Trapped in the Bottleneck

I sit near the door in a fluorescently lit seminar room. The walls are all white, one green chalkboard. A table and several high back wheelie chairs fill nearly all the available space in the room. There are seven of us, no extra seats. The graduate seminar is full. I look around at the Patagonia and LL Bean fleeces, the faded t-shirts with conservation emblems on their sleeves. The quarter-zip sweaters identifying research groups. I pull my skirt over my knees. I am not supposed to be here. 
‍

The professor sits at the head of the table and leans back in his chair. He speaks to one of my classmates like an old friend—already knows everything about her lichen research. I’m glad you decided to take this course, he says. I think it will be good for your project. 

We have emailed briefly, the professor and I, so I know that he is one of those younger, informal types. He doesn’t bother with formatting in his emails, eschews punctuation. He signs his name as a single initial, forgoing credentials—the confidence speaks for itself. I don’t see any reason why not, he said when I asked if I could join the class. He looks straight at me when he suggests we go around the room and introduce ourselves. 

Of course they know each other already; I am the stranger. I feel my cheeks get hot when I describe myself. I see the contempt on their faces—an English major? 

This is evolution, the professor says. He leans back in his chair as he speaks, looks up at the ceiling as if he can see the stars through it. He describes the course, the field, his research; he suggests that everything I think I know about change is wrong. 

*

Around ten thousand years ago, glaciers receded. Ice crawled away from the belly of the Earth and closer to its poles as temperatures warmed. 

Geologists sometimes refer to glaciers as natural archives for the way the ice indexes the elements of the past. The scientists drill thick cores from the glacial sheet and count the layers of compacted ice, send their instruments into the tiny pockets of air preserved within. Their composite data reveals the shape of a changing climate: Here the ice creeps forward, here it pulls back. 

The movement of water contorts every cell of life on this planet, from the rivers and lakes on which we build communities to the molecules that surround our DNA. And the end of the last glacial period, when all that water changed shape, other contours of life on earth changed too. 

During this time, the end of the Pleistocene, there were many extinction events: The dire wolf, the woolly mammoth, mastodons, sabretooths, the North American lion. The disappearance of the planet’s megafauna is perhaps the most well-known loss, but certainly not the only one. Many species were overwhelmed by the changing shape of water and its influences on the surface of the Earth. 

In my undergraduate class in evolution, my professor directs our attention to a diagram of a cheetah’s jaw and teeth in our textbook. Cheetahs have become a common example of limitation, of stunted potential. Under the influence of shifting water, shifting land, shifting resources, cheetah populations were decimated, leaving only a small number to breed with significantly reduced genetic diversity. Today their numbers are still low, their genetic diversity has not recovered. Because of this, cheetahs are highly susceptible to widespread disease and other environmental threats—what is likely to kill one cheetah is likely to kill all cheetahs.

This is what is known in evolutionary theory as a bottleneck. 

*

Bottlenecks are a kind of genetic drift, a term born of probabilities and frustrating math—frustrating especially to those of us who much prefer the narrative of evolution to its mechanism. The common analogy used to explain genetic drift is to imagine marbles in a jar: blue and red, equal parts. This is your population. Then you are to imagine catastrophe—wildfires, floods, swaths of ice expanding under foot—these events remove marbles from the jar. These are the dead marbles. The marbles that can no longer reproduce, can no longer reach their red selves or their blue selves ahead into the future. What is left? Only one blue marble remaining in a small sea of red. How vulnerable that colour becomes. 

*

When I take biology classes in my undergrad, I am told that I am an excellent writer. My professors leave glowing comments on my essays, fail to notice the comma splices and sentence fragments for which I am penalized in my English classes. The essays I turn in are not special; they lack insight and new ideas, only refurbish findings from reviews and summaries. I write a paper on bamboo for my Plants in the Modern World class and the A feels almost automatic. 

Communication is an extremely important skill. That’s what kind people tell me about my English degree. The unkind ones are not shy to remind me that there is utterly no utility in literature. But how will you make money? they ask. 

I meet my husband when he is at the tail end of his master’s in biology. He is all science and no prospects. The money he does manage to find is short-term and barely pays his commute to the lab. I did this to be practical, he tells me. I went into science because I was promised it would lead to stability. 

Among the science students I often feel like the lone blue marble. No wonder the arts are dying off so swiftly. We have all been fed the same lie: that the future is a stable, certain thing—one we can prepare for. 

*

In my plants class, I learn that not all bottlenecks are born of random mayhem, some are calculated. Monocultures in agriculture occur when humans decide to nurture only a single variant of a crop, removing all the blue marbles themselves. The red marble that has come to dominate our global food industry? The Cavendish banana, the only variety of banana commercially grown, the only banana you’ll ever encounter in a grocery store, the only banana you’re likely ever to know the taste of. In the case of the Cavendish banana, the blue marbles were removed from the jar for a reason—to streamline production and reduce the cost for consumers. Monocultural crops allow for more consistent harvesting methods, more consistent products, more consistent satisfaction. The Cavendish banana remains one of the most reliable and inexpensive pieces of produce on the market. We have come to depend on it. 

The Cavendish banana does not reproduce on its own; there is no re-braiding of DNA required for one plant to beget another, no opportunity for change or mutation. Humans continue the propagation of the organism so that we might continue to harvest them, each one a clone indistinguishable from the others. The Cavendish banana has also come to depend on us. 

*

In the public imagination, an adaptation is a change that enables something to become more suited to its environment. In winter, we adapt to the cold by moving inside. In an unfamiliar country, we may adapt to the culture by learning the language. We like to think of adaptation as something we can control, something we can change to improve our suitedness to the imagined future. When we escape cold by the fire, we anticipate that cold will continue, at least for the time being—and if it were to suddenly become warm, we could just as easily extinguish the flames. We can always adapt, always recalibrate to the next circumstance. 

In my evolution seminar, it is the confusion around the term adaptation that I am told makes the theory infuriatingly unintuitive. Species do not adapt by putting on sweaters, or by growing additional limbs or organs to suit their desires. Eyes did not spring into existence because an early arthropod thought it would be nice to have a look around. 

In class, we are given the example of songbirds: during a warm spring, the birds who hatch early have an advantage—finding the best food, the best shelter. The early hatchers outcompete the late hatchers, since they control the resources. In our era of unstable climate, though, thaws are unpredictable and vary year to year. During the warm spring, the early hatchers survive, they reproduce, they lay eggs containing genetic material of more early hatchers. Then the next spring is cold—the thaw is late. All those early hatchers? They freeze to death in their nests, they starve waiting for seeds to emerge from under snow. Not so advantageous anymore—the songbird population rapidly dwindles. 

Adaptation is not instantaneous. A trait surviving does not mean that trait is better, or that the individual possessing it was stronger. Evolution is an arbitrary arms race. In a jar full of marbles of all colours, one year pink may be the desirable colour, the next it may be green. 

The kicker: evolution can only act upon what’s already in the jar. If the greens were all abandoned in favour of pink in the previous season, green marbles cannot spontaneously reoccur to suit the vogue. 

My professor says: You have to think about it carefully, every time. 

What he means is: You cannot assume you already know how the story will go. 

*

The Cavendish banana, like the cheetah, is vulnerable. What kills one banana kills all bananas. Farmers and scientists have already isolated their assassin: Panama disease, a fungus that quickly spreads through crops, rotting them from the inside out. 

In afflicted fields, the infected banana plants are burned in rice hull pyres—omens of the plague sure to follow. For smaller family-operated farms in Central America, where most commercial bananas are sourced, Panama disease puts whole crops and livelihoods at stake. 

Not empty fearmongering either: not too long ago the Cavendish banana’s predecessor, the Gros Michel, fell to the same fate. A monoculture like the Cavendish, the Gros Michel was rendered functionally extinct by the disease. The bottleneck so tight it choked. 

*

In my seminar I am to write two papers, short and long, and complete a presentation. The idea of the presentation makes my skin crawl. I can only imagine making mistakes, using the wrong terminology, stumbling over the vernacular. I miss my English classes and our shared vocabulary, the words I would rather use: metaphor, syntax, analogy. 

I decide to do a project in my wheelhouse. I present on the theories of the evolution of language. I read several, but none speak to me and my efforts towards rigorous scientific scrutiny as much as Noam Chomsky’s model of merge and recursion. He directs my attention to mutation. A sudden change in our genetic code that gives rise to the capacity for complex systems of language, complex systems of thinking. 

Just like the mutations that led to precursors of the human eye, the mutation for language affords such an advantage within its environment that those that have it efficiently outcompete those that don’t. That we can speak to one another, communicate and ideate, makes us powerful. A power we are still harnessing. 

At the end of my presentation, discussion shifts: instead of mutation, we talk about the way language itself grows and contorts over time. 

I study literature. I am already very familiar with how a language twists and turns, how certain branches wither and die, how others swell and absorb, how words, also, are a response to the environment. 

*

A few years later, after a failed dalliance with medicine, I am back in the comfort of my books and in the process of becoming a scholar of Canadian literature. For my comprehensive exams, I have over one hundred books to read to become an expert. Hundreds of years of culture and history to attend to as I read. At times I feel like I am drowning in the pages. 

I read chronologically. I begin with the white settlers, the first contact, the long poems of early prospectors admiring the potential of the land, the way they keenly render it empty; the way they violently erase a world. 

When the glaciers receded across North America at the end of the Pleistocene, it was out from under the feet of the Indigenous peoples residing on the land. That is how far back their tenure goes. As part of my studying, I read the poems and songs that have been carried forward for so long that some of them may have touched the ice.

The poetry is what begins to make my mind whirl. I read thousands of poems, some of them hundreds of pages long. I read the great debates between the Romantics and the Modernists. The fury that can spark over a few lines on a page. This literary history is full of drama: In the 1940s, two Montreal magazines find themselves in a kind of literary war over the direction of Canadian poetics. Writers and their work become cannon fodder. One poet is panned for her pedestrian stylistics, another accused of covert homosexual desire. Even still, these magazines foster artistic communities and offer space for bursts of experimentation and linguistic alchemy. They pave the way for other groundbreaking poets such as Leonard Cohen, bpNichol, and Michael Ondaatje.

In the eighties and nineties, it was feminist poets like Nicole Brossard and Daphne Marlatt who became particularly attentive to how their language and poetry reflected their experience—how the words and sentences themselves needed to be reshaped and reimagined on feminist terms. Influenced by French continentalists like Hélène Cixous, they wrote their bodies into being in ways they had never been before. Marlatt took a line and stretched it beyond all convention. A sentence not direct and contained but effusive and winding, meandering, changing. What a world-making—to desire the future and write it into existence. 

In the evolution of language, poets are the red marbles that write themselves green, that write themselves yellow, that write themselves out of the jar, that unwrite the jar completely. 

*

Before I begin my PhD, I spend a year as a teaching assistant. In an early meeting with one of the professors I am working for, I am introduced to a new software, already creating havoc within the institution. 

Have you heard of this? The professor asks as she turns her screen to show the other TAs and me. A white screen with a hexagon made of interlocking segments. I shake my head. Unfortunately, she continues, many of the students already seem to know about it, and it produces much more coherent writing than you’d expect. The good news for us is there’s no substance. It makes them easy to spot. To penalize. 

I don’t worry about it. First semester goes by without any incidents, and I wonder if the fad has passed, if everyone has realized the same thing that I have. Second semester begins and I quickly learn that they have not. The AI text is everywhere, always hollow, always obvious. We struggle under the burden of proof when flagging academic misconduct. There is no paper trail or smoking gun; we are looking at a crime scene full of blood splatters before the invention of forensics. 

The frustration ultimately makes me curious. I never download the tool, but I read about its inception, watch the TEDtalk by its creator, imagine if Victor Frankenstein had been proud of his monster. I feel that the word generative is misleading; reconstituted would be more appropriate. The beast akin to No-Face from Studio Ghibli’s Spirited Away: consuming as much as it can, eating every sentence in existence, stripping each of its semantics, turning language into nothing more than probabilities. 

Another word that comes to mind: recombination. A reshuffling of genes that leads to novel genetic material. A daughter with the same nose as her father, the same hands as her mother. In the LLM words seem to be recombined like DNA. New sentences made up only of what came before. 

*

At the end of my presentation, the professor cuts me off prematurely; he does not like how long he has gone without talking. I worry this means my work is being rejected—that he is attempting to save me from an embarrassing failure. The grade I receive confirms my misjudgement: the presentation was unfaultable, highly researched. His interruption only belies that it was a topic too far outside his specialty. 

He is the one who shifts the conversation away from biology and towards culture. We can also talk about how language “evolves”—he puts the word in air quotes, locking it into its metaphor. 

My professor patronizes me with a smile as he ends my presentation before I reach the conclusion I’ve been preparing. He says, Very interesting, how we can talk about these other kinds of “evolution,” too. 

What he does not say: Of course, that’s not the kind of serious scientific work we do. 

*

The LLM is not a simple pattern replication machine—its algorithm does not always produce the most probable word to follow its precursor; no, sometimes the algorithm locates a less probable word (though one still proven to have an identifiable relationship with the word that precedes it) to use instead. This is what makes the LLM so intoxicating to its users: it is not merely a replica of text already in existence elsewhere. Each generated piece of text offers variation, difference. The language suddenly autonomous, developing without human intervention, what else do we call that but intelligent?

Another answer: once the LLM begins to consume its own output like the ouroboros that it is, it becomes informed more and more by its own recombinants, feeds on a diet of red marbles until only red marbles remain. Becomes another creature trapped in the bottleneck. 

*

In my undergraduate course on evolution, my professor is a small, soft-spoken man. Unlike many of the other biology professors, he prefers the chalkboard to PowerPoint. During one lesson, he draws the outline of a giraffe on the board. Within it, a long, looping line to indicate the animal’s laryngeal nerve. The laryngeal nerve, rather than taking the most direct route through the body, travels from the skull down to the heart, loops under the aorta, only to turn around and travel upwards to the larynx. While human anatomy includes this same detour, in giraffes the journey is much longer. 

My professor taps his chalk on the board to emphasize his point: Not a design flaw, but evidence of all mammalian origins of development. The route of the laryngeal nerve is preserved from our bodies’ aquatic legacy, when our hearts’ orientation to our skulls was horizontal rather than vertical. 

Evolution is not forward-looking; it responds only to its past. Our presents are interminably implicated by those responses. We can never liberate ourselves from them. The giraffe cannot go back in time and change the configuration of its nerves—and even if it could go back those millions of years and attempt to encourage a different developmental path, there would be no way of ensuring she ever evolved again, or if evolution out of water was even possible without this U-turn in the neck. 

*

My work has been described as frivolous, superfluous, nonessential. I am reminded regularly by arrogant scientists, by shrinking arts departments, and dwindling funding that there is no utility in writing, in studying literature. I have been told that the AI will soon replace me. It’s odd to me that they can’t see it—the perils of encouraging a monoculture of science and technology. 

*

Ten thousand years ago, water began reshaping the earth. Today, it is the data centers moving water—away from communities and farmland and into cooling chambers to offset the enormous energy expenditure and heat production of AI. A different shape. One we have done deliberately to ourselves. 

When I watch my students, my friends, or my colleagues using LLMs, I think about the difficulty of writing, of choosing words and structuring sentences. When I receive an AI-written email or paper completely without flaw, inbred and recycled, I wonder what sentences the sender might have chosen, had they been the one to choose them. What if it had been something imprecise, imperfect, unexpected? Poetry can arise in the most unusual places. The language that rewrites the jar. That unwrites the jar. What if it had turned out to be exactly what we needed?

‍