labor, Slavery and Regional Life
- How to Apply Trigonometry: Practical Uses and Insights into Engineering and Astronomy
- Hawaii SBAC Grade 4 Math Free Worksheets: Printable Practice Worksheets with Worked Solutions
- The Best Grade 7 Math Book for Nevada Students
- 5th Grade DCAS Math Worksheets: FREE & Printable
- The Best Grade 3 ELA Practice Tests for Mississippi Students
- 7th Grade WY-TOPP Math Worksheets: FREE & Printable
- BMI Calculator for Men


What people say about "Turing Test Explained: Has AI Passed It in 2026? - Effortless Math"?
No one replied yet.