Unpredictable Futures: Bonhoeffer and Bots

With Poets Training Generative AI, It’s Worth Asking: What Makes Us Human?

Everyone was waiting for just one thing. World War II had interrupted the work, and now Dietrich Bonhoeffer found himself behind bars. It was 1943. The German theologian started his magnum opus, Ethics, in 1940. Busy assisting Jews and shirking as much official work as possible, however, Bonhoeffer found little time to write.

But here in Tegel Prison things changed. For a year and half he lived, as Devin Maddox and Taylor Worley say in a wonderful article for Christianity Today, “in relative comfort, allowing him the time and space to read and write prolifically for most of his imprisonment.”

So, he jumped back into Ethics, right, the partially completed, urgently needed book everyone anxiously awaited?

Dietrich Bonhoeffer

No.

As Maddox and Worley detail, along with piles of letters, Bonhoeffer wrote poetry, worked on a play, and started a novel. Given his passive and active resistance against the Nazi juggernaut, including joining an unsuccessful plot to assassinate Hitler, this might strike us as curious, maybe even disappointing.

“Bonhoeffer spent his final months,” they say, “creating art.”

Wasted Effort?

The reason some might judge this pursuit frivolous comes from our tendency to reduce people to what we deem their primary purpose. If you’re an activist theologian working against a maniacal and murderous regime, we expect you to keep up the routine—not drop out and start telling stories, let alone counting iambs.

Activist theologians are on the clock, after all. And as soon as the authorities could nail Bonhoeffer for his involvement in the assassination plot, time ran out. His enforced writer’s retreat was cut short, and he was shunted off to more forbidding confines, including Buchenwald and eventually Flossenbürg.

The Nazis hanged Bonhoeffer there just before the end of the war in April 1945. And that book everyone was waiting for? He never finished it. Nor, it’s worth saying, did he finish the novel. So, was it all a waste? We did get some penetrating poetry. Here’s a bit from one of those prison poems, “Success and Failure”:

Success is full of foreboding,
failure has its sweetness.
Without distinction they appear to come,
the one or the other,
from the unknown.
Both are proud and terrible.

People come from far and wide,
walk by and look,
pausing to stare,
half envious, half afraid,
at the outrage,
where the supernatural,
blessing and cursing at the same time,
entangling and disentangling,
sets forth the drama of human life.
What is success and what is failure?

Time alone distinguishes.

An apt thought for a man in prison for virtues misconstrued by an unjust state as crimes. But it’s the motivation behind the lines that matters most of all. Why write poetry when more productive pursuits await?

“What appears to us as such a curious choice must have been a desperate attempt to hang on to his own humanity,” write Maddox and Worley. “In the midst of threatening death, creativity allowed him some recovery of his most authentic self. . . . Creative expression is our life-giving cry of freedom.”

The Consolation of Art

Bonhoeffer’s story reminds me of another writer and occasional theologian imprisoned for supposed involvement in a coup. Boethius was partway through translating Aristotle’s corpus when he was accused of trying to undermine the sixth-century Ostrogothic king Theodoric the Great.

Behind bars and awaiting his own execution, did Boethius resume his translation efforts? It might have changed the course of Western history if he had.

Boethius behind bars

As it stands, it took the Arabic translation movement and half a millennium of additional time for the Latin West to access the Greek philosopher because Boethius wrote The Consolation of Philosophy instead, interspersing prose and poetry to illumine life’s reversals and why our favored paths to happiness reliably disappoint.

As with Bonhoeffer, art trumped productivity. Scholasticism could wait.

Maddox and Worley provide a theological rationale for this tradeoff. As humans, we bear the image of a God who creates. Creative engagement allows us deep and needful access to what makes us human; in creating art we are returning to form.

But one needn’t hold Christian convictions to recognize art as fundamentally generative and affirming of our humanity. And what makes us human is perhaps more relevant now than ever.

Do Robots Write Poetry?

I’m unsure if we ever satisfactorily answered whether androids dream of electric sheep, but let’s complicate it with another puzzler: Do robots write poetry? As it happens, Bonhoeffer and Boethius’s prison pastime intersects with a fascinating development emerging from Silicon Valley.

“A string of job postings from high-profile training data companies, such as Scale AI and Appen, are recruiting poets, novelists, playwrights, or writers with a PhD or master’s degree,” reports Andrew Deck for Rest of World. “Dozens more seek general annotators with humanities degrees, or years of work experience in literary fields. . . .”

The reason? “The companies say contractors will write short stories on a given topic to feed them into AI models,” explains Deck. “They will also use these workers to provide feedback on the literary quality of their current AI-generated text.” One company, Mercor, currently pays poets $150 an hour for such services.

Poets training robots: It’s a sci-fi story come true. It also exposes something interesting about the basic divide between human creativity and its algorithmic derivatives. “Tools like ChatGPT were built to mimic human writing,” says Deck, “not to innovate on it.”

As Mercor’s 22-year-old CEO Brendan Foody tells economist Tyler Cowen, these poets aren’t just writing poems; they’re also building rubrics—criteria to train large language models on what good poetry supposedly sounds like. “When one of the AI labs wants to teach their models how to be better at poetry,” he says, “we’ll find some of the best poets in the world that can help to measure success via creating evals and examples of how the model should behave.”

Does it work? Depends on what you expect.

Robot writing poetry, maybe. Generated by ChatGPT.

Insufficient Training Data

Poetry—or any creative work worth the name—doesn’t really follow rubrics, not exactly. Every art has norms and forms the artist follows, but for the final product to succeed the artist must use the conventions to do something unconventional, something that subverts expectations and surprises the audience. Otherwise, what’s the point? Is there anything more disappointing than a novel whose end you easily anticipate?

When Cowen asked Foody how long until an LLM can write something as good as the median Pablo Neruda poem, both agreed it’s less than a year away. But the best Neruda poems? “Much further out,” says Cowen. Foody agrees. The rubric gets you to the median, but to surpass the average you need something that better evals and more training data can’t manage. What is it?

When we ask ChatGPT to write us a poem or tell us a story, it’s doing so based on all the data upon which it’s been trained, including innumerable books of poetry and other works of literature. One dataset used by Meta, Bloomberg, and others contains over 191,000 books—much to the frustration of many of their authors. And that frustration is illuminating, actually essential.

Where, after all, does poetry come from? Thinking of Bonhoeffer and Boethius, from a place of longing and suffering, of fraying hope and discouragement and the need for self-assertion amid humiliation and abasement.

We can train AI models to mimic the product of those emotions, but I don’t see how even the most advanced models can originate such products without experiencing a feeling as simple as frustration, let alone the higher yearnings and struggles of the heart. The AI don’t have—and can’t get—the training data. They don’t experience fear and disappointment, anger or love, or any of the raw encounters with life that humans transmute into art.

And, atop that limitation, they’re missing something else, something Bonhoeffer and Boethius show us.

The Choice to Create

I’m a fan of AI, particularly large language models, as I’ve made clear here before. As I argue in The Idea Machine, large language models represent the latest development in a long line cognitive tools humans initiated 3,000 years ago with the invention of writing to help domesticate data and manage the wild tangle of information we create.

What’s more, I see tremendous upside from AI’s ongoing development as a creative partner in all sorts of human endeavors, including writing and translation, medicine, manufacturing, finance, energy, engineering, and more. The whole point of technology is to emancipate human labor by enhancing human capabilities. As our tools allow us to transcend our native limitations, we are free to pursue other, higher aims—such as, for instance, art.

Critics of AI say we risk dehumanization as technological capabilities expand. I agree, which is why mindful engagement is required. But while the future of AI is every bit as undetermined and unknowable as all other futures, there still stretches a vast gulf between man and machine, something Foody acknowledges.

Models are quickly advancing, capable of automating, say, 50 or 75 percent of what human experts can do. But, admits Foody, they “will really struggle with that last 25 percent.” Some of that struggle is permanent because, as stated above, it involves the raw human data AI can’t access.

But the examples of Bonhoeffer and Boethius point to the deeper challenge for robot poets. Knowing all the reasons for doing one thing—say, completing Ethics or translating Aristotle—we can nonetheless blow it off and do something else: write a poem, or paint, or garden, or make music, or bake bread, or dance, or really anything else. The choice to create, especially when supposedly more important work awaits, is every bit as human as the ability to do so. And in some sense, that’s the whole 100 percent, not the last 25.

Wasted effort? “Time alone distinguishes.”

---

Thanks for reading! If you enjoyed this post, please hit the ❤️ below and share it with your friends (including anti- and pro-AI types alike).

Not a subscriber? Take a moment and sign up. It’s free for now, and I’ll send you my top-fifteen quotes about books and reading. Thanks again!

Before you go, see these:

Tour My New Book, ‘The Idea Machine’

Bookish Diversions: Use AI, Lose Your Book Deal—and Maybe More

Machines Don’t Believe. Is That Our Secret Advantage?