Alan Turing was the father of computability. Working with Alonzo Church, Turing showed how different ways of approaching mechanical computing (his own Turing machine and Church’s lambda calculus) were equivalent to each other, and delineated the space of things that could be computed using such methods given nearly infinite time and resources. To this day, new programming tools much demonstrate “Turing completeness” (a minimum set of basic operations) to be considered general purpose.
Turing was a brilliant mathematician and helped Britain break the Enigma code used by Axis forces during WWII. Jolly old England later charged Turing with homosexuality (which at the time was a criminal offense) and died two years into a mandatory regimen of chemical castration, in 1954. If you needed a reason to hate the English and didn’t have one already, the extinguishing of this light in the world is a good candidate.
In 1950, Turing laid a set of ground rules for determining when the science of computing had achieved what we would call “strong” artificial intelligence (i.e. AI that seemed human and could be interacted with like a human — a “thinking machine”). The basis of this test was the Imitation game, requiring three people: two acting participants — one of each sex — and one judge. One acting participant attempts to disguise himself/herself while the other attempts to assist the judge in determining which one is correct. The only form of communication available to the judge is handwritten notes.
In the Turing test, the trickster participant is replaced by a conversational AI. The judge converses with each participant with a text-only communication medium, then attempts to determine which participant is the AI. An AI is considered to pass the test if it has a statistically significant record of fooling judges into believing it is the human participant. Turing predicted that by the year 2000 the best AI systems would have a 30% pass rate.
He was wrong.
[Ed note: This section has been updated from its original 2016 text to reflect factual changes] Nobody had made a conversational AI that could fully pass the test by 2000. Even by 2016, in all cases where chatbots seemed plausibly human in the runnings of such a test, certain smokescreen stipulations were introduced, such as the AI being presented as a nonnative speaker of the language used. A rigorously constructed Turing Test was not passed by a conversational computer program until 2025, when ChatGPT 4.5 fooled judges into thinking that it was the human three-fourths of the time in a Stanford experiment. Though this is a landmark for computers imitating humans in conversation, it is still merely imitation rather than “strong” AI, as a language model like GPT does not exhibit independent thought and only responds with plausible text to prompts from human input. While not a concisely named eponymous rule, Armstrong and Sotala in 2012 noted in a paper that strong AI is always predicted to happen 20 years in the future. It will probably remain so.