AI and literature: the muse in the machine

Image / FT

To generalise heroically, the entire arc of world literature can be summarised as follows: writing about gods, then kings and queens, then ordinary people and, finally, ourselves.
But having exhaustively explored the divine and almost every dimension of the human, we are plunging deeper into a new era of literary history: writing about machines. And that may even presage the most startling evolution of all: machines writing about humans and perhaps, one day, machines writing about machines.

Literature, sometimes described as crystallised emotion, contains the very essence of the human experience. So what will it mean when machine intelligence trespasses into human territory? Will it open up exciting new vistas of understanding and insight? Or will it only highlight our own diminishment in the cosmic scheme of things?

Almost every day brings a startling new advance in the field of artificial intelligence (AI), the seemingly magical general-purpose technology of our times. In 1997, the world was stunned when IBM's Deep Blue computer beat the greatest chess player of all time, Garry Kasparov. Although impressive in its way, Deep Blue was little more than an immensely powerful rules-based calculating machine, a "$10m alarm clock", in Kasparov's irritable phrase.

But more recent advances in Deep Learning techniques, combined with an explosion of data from our smartphones and computers and massive increases in computing power, have enabled machine-learning programs to perform an increasing array of tasks as well as (if not better than) any human: interpreting radiology scans, flying aircraft, identifying images and recognising speech (just ask Siri).

Google DeepMind's defeat of two of the best players of the ancient — and fiendishly complex — Chinese game of Go in 2016 and 2017 also captivated a global audience. Using a very different technique from Deep Blue, AlphaGo's success stemmed from "learning" by itself, defying 2,500 years of accepted wisdom about the game. Little wonder that some Chinese researchers called this a "Sputnik moment" for the country, stimulating a massive increase in spending on AI amid talk of a new technological arms race.

All this buzz about AI has also sparked the imagination of some of our most inventive novelists, among them Jeanette Winterson and Ian McEwan. Once the exclusive preserve of science fiction, thinking robots have now entered the literary mainstream. McEwan's most recent muse is none other than Demis Hassabis, the luminous founder of Google DeepMind.

This latest fascination with AI is that it reframes the eternal debate about what it means to be human, challenging our conceptions of identity, creativity and consciousness. AI is both the apotheosis of the rational scientific thought of the Enlightenment and, possibly, its damnation too. Are we approaching the day, the Israeli historian Yuval Noah Harari has asked, when computers will know us better than we know ourselves?

IJ Good, the mathematician who worked alongside Alan Turing in the code-breaking team at Bletchley Park during the second world war and was a pioneer of AI, was one of the first to grasp the full significance of an "intelligence explosion".

Once ultra-intelligent machines surpassed the intellectual level of humans, he wrote, then they themselves would be able to invent even better machines, leaving us far behind. "Thus the first ultra-intelligent machine is the last invention that man need ever make, provided that the machine is docile enough to tell us how to keep it under control," he wrote in 1965.

That moment of technological singularity, should it ever happen, has been compared with the creation of a new life form, elevating Homo sapiens to sit alongside God. Unarguably, it would be the most significant event in 300,000 years of human history. According to the scientific writer James Lovelock, it would also signify the end of the 300-year Anthropocene age, during which Homo sapiens has wielded planet-changing technologies, and the arrival of the Novacene, the age of hyper-intelligent machines.

In this view, we will be swept away in an avalanche of algorithms sliding towards a post-human future in which we are likely to have as little understanding of electronic lifeforms as polar bears have of snowploughs.

Even before the age of AI, humans have been fascinated by artificial creations that come to life. In ancient Greek mythology, Pygmalion created Galatea, a beautiful ivory statue that was animated by Aphrodite and became his wife. In Jewish folklore, the golem was an amorphous lump of clay that was brought to life by humans.

Czech playwright Karel Čapek wrote about a factory that manufactured artificial workers in R.U.R., popularising the word "robot" after his play was staged in America in 1922.
And, perhaps most famously of all, Mary Shelley wrote a Gothic horror in 1816 about how Victor Frankenstein, a brilliant young scientist, created a hideous, out-of-control monster from body parts discarded from "the dissecting room and the slaughterhouse".

It was this literary masterpiece that inspired Winterson to write her latest novel Frankissstein, interweaving Shelley's experiences with a love story set in Brexit-era Britain in which a young transgender doctor called Ry (wryly described as "future-early") falls for a mysterious Professor Victor Stein, who is, naturally, an AI expert. "This story is an invention that sits inside another invention — reality itself," Winterson writes.

In an interview with the BBC, the author described Shelley's original tale as a message in a bottle that can only now be fully understood, 200 years after it was written. Just as the 19th-century mathematician Ada Lovelace wrote code for a computer that had not yet been built, Shelley was anticipating a world that had not yet been created. "These women were really looking past the present, jumping forward into a future that did not yet exist," she said.

For Winterson, who has movingly explored the nature of her own individuality and sexuality in her memoir Why Be Happy When You Could Be Normal?, the dawn of the AI age is something to be welcomed, even cherished.

The first ultra-intelligent machine is the last invention that man need ever make
"The great thing about AI, if we ever do it, is how can you worry about gender, whether you love a man if you are a man, whether you love a woman if you are a woman, when we are about to share the planet with self-created, non-biological life forms which will not have a gender? This could change everything. I love this," she told the BBC. But many other writers, including McEwan, have a far more ambivalent, and darker, view of the emergence of humanoid robots, the visible embodiment of AI. In his novel Machines Like Me, McEwan tells the story of Charlie, a happy-go-lucky drifter who buys one of the first synthetic humans.

Adam, as he is tellingly named, was "advertised as a companion, an intellect­ual sparring partner, friend and factotum who could wash dishes, make beds and 'think'." He was also capable of sex and possessed functional mucus membranes in the maintenance of which he consumed half a litre of water each day. "Before us sat the ultimate plaything, the dream of ages, the triumph of humanism — or its angel of death. Exciting beyond measure, but frustrating too."

Although it is not supposed to happen, according to the instruction manual, Adam proceeds to fall "in love" with Charlie's girlfriend Miranda and even starts writing her poems, some 2,000 haikus in all. Or does Adam actually fall in love? Can a robot ever become truly conscious and experience and express emotions? Cue a couple of thousand years of philosophical disputation on the nature of consciousness.

Charlie, though, appears to be of the "quacking duck" school of philosophy. If it looks like a duck, walks like a duck and quacks like a duck, then it probably is a duck. Or, as Charlie puts it: "If he [Adam] looks and sounds and behaves like a person, then as far as I'm concerned, that's what he is."

For McEwan, the onward march of science spells the retreat of human self-regard, a "series of demotions leading to extinction". Once, humans were at the centre of the universe with the Sun and planets "turning around us in an ageless dance of worship." Then "heartless astronomy" reduced us to an orbiting planet around the Sun, just one among many other rocks.

Humans, though, still clung to the conceit that they remained brilliantly unique, "appointed by the creator to be lords of everything that lived". But then soulless biology confirmed that we are little different from other creatures, "sharing common ancestry with bacteria, pansies, trout and sheep".

Our last redoubt was consciousness. But even here we are likely to prove hapless accomplices in our own self-abnegation. "The mind that had once rebelled against the gods was about to dethrone itself by way of its own fabulous reach. In the compressed version, we would devise a machine a little cleverer than ourselves, then set that machine to invent another that lay beyond our comprehension. What need then of us?" Hear the echoes from IJ Good's intelligence explosion.

Many AI experts reading these novels would surely say that the authors are letting their imaginations run way ahead of themselves. The success of Google DeepMind in creating neural networks to defeat one of the strongest Go players in history may be a remarkable achievement. But that does not herald the imminent arrival of an omnipotent new life form. Machine learning programs may be becoming exceptionally good in many narrow domains, but they are still comically weak in terms of general intelligence. After all, modern robots, rather like the machine-hybrid Daleks in the TV series Dr Who, still find it hard to climb the stairs.

Alison Gopnik, a developmental psychologist at the University of California, Berkeley and an expert in the "theory of mind", argues that natural stupidity will continue to pose a far greater danger to humanity than AI for the foreseeable future. "There is not much basis for either the apocalyptic or the utopian vision of AIs replacing humans. Until we solve the basic paradox of learning, the best artificial intelligences will be unable to compete with the average human four-year-old," she writes in Possible Minds, a collection of essays on AI by many of the world's leading experts.

Still, machine learning is continuing to make great strides in many narrow areas, including writing. Some AI programs are already writing news reports for major US agencies. Bloomberg News says that almost one-third of its content is written with the help of robot reporters. The system it uses, known as Cyborg, can pull the essential facts out of routine financial statements and write basic news stories in double-quick time.
Sometime before our Sun burns itself out, it seems certain that electronic intelligence will supersede the human kind

The irony is that many of these computational news reports are also read by robots linked to automated financial market trading programs. Important information is already being extracted, analysed, reported and read with almost no human intervention. We are already beginning to write ourselves out of the script.

But the severe limitations of literary AI have also been highlighted by an intriguing project in 2017 conducted by Ross Goodwin, who describes himself as a creative technologist and gonzo data scientist. In an attempt to emulate the great road trip novel of Jack Kerouac, Goodwin drove from New York to New Orleans collecting data from a camera, a microphone, a computer clock and GPS. Those data were then fed into a neural network that had machine-read hundreds of books to spot literary patterns. Out spewed a real-time novel, 1 the Road.

The novel's opening sentence certainly has some resonance: "It was nine seventeen in the morning, and the house was heavy." But the overall result, according to Brian Merchant, the Atlantic magazine writer who followed the experiment, was like "[Tom Wolfe's] Electric Kool-Aid Acid Test meets Google Street View, narrated by Siri". Kerouac it was not.
No matter how much we laugh at robotic failings today, we should still marvel at how fast AI has developed over the past decade and wonder how far it may yet evolve. We flatter ourselves that electronic intelligence will always take shape in humanoid robots, such as McEwan's Adam. But it is far more likely to develop in disembodied forms that we can scarcely comprehend.

Few scientific writers have a longer perspective than Lovelock, author of Novacene, who has just celebrated his 100th birthday. The originator of the Gaia theory, which argues that the Earth is a single system of living organisms and inorganic surroundings, Lovelock believes it is only a matter of time before we hand the "gift of knowing" on to new forms of intelligent beings.

At 4.5bn years of age, Earth is estimated to be halfway through its existence. At some point before our Sun burns itself out, it seems certain that electronic intelligence will supersede the human kind, he argues. If we are lucky, humans might provide entertainment for — what he calls — the cyborgs, "just as flowers and pets delight us" today.

But Lovelock urges us not to be depressed. We have played our own astonishing role in the history of the cosmos and have enjoyed our time in the sun. "Just as we do not mourn the passing of our ancestor species, neither, I imagine, will cyborgs be grief-stricken by the passing of humans."

If we are to take any consolation, he suggests, it is to come from the words of Alfred Tennyson's poem on Ulysses, as the great warrior and explorer slipped into old age.
"Tho' much is taken, much abides; and tho' / We are not now that strength which in old days / Moved earth and heaven, that which we are, we are . . . "

Maybe cyborgs will still write novels for fun about humans. Or, if they see the need, they may even write books about the internal lives of machines. The only trouble is that we humans would not be able to read them, still less understand.

Copyright The Financial Times Limited 2019