How AI Ate the World (eBook)

2

THE FATHERS OF AI

Today, a number of individuals currently debating the potential and pitfalls of our generative AI revolution have been given the mantle of 'godfathers of AI' by the media. But expand the timeline back to the 1940s and 1950s, and AI had two godfathers who predate those currently laying claim to the title: one in the UK and the other in the USA. Both were developing their thinking just after World War II. And in a way, both were outsiders ahead of their time.

In Britain, Alan Turing, a mathematician who helped crack Germany’s Enigma codes during World War II, thought the human brain was, in many ways, a biological version of a digital computing machine. 

Turing was a strange and isolated figure who was decades ahead of his time. Growing up, he and his brother, John, rarely saw their parents (his father was part of the Madras government in India) and were left in the care of a retired colonel who lived in Hastings. Turing’s mind was prodigious, and a natural fit for mathematics but he didn’t fit in at school, where the classics, and writing, were seen as more important than numeracy: his English teacher said his writing was ‘the worst I have ever seen, and I try to view tolerantly his unswerving inexactitude and slipshod, dirty, work.’ His headmaster was no kinder: he was 'the sort of boy who is bound to be a problem for any school or community.’

Yet the young Turing excelled at physics, quantum mechanics and the working of the mind. He devoured Einstein’s theory of relativity, and went to Cambridge University where he graduated with a maths degree. After he was dragooned into Bletchley Park’s Government Code and Cypher School, he earned his PhD at Princeton in the US. Although Turing became known for his codebreaking prowess in cracking Enigma, he would play several key roles in the early years of British computing. His work was often theoretical, because the hardware required to bring his inventions to life simply didn’t exist in his lifetime. He ended up having an outsized impact on the development of AI.

Take one moment in 1948 as an example of Turing developing ideas on paper that he couldn’t yet put into practice. Since the early 1940s, Turing had believed that machines could mimic how human brains worked through programmed instructions to conduct some tasks, like playing games. One of those games he believed was a potential early use for computers was chess. In 1948, while working at Cambridge, Turing and a colleague, David Gowen Champernowne, developed a computer program called Turochamp that would follow rules laid down by its creator to play the right moves in a game of chess.

Turing and Champernowne didn’t get to test out Turochamp (named after its inventors) on an actual computer because the tech at the time wasn’t powerful enough to run the program. Turing would actually die before the program was loaded into a computer. But it worked – and chess, as we’ll learn in a subsequent chapter, would be intrinsically tied up with the development of AI.

Each step along his journey, from breaking Enigma to developing Turochamp, imbued Turing with the skills he’d need to start helping create alongside others, in broad outline, what would become the field of artificial intelligence. Unlike others who believed we humans are born with innate abilities – the so-called ‘nativist’ approach – Turing believed humans are born with an ‘unorganised machine’: the brain, which is trained to become a universal machine by our lives and experiences. Turing reckoned there wasn’t much difference between the cold logic of the early computers he was tinkering with in the 1940s and the human brain.

He put forward his ideas in a discussion at the University of Manchester, where he was deputy director of the computing laboratory, on 27th October 1949. The others present were Turing’s colleague, Max Newman; a chemist and philosopher called Michael Polanyi; and a zoologist-cum-neurophysiologist, John Zachary Young. Polanyi and Young argued against Turing’s idea that the human mind was programmable like a computer; Turing replied that they had simply not yet thought about it. ‘The mind is only said to be unspecifiable because it has not yet been specified,’ Turing reportedly said, according to contemporary minutes.

Turing didn’t manage to convince either philosopher or neurophysiologist that the human brain and the logic that dictated computers were similar. He did, however, convince himself. He kept plugging away at the problem. He designed a test to ascertain whether a computer could, in theory, ‘think’ the same way a human does. Turing thought determining whether computers could think outright was too tricky, but designed his test to see whether it could imitate the process. He put forward his test in an academic paper in the Mind journal titled ‘Computing machinery and intelligence' in 1950. The ‘Turing test’ was a version of the Imitation Game, a classic parlour game in which a player would hide their gender while answering questions from another, who would try to guess whether they were man or woman from their answers.

Turing’s test asked a simple question: could a computer talk like a human? The idea was that a human judge would take part in a text-based conversation, mediated through computers, with different players. The judge would look at the responses to their questions, and decide whether they sounded human, even if they weren’t. If the judge considered a computer indistinguishable from human participants, it would be considered ‘intelligent’.

Turing believed that was the benchmark for intelligent behaviour from a computer – and he believed that while it wasn’t likely to be possible in 1950, it would be by the year 2000. At that point, he reasoned, computers would have 100 megabytes of memory, which he thought would be enough to pass the test:

An average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning.… I believe that at the end of the century the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking without expecting to be contradicted.

Sure enough, Turing was right about computer memory, also called RAM, or random access memory. An Apple PowerBook G4 released in 2001 had 128 megabytes of RAM as standard. Turing was wrong, however, in his confident prediction that a computer could pass the Turing test with that level of memory – at present, no artificial intelligence system has managed to convince someone they’re human, despite their computational power now far outstripping the 100 megabytes he calculated would be sufficient. (They have got closer, however.) But Turing wasn’t the only computer scientist pushing the boundaries of the discpline.

...

John McCarthy was the US-born child of pro-communist Irish and Lithuanian immigrant journalists. Young John’s childhood was steeped equally in radicalism and science, encouraging him to question things frequently without worry, and to embrace technological innovation without fear.

‘There was a general confidence in technology as being simply good for humanity,’ he recalled. A precocious child, his parents gave him a Soviet technology book published in 1933 called 100,000 Whys: A Trip Around the Room, the English translation of which Nature called a ‘small guide to general knowledge [which] will serve in lessening to a slight degree the load of ignorance which so many carry’. (McCarthy read the book in its native Russian; his parents' Communist zeal meant that Russian was spoken frequently in their home, and John junior was fluent.)

100,000 Whys helped stoke in young John McCarthy a passion for technology that continued through his teenage years. Like Alan Turing, McCarthy showed a huge aptitude for mathematics. Unlike Turing, whose interest was downplayed and diminished by his teachers, McCarthy was given the chance to foment that interest. 

While at high school in California, McCarthy managed to get his hands on the recommended reading list for maths degrees at the California Institute of Technology (Caltech). He then bought the books and taught himself university-level mathematics – which would come in handy when he applied to Caltech in 1944. He graduated high school two years early, despite starting his schooltime education a year later than his peers because of sickness. A life lived on fast-forward continued at Caltech: he was allowed to skip the first two years of study when the admissions tutors saw what he could do.

But it wasn’t all plain sailing at the university. McCarthy was booted out of Caltech because he refused to attend physical education classes. He had a short stint in the army before returning to university and graduating with a bachelor’s degree in mathematics in 1948.

In September that same year, at the same time as Alan Turing was eagerly sketching out his idea for Turochamp, McCarthy, then 21, attended a talk by the celebrated Hungarian-American mathematician and computer scientist John von Neumann. At the time McCarthy, who would become a central figure in the development of artificial intelligence, was back at Caltech taking a one-year master’s programme.

Caltech was where von Neumann delivered the Hixon Symposium on Cerebral Mechanisms in Behavior. His subject was self-replicating automata: hypothetical machines that could manufacture copies of...