Ian Brown is a Globe and Mail feature writer.
In April I wrote a long story about humanity’s new partner, artificial intelligence. The experience almost crushed me. I’m still trying to figure out why.
It was in many respects an ideal assignment: challenging (it’s a complicated subject), interesting (AI’s capabilities are amazing) and populated by smart if sometimes eccentric computer engineers. But my AI queasiness was persistent, an unavoidable weather system advancing on the horizon. Thanks to the new AI, all of humanity seemed to have aged instantly – was older, slower, less capable. We’d been handed a collective reprimand and a future pink slip.
At first I resisted my uneasiness. Don’t be silly, I told myself: There’s no point pushing back against something that already exists and that will be hugely valuable (for medicine, for self-care, for teaching, for countering climate change, for you name it). Adapt. Use it to your advantage. Yes, some of the original architects of AI, including Geoffrey Hinton (formerly of Google), fear the technology will cause the extinction of the human race. But other billionaire innovators, such as Marc Andreessen (co-founder of Netscape, venture capitalist), insist “AI will save the world.” As long as the tech titans are at loggerheads over what AI portends, it’s obviously too soon to tell.
And yet I kept skittering away from the subject, the way a rat runs from the light. A big part of the problem was that writing about AI demanded I spend time examining what machines, rather than human beings, had created. And while there is endless touting of AI’s wizardry – one new form of machine intelligence can spot lung tumours previously invisible to doctors – those brilliant developments aren’t so visible. What we the general public mostly see instead when we search “AI” on Google are the creations of the “generative” machine-learning that drives the likes of ChatGPT and its arty cousins, DALL-E (which makes pictures) and VALL-E (which reproduces voices).
The results were not inspiring. They were sometimes arresting – they’d been created by a machine with almost zero human input, after all – but they were rarely exceptional or admirable, and they never held my attention. AI wrote “essays” and “stories” and jammed unlike things together (the Pope, in a parka) and made politicians do and say things they never did or said, but its creations were depressing in their sameness, in the thinness of their humour and in the adolescent repetitiveness of their parody: Look how real we can make a lie look, heh heh heh.
They felt mediocre. It’s all very well to say “paint a picture in the style of Picasso of two cats driving off a cliff.” But the resulting picture has nothing at stake and therefore no yearning and no genius: The image can be instantly replaced by another one. It feels surreal because it’s surrealism. Of all the periods of painting, surrealism is the most enervating because it is the most random, the least intellected and thought-out, the genre that eschews “meaning” for the alleged value of artistic happenstance.
I’m not trying to be a snob. There is nothing automatically wrong with mediocrity. It means “average” or “without distinction,” from the Latin for “the middle of the mountain,” and it makes the world go round. But there is nothing special about it. And what got me down, evening after evening, was the middling aspiration of AI – the non-existent odds of anything intentionally sublime popping up, the insufficiency of hoped-for excellence or grandeur or beauty to inspire even my mildest awe.
I mean, don’t you always want at least the possibility of being moved to be there when you read a book, look at a painting, watch a TV series, listen to music or even open a newspaper (I know, fat chance)? That you will be snatched out of this world, out of your present moment, into a bigger world and a bigger moment, into a perspective that … puts things into perspective?
Don’t you always secretly (or not so secretly) want to stumble across something such as Nicholson Baker’s four-page paean, in his action-packed little novel The Mezzanine, to the micro-specific pleasures of drinking milk as a child and “the wayside greatness of the milk carton” and how “the small diamond shape of the spout is a perfect fit for the nose”?
Don’t you long to find something as good as – and now I’m just pulling random books off my shelves – the consul’s drunken letter to his ex, Yvonne, in Malcolm Lowry’s Under the Volcano, in which he describes his hellish “nights the color of grey hair” lying awake missing her in Mexico, or Sharon Olds’s description of her father’s stillness after he learns his illness is incurable – how “he sat, motionless, alone, with the dignity of a foreign leader”? Don’t you hope for at least a drop of unexpected originality along the lines of Philip Larkin’s description in his poem To the Sea of the eternal pleasures of walking on the beach? ”The small hushed waves’ repeated fresh collapse/Up the warm yellow sand…/Still going on, all of it, still going on!”
Admittedly those are classics, but that’s the great thing about human-generated classics: They’re reliably moving – unlike the endlessly averaged output of the paraphrasing, plagiarizing input genius ChatGPT. And even if the chatbot does manage to write something as fetching as those glimmers, it isn’t intentional, and therefore it isn’t the same. AI “creates” by combining words according to the statistical likelihood of their having been strung together in the past, which is why its answers to our prompts have that weird (bathtub) ring of distant familiarity, and mostly read as if they were squeezed out of a cliché grinder.
The one thing ChatGPT writes well (at least so far) is transitions between arguments. That makes sense: ChatGPT’s writing is intended to save time, to be productive and smooth and bureaucratically efficient. This is the dream of capitalism, and the motto of homo technicus (and the kind of thing Mr. Andreessen seems to think will save the world). Whereas Baker, Lowry, Olds, Larkin and most other human artists are usually trying to stop time, to be the opposite of efficient, because they want to make readers, watchers and listeners stop and then stall long enough to experience a moment of thought, feeling, recognition, connection.
ChatGPT, in other words, is antithetical to the goals of memorable human art. It wants instead to knit us more inextricably into the system, into the productivity machine. You can hear the invisible co-opting at work in the background, humming away. I think that’s one reason AI gives me a headache. It claims to help people “write” better, but its design undermines the difficult, inefficient, repetitive, time-consuming habits that produce the best writing.
And yet we’re lapping ChatGPT up as fast as some of its inventors cry out for government regulation and others pursue a more “spiritual” AI. An estimated 1.6 billion users visit the ChatGPT website every month.
A few experimentalists have taken the creative possibilities of chatbots seriously. Early in June, Canadian novelist and journalist Stephen Marche released Death of an Author, a murder mystery he wrote with the help of a passel of chatbots.
The task wasn’t easy. The bots couldn’t devise a decent plot, so Mr. Marche had to come up with one. He then fed and refed the plot into the chatbots, using different ones to develop the prose in different ways. Sudowrite was good at filling in details (colours of sweaters, smells in the air); Cohere, a Canadian AI (which in June completed a very successful round of venture capital funding), was more adept at polishing sentences.
To give the final manuscript verve, Mr. Marche trained his writing machines on multiple examples of snappy writing – beauties such as Raymond Chandler’s famous description of a blonde so sexy she could “make a bishop kick a hole in a stained-glass window.” Then he asked the bot to write specific descriptions, such as the smell of fresh coffee. The result: “The smell of coffee was like a fog burning off a field.” Which is … okay. It’s a mixed simile, but it’s not … bad. It’s a notch or even two above mediocre. But ChatGPT did not produce even that line on its own: It needed Mr. Marche, who holds a PhD in English and understands literary structure as only a PhD can.
“It’s very much an act of making, of construction,” he told me not long ago. We were talking on the phone after breakfast. “It’s not easy to do, to do what I did with the thing.” He sounded proud but – I admit this is subjective – slightly bewildered, like a guy who set out to rebuild a Toyota Camry from parts without any instructions, only to discover what a daunting job it is to rebuild even a Toyota Camry. Mr. Marche won’t call himself the co-author of the book because he didn’t write the words, but he doesn’t plan to split the royalties.
“Halfway readable” was Dwight Garner’s recent verdict on Death of an Author in The New York Times. “Clever, for sure, but it left me feeling hollow.”
Some people think ChatGPT will be a boon to genre writers, who produce more formulaic fare. Mr. Andreessen, for instance, believes the “creative arts will enter a golden age, as AI-augmented artists, musicians, writers and filmmakers gain the ability to realize their visions far faster and at greater scale than ever before.”
I worry instead that AI and ChatGPT and its clicking cousins portend the end of what we have long understood and valued as our cracked but sustaining humanity. AI can do tons of stuff humans can do, and do it better and more efficiently, but there isn’t any mystery as to how it does them. And the value of that mystery is always undersold.
We know why the chatbots combine words the way they do: we tell them to. But why did Flaubert write in the way he did about his chosen subject, a bored bourgeois woman, and why did he choose that subject? For all our technological sophistication, we still don’t know what sparked his human creative impulse. And that small but persistent mystery gives us hope that fresh human genius may turn up again just as unexpectedly.
We almost don’t want to know, most of the time, how transcendent beauty is made, or where grace comes from, because our ignorance lets us believe we’ll find it in unpredicted places, maybe even in ourselves. If that beauty is just a mathematical probability that can be chunked out faster by an artificial intelligence, we have less reason to believe in our (haphazard) human capabilities, and less reason to want a human-run world – hence the demoralizing miasma, the depressing forest-fire-like fog AI has already cast across the human future.
Of course, AI is just the latest technological revolution (it joins the printing press, the car, the machine gun, the television, the nuclear bomb, the computer and the smartphone) to do what technological revolutions always do to human confidence: They shatter our conception of who we are.
Meghan O’Gieblyn, in her mind-cracking 2021 book, God, Human, Animal, Machine: Technology, Metaphor, and the Search for Meaning, describes the standoff well. “The true trauma of [the] disenchantment,” she writes, “is that the world, as seen through the lens of modern science, is devoid of intrinsic meaning.” We humans want our world to mean something, because we want to matter, however tangentially, while we are here in the now.
The new AI bots imply (falsely) that our meaning can be reproduced at the click of a button. In this way AI trolls all human endeavour. Behind the fear of what it will do to the world we live and work in is a bigger alarm, because gizmos such as ChatGPT devalue all the creation and beauty we ever made and believed in, that ever gave us hope, that ever soothed us. That flaming ball of human glory, the computer masters insist, was really just inefficient combinatorial statistics.
Let me put this another way. The other night, my wife and I invited a few folks over for my son Walker’s 27th birthday. Walker can’t celebrate on his own: He was born with a rare genetic defect, and because of the profound intellectual disabilities it causes, he can’t talk, or swallow by mouth, or walk very fast, or stay focused on one thing for long, or live without 24-hour care. The small group we invite to these occasional parties tend to be people who watched him grow up, who witnessed the crushing, unpredictable, terrifying but oddly graceful touch-and-go nature of his becoming. (We never knew if he’d make 27, for starters.)
It wasn’t a wild party: just some pals, sitting and chatting in the backyard. Most of the guests hadn’t seen Walker in three years because of the pandemic. None of the usual rivalries and antic comings and goings and momentary fractures and jostlings that characterize a good party were in evidence. This was largely because of Walker, who did what he always does: visited, and turned away, and visited, and turned away, and made strange noises, and laughed, and sometimes seemed out of it, and sometimes seemed as happy as he has ever been.
As often happens, his strangeness and his imperfections – his unavoidable flaws, his apparent but unsatisfiable and therefore sad yearnings, his unapologetic unpredictability and the dissolving effects he can have on others – made us recalibrate our definition of a successful human encounter. He softened everyone and lowered our ridiculous standards, but deepened our connections. He joined us, wanting to be with us, but we also joined him, on his terms, as we managed to share what I can only describe as the forgotten beauty of his slowness.
He reminded us of the contingency of every life and all privilege, that his disabilities were different only in kind and degree from our own, and that none of us have much control over what happens. This in turn made us more forgiving and less judgmental. That is my theory, anyway. My point is, a few hours in his disabled presence felt like the opposite of all-powerful AI. And it felt a lot better.
I think what I am trying to say is this: The average North American now spends seven hours a day in front of a screen, abetting one or another giant of surveillance capitalism. The distracted itching and scratching of that online existence supply AI and ChatGPT with the hosepipe of data that enable them to do what they do. Our incessant distractedness in turn makes it easy to forget how much we need non-mediocre art created by actual, living, non-perfect human beings, who are capable of identifying and describing and sharing what we feel, given that we are excitable mortal beings who will one day die.
Perhaps I should underline this point: AI doesn’t die. It isn’t mortal, and so it isn’t close to human. AI can’t feel the tightening grip of oncoming death, can’t hear or feel the whoosh of time rushing by, can’t sense the always present spectre of our sometimes sudden but always inevitable end.
AI has no such humility. So nothing it creates can be as urgent, as daring, as existential as human creation. Furthermore, while AI might be able to create a description of what it feels like to know one’s time is evaporating, it does not and cannot share the feeling behind its description. Whatever AI describes, however much it might move us, we know at bottom the feeling is not shared and is therefore inauthentic, is finally, frankly, faked.
Do you want to live in a world you know is faked? Probably not. We need to know we can share our short and speedy rocket ride through life and consciousness with each other, through the brief blink of life that separates the unknowable darknesses at either end of it. You can’t do that with a machine, not even if you pretend, not even if it makes you very, very rich.
But we go online anyway, and embrace contraptions such as AI, and forget about the irreplaceable sensation of being a living being – until we look up from the feeds and reels and VR goggles and chatbots and wonder where all the precious time that tech was allegedly saving us has leaked away to. Whereupon we need human beings making human art to remind us that everyone makes the same mistakes, that regret and remorse and longing are part of our common felt inheritance.
Which brings me to the antidote I eventually devised to feel better about AI’s pushy takeover of our human dimension. I offer it to you freely, esteemed reader, my old pal and beloved companion, if indeed you are still with me.
Here it is: I simply went about the rest of my physical, non-screen life. It turns out that, at least compared with the average output of ChatGPT, my daily physical life is a thing of timeless genius.
I walked my dog for an entire afternoon in a steady rainstorm and stayed completely dry and therefore marvelled at the brilliant effectiveness of contemporary outerwear.
I had lunch with my daughter, which is always thrilling. We talked about money and books and boyfriends (which is what I do to avoid the fact that my time with her is inevitably shrinking; if my luck is average, I have about 750 weeks left).
I ate unexpectedly fantastic dim sum, the best I’ve had in a long time, in a desolate white plastic room on the hard-to-find top floor of a bleak shopping mall in Toronto’s downtown Chinatown with a smart guy who talked like a friendly machine gun.
My wife and I spent a weekend in the country at the luxuriously redone farmhouse of some friends, a property that made me so envious my chest hurt. It was helpful to remember what envy felt like, how it hangs in the forehead and just below the solar plexus, and helpful to recall that you don’t have to do anything about it, because it eventually fades away.
I went to a public lecture given by a moralizing Catholic priest and was surprised by his incapacity for forgiveness; hosted a luncheon of lucky people who had all managed to read books for a living; saw Suzie Raudaschl and her band, Bridal Party, sing at a downtown Toronto club called, get this, Baby G, where most of the people in the audience were at least 40 years younger than me and were wearing baggy vintage garb that made them look like peasants in a Brueghel painting.
I saw my super-gregarious ophthalmologist, Robert Wagman, one last time before he retired (which was unexpectedly moving); my dentist installed another crown in my mouth (unexpectedly expensive).
Spring came, sort of, and I stood under the weeping Japanese cherry in my backyard on a dangerously frosty morning and looked up through its stiff and blooming branches at the pinkened sky. (The cold snap actually preserved the blossoms for weeks longer than they would have lasted otherwise.)
This isn’t a complete list, of course. None of those everyday human experiences were especially purposeful, productive or efficient; they did not save time and were not notably intelligent. Still, they reminded me not to grant human status to computers. They don’t deserve it yet, and I suspect they never will.