ChatGPT is a Fur Coat
Should you wear it?
I’m Tim Gorichanaz, and this is Ports, a newsletter about design and ethics. You’ll find this week’s article below, followed by Ports of Call, links to things I’ve been reading and pondering this week.
User interfaces are disguises. They get us to focus on what actions are possible—booking a ride, sending a message—while hiding from us how the system works. And they hide what’s going on so well that it doesn’t even occur to us as a possible question.
As the inner workings of a system get more complex, what the interface hides gets, by definition, thornier and thornier. AI systems are thorniest of all. The article “Anatomy of an AI System” shows this well, offering a map of the immense machinations that occur every time you give Alexa a simple command. In a similar vein, the book Ghost Work, by Microsoft researchers Mary Gray and Siddarth Suri, documents the unseen networks of human labor that power the “autonomous” side of the internet.
What ChatGPT Hides
Like everyone else, I've been thinking a lot about ChatGPT lately. The user interface couldn’t get much simpler: a text field and a Send button. But what this interface hides from us is more complex than perhaps any other system humans have devised. First and foremost, a GPT model is a huge neural network whose connections we have barely begun to map.
But beyond that, ChatGPT’s interface obscures—if you’ll permit me a bit of hyperbole—the theft and slavery that goes into creating a large language model. Theft, because large language models require troves of input data—45 gigabytes of text for GPT-4—and this data doesn’t come from nowhere. It was written by people over the ages and is being used without notice, consent or compensation. This issue is at the root of a growing pile of lawsuits over generative AI and an emerging shift in tech company policy toward charging for API access.
And slavery, because the method of training used in large language models requires human input along the way. In the case of OpenAI’s GPT models, the company has employed Kenyan crowd workers paid less than $2 per day. Could OpenAI be worth $29 billion today if it weren’t for an underclass living half a world away?
Given the harms inherent in ChatGPT, should we use the product? On one hand, using ChatGPT would seem to endorse theft and slavery. On the other hand, you might say that the theft and slavery have already occurred, so we might as well make use of the results so that it won’t have been for naught.
There’s an analogy we could make to fur coats. People wore them in ages past as a matter of survival. The fur came from animals who also provided food, fuel and more. But when fur coats became an industrial operations, they became less defensible. A fur coat may be beautiful and warm, but it can only be had through harm. And wearing the coat, even though the animal's death can no longer be prevented, is in a way endorsing the system that produced it. Still, apparently 100 million animals are killed each year for their fur.
Designs of War
The Japanese animator Hayao Miyazaki—of Spirited Away and My Neighbor Totoro fame—has struggled with similar questions when it comes to airplanes. He always loved drawing airplanes—military fighter planes, to be precise. As a person born at the start of World War II, he also knew firsthand of the destruction they wrought. But—so beautiful, so fun to draw.
In Miyazaki’s most recent film, The Wind Rises, warplanes take center stage. Indeed Miyazaki was criticized in Japan when the film was released for the film’s seeming glorification of war. The claims are not unfounded, and yet Miyazaki is an outspoken pacifist.
As Susan Napier writes in Miyazakiworld, the animator is not unaware of the irony. It’s even made explicit in The Wind Rises in an exchange between Caproni, a famous airplane designer, and the protagonist Jiro, an aspiring airplane designer.
Caproni asks Jiro, “Which would you rather live in? A world with or without pyramids?”
“What do you mean?” Jiro says.
“Humanity has always dreamt of flying,” Caproni says, “but the dream is cursed. My aircraft are destined to become tools for slaughter and destruction.”
“I know.”
“But still, I choose a world with pyramids in it. Which world will you choose?”
Jiro says, “I just want to create beautiful airplanes.”
Napier explains the significance of this exchange:
Caproni’s implication is that the technological and aesthetic achievement of the pyramids was achieved only through hideous human suffering and sacrifice. Overall, the film seems to be arguing for the pyramids, no matter what cost, a thorny message that may be hard for some audiences to accept.
It may be hard for us to accept, yet it’s a question that we answer through our actions all the time. Our revealed preference is for a world with pyramids.
The philosopher Vilém Flusser wrote an enigmatic little collection of essays called The Shape of Things: A Philosophy of Design. In one of the essays, he points that all design ultimately has its roots violence and warfare.
The first human creations were about hunting animals and fighting over territory. Knives and arrows made that more effective and efficient:
If in their day our ancestors in East Africa 100,000 years ago had not designed arrow-heads that were at the same time elegant, user-friendly and good (and that could therefore kill with elegant convenience), then we would probably still be laying into each other or into animals with our teeth and nails. …
There are, however, people who are against war. They are not willing to be killed by rockets (although, when asked, they cannot say what kind of death they would prefer). Such people are prepared, in the interest of peace, to accept bad design. They are downright pleased if rockets, paper-knives and arrow-heads get worse and worse and thus become less and less elegant, less and less convenient. They are good people in a totally different sense of ‘goodness’ from the one intended. These good people are good for nothing but for simply existing. They are anti-designers.
What Flusser seems to be saying is that design is a toolkit, a set of processes, for making things more usable and efficient, and that this toolkit enables both designs for good and those that cause harm. You can’t have just one, because in the end they are inseparable. “Where the danger is, there also grows the saving power,” wrote the German poet Hölderlin. Think of the way nuclear technology at once enables nuclear power (which may be the only way out of our climate crisis) as well as nuclear weaponry.
“There may be a way out of this dilemma,” Flusser writes. “Either war and an elegant, user-friendly life in the midst of good objects, or everlasting peace and a squalid, inconvenient life in the midst of badly functioning objects.”
Well, it’s not much of a choice, is it? That’s why we choose the pyramids and fur coats.
Ports of Call
Debates: Hear Me Out is a podcast from Slate that hosts some of the most fascinating conversations I’ve heard lately. Guests make the case for positions that may at first seem overreaching or ludicrous—why childbirth should be free, why the U.S. invasion of Iraq was good, why some U.S. states should secede. You may not be convinced by the end of the episode, but it will certainly help you understand your own position more deeply.
A Coronation: Not being British or descended from any British subjects (my ancestors crawled over from Eastern Europe in the early 20th century), I don’t have intimate feelings one way or another about the monarchy. But King Charles III’s coronation is tomorrow, and I’ll watch it online, if for no other reason than that it constitutes a 1,000-year tradition and combines ritual with mystical belief—how could you not be fascinated?
AI-Generated Commercials: You may have seen the 30-second beer ad that’s been making the rounds on social media. You can watch it here, as well as one for a pizza place. For some, further proof of our impending doom. For others, proof that all of this is overblown.