1818 novel by Mary Shelley Frankenstein – the urtext of science fiction – is about creating artificial life. And Fritz Lang’s founding film of 1927 Metropolis established an astonishing number of fantasy horror tropes with its Maschinenmensch – the “human-machine” robot that sows murderous chaos.

The Turing machine and the Turing test

However, the creation of AI remained firmly in the realm of science fiction until the advent of the first digital computers shortly after the end of World War II. At the center of this story is Alan Turingthe brilliant British mathematician best known for his work deciphering Nazi ciphers at Bletchley Park. Although his decryption work was vital to the Allied war effort, Turing deserves to be at least as well known for his work in the development of computers and AI.

While studying for his doctorate in the 1930s, he devised a mathematical device now known as the “Turing machine”, providing a model for computers that is still standard today. In 1948 Turing took a job at the University of Manchester to work on Britain’s first computer, the so-called “Manchester baby”. The advent of computers sparked a wave of curiosity about these “electronic brains”, which seemed capable of dazzling intellectual prowess.

Alan Turing deserves to be at least as well known for his work on the development of computers and AI

Turing apparently became frustrated with dogmatic arguments that intelligent machines were impossible and, in a 1950 journal article MIND, sought to settle the debate. He proposed a method – which he called the Imitation Game but is now known as the Turing Test – for detecting a machine’s ability to display intelligence. A human interrogator engages in conversations with another person and a machine – but the dialogue is conducted through a teleprinter, so the interrogator does not know who is who. Turing argued that if a machine could not be reliably distinguished from a person by such a test, that machine should be considered intelligent.

At the same time, on the other side of the Atlantic, the American academic John McCarthy had become interested in the possibility of intelligent machines. In 1955, while seeking funding for a scientific conference the following year, he coined the term ‘artificial intelligence’.

McCarthy had high expectations for his event: he believed that after bringing together researchers with relevant interests, AI would be developed in just a few weeks. In fact, they made little headway at the conference – but McCarthy’s delegates gave birth to a new field, and an unbroken thread connects these scientists through their academic descendants to today’s AI. .

Development of “neural network” technologies

In the late 1950s, only a handful of digital computers existed in the world. Even so, McCarthy and his colleagues had by then built computer programs capable of learning, solving problems, solving logic puzzles, and playing games. They assumed that progress would continue to be rapid, particularly as computers were rapidly becoming faster and cheaper.

More like this

But the momentum faltered, and by the 1970s research funding agencies had grown frustrated with overly optimistic forecasts of progress. Cuts followed and the AI ​​gained a bad reputation. A new wave of ideas sparked a decade of excitement in the 1980s but, once again, progress stalled – and, once again, AI researchers were accused of overstating expectations for breakthroughs.

Things really started to change in this century with the development of a new class of “deep learning” AI systems based on “neural network” technology – a very old idea itself. . The brain and nervous system of animals comprise a considerable number of cells called neurons, connected to each other in vast networks: the human brain, for example, contains tens of billions of neurons, each possessing on average of 7,000 connections. Each neuron recognizes simple patterns in the data received through its network connections, prompting it to communicate with its neighbors via electrochemical signals.

The human brain contains tens of billions of neurons, each of which has, on average, around 7,000 connections

Human intelligence is born in some way from these interactions. In the 1940s, American researchers Warren McCulloch and Walter Pitts were struck by the idea that electrical circuits could simulate such systems – and the field of neural networks was born. Although they have been studied continuously since McCulloch and Pitts’ proposal, it has taken new scientific advances to make neural networks a practical reality.

In particular, the scientists had to figure out how to “train” or configure the networks. The necessary breakthroughs were made by British-born researcher Geoffrey Hinton and his colleagues in the 1980s. This work sparked a short-lived wave of interest in the field, but it died out when it became clear that the computer technology of the time was not powerful enough to build useful neural networks.

“Deep learning” AI systems

At the dawn of the new century, this situation has changed: we now live in an age of cheap and abundant computing power and data, both of which are essential to building the deep learning networks that underpin tend recent advances in AI.

Neural networks are the core technology that underpins ChatGPT, the AI ​​program launched by OpenAI in November 2022. ChatGPT – whose neural networks comprise around a trillion components each – immediately went viral and is now in use by hundreds of millions of people every day. Part of its success can be attributed to the fact that it looks exactly like the type of AI we’ve seen in the movies. Using ChatGPT simply involves having a conversation with something that seems both knowledgeable and intelligent.

What its neural networks do, however, is pretty basic. When you type something, ChatGPT just tries to predict what text should appear next. To do this, it was trained using vast amounts of data (including all text published on the World Wide Web). Somehow, those huge neural networks and data allow it to deliver some extraordinarily impressive answers – for all intents and purposes, passing the Turing test.

The nightmare of losing control

The success of ChatGPT has brought to the fore an overriding fear: that we might bring something to life and then lose control of it. It’s the nightmare of Frankenstein, Metropolis And The Terminator. With ChatGPT’s bewildering capability, you might believe such scenarios could be at your fingertips. However, although ChatGPT is remarkable, we shouldn’t attribute too much real intelligence to it. It’s not really a spirit – it’s just trying to suggest what text might appear next.

The success of ChatGPT brought to the fore an overriding fear: that of bringing something to life and then losing control.

He doesn’t wonder why you ask him about curry recipes or Liverpool Football Club’s performance – in fact, he doesn’t wonder anything. It has no beliefs or desires, nor any purpose other than to predict words. ChatGPT will not come out of the computer and take over.

That doesn’t mean, of course, that there aren’t potential dangers in AI. One of the most immediate is that ChatGPT or the like can be used to generate industry-scale disinformation to influence upcoming US and UK elections. We also don’t know to what extent these systems acquire the myriad human biases that we all display and are likely evident in their training data. The program, after all, does its best to predict what we’ll write – so widespread adoption of this technology can essentially serve to mirror our biases. We may not like what we see.

Michael Wooldridge is professor of computer science at Oxford University and author of The Road to Conscious Machines: The Story of AI (Pelican, 2020)

This article first appeared in the August 2023 issue of BBC History Magazine