Fermented Code: Modelling the Microbial Through Miso
Fermentation is the focus of the first season of Compendium, an evolving collective archive initiated by Serpentine’s Synthetic Ecologies Lab. In this companion essay to Compendium Season One: Microbial Lore, writer Claire L. Evans journeys outward from a batch of homemade miso, considering fermentation from the microscopic level to vast scales of time, space and transformation.
It seemed like a good idea: make misery miso.
It was April 2020. I was growing scallions on the windowsill and reviving my sourdough starter, both nervous reactions to the uncertainty of the moment. I was so anxious, I’d wake up every morning with my heart racing, my somatic system primed for reality’s daily onslaught. By making miso, I thought I’d sublimate the year’s dread into something meaningful. If nothing else, it would pass the time. Making miso is an ancient ritual, but as I mashed my boiled soybeans, I imagined the future and all its flavours. At the end of the year, I thought, I’d decant the miso from the stoneware crock in the cupboard and dole it out in jars to my friends, proclaiming, “we made it!”
By the time Christmas came around, I already knew that something had gone terribly wrong. I lifted the weights, peeled plastic wrap from the surface of the miso and scraped the thick layer of salt separating it from the air. Over the last few weeks, instead of the warm, sweet odour of miso, my kitchen had slowly filled with a pungent rot – a smell closer to a strong blue cheese. A dark blue-green mould crept malevolently along the edges of the crock. My misery miso had lived up to its name, and to the year of its creation. Like me, it was miserable.
Of course, you can’t ruin a miso with bad vibes alone. But as the American fermentation revivalist Sandor Ellix Katz writes, fermentation “provides us with a powerful metaphor with infinite regenerative power.” Social and political change bubbles like an active ferment, transforming the world over time. And the persistence of fermented foods across 2,000 years of human culture suggests that resilience is a consequence of entanglement, rather than isolation. Seoul-based collective Rice Sisters Brewing Club uses these metaphors in practice, staging artistic ‘social fermentations’ exploring how social models might ripen and transform with the help of a starter. Fermentation, as they explain it, “represents the clash of agreements, disagreements, and sparks of resonance when different beings meet.” But in the case of my misery miso, the metaphor was more straightforward: if you’re going to cook with time, make sure you’re having a good time.
Around the same moment that I started making my miso, I developed a keen interest in the overlaps between computation and biology. This was an extension of the same impulse that had driven me to the miso to begin with: a growing awareness that the solutions to the future’s most intractable problems would not be found exclusively in silicon chips or Silicon Valley, but by drawing inspiration, as Katz and others suggest, from the webs of life beneath our feet, in our gardens and in our guts. I busied myself reading about slime moulds, those odd unicellular organisms whose searching protoplasmic tubes can easily solve complex mathematical and networking problems, and spoke to researchers building living robots out of frog cells.
Just as the misery miso was making its final turn, I was thinking and writing about Artificial Life, a scrappy discipline of computer science that concerns itself with reproducing life in silico. In the course of this research, I had a long conversation with an iconoclastic evolutionary biologist-turned Artificial Life researcher named Tom Ray. In the early 1990s, Ray created a silicon ecosystem called Tierra on his Toshiba laptop; it was home to a species of fiercely competitive code-based lifeforms. Tierra was not a model of evolution, Ray explained to me – it was an instance of it. Although his ‘organisms’ were nothing more than snippets of self-replicating code, their complex interactions sparked adaptations, predations, mutations, and
all manner of evolutionary behaviours. As far as Ray was concerned, Tierra’s code was very much alive. Observing it, he felt as though he were back in the jungles of Costa Rica where he’d done his field work, watching butterflies, beetles, ants, and vines in their dance across the
Tierra was a watershed project in the history of Artificial Life. Today, the field is a veritable zoo of flocking, battling, evolving virtual creatures, and researchers studying emergent interactions between such organisms, hoping to better understand the dynamics of complex systems and the open-ended evolutionary process that produced human-level intelligence. Many also believe that biological processes can build better software. But what drove Ray to create Tierra was simply impatience. In his early fieldwork, he’d been limited by time. Evolution plays out over millions of years, and even in the jungle it’s impossible to observe a whole evolutionary cycle. That’s why biologists so often work with short-lived organisms, like fruit flies, whose evanescent lifespans speed the passage of generations, making it easier to observe genetic changes. For Tom Ray, building a virtual ecosystem was a way to accelerate evolution, to watch it play out over rapidly successive generations of mutating code. To cook with time, if you will.
Which brings me back to my misery miso. Over seven months, that cursed crock had hosted an epic microbial saga played out by bacteria, mould, enzymes, and yeast. Several strains of halophilic bacteria had faced a parade of minor salt-averse species before they’d all, ultimately, been bested by some more tenacious local mould. Generations of microorganisms lived and died, fought and cooperated, to create the outcome I ultimately scrapped. Seven months in the cupboard, to them, traced the rise and fall of an entire culture. What is time to a spore? To a bacterium? Depending on where you stand, fermented foods are either products of slow and uneventful processes – a mere waiting game – or eventful records collectively formed by thousands of lives.
Indeed, fermented foods have a way of making time observable, even edible. First there is the now of packing the crock, punching vegetables under brine, or kneading starter into autolysed dough. This is done in anticipation of some future moment days, weeks, months, or even years ahead. And as the miso, kimchi or sourdough matures, another timescale unfolds, measured in the reproductive lifespan of yeasts, in the proliferation of bacterial lives, in cycles of latency and efflorescence. This is what produces a ferment’s unique flavour and aroma. Fermentation allows us to taste time, but it is a hedge, too, against time. In cultures around the world, with the cooperation of helpful bacteria and ambient yeasts, fermentation prolongs the shelf life of food, transforming an overabundance of produce into tart, pungent, savoury staples that enliven the palate. A wild mountain vegetable, preserved in brine, carries the flavour and nourishment of springtime deep into the winter season. Fermentation is a controlled entropy that transcends the inevitability of rot. It slows, prolongs and preserves.
The aroma of my misery miso permeated the kitchen for weeks. It drifted into my office as I read about creatures made from code fighting for resources in a universe the size of a Toshiba laptop. The smell was sweet, damp, foul, alive. It betrayed in one sniff the complexity of the microbial community from which it emanated. In its presence, I found myself wondering what my miso – a universe the size of a pound of boiled beans – would look like as a simulation. A community of microorganisms digitised and set loose in a virtual crock! Why not? It certainly wouldn’t smell. And that way, I could run my miso again and again, tweaking the parameters and accelerating time until I got it right. This fleeting curiosity took hold of my mind like mould on bread. What if I could ferment code? Or, conversely: what if I could model miso?
Fermented foods exist on a spectrum of complexity. On one end of that spectrum are natural microbial ecosystems, which can consist of thousands – even hundreds of thousands – of interacting species. On the other are small ‘model systems’ of microbes studied by biologists in the lab. These are not models in the mathematical or computational sense, but rather in the same sense that E. Coli is a model organism: they serve as simple living representations of more complex systems. By studying the interactions between species in these pared-down model systems, biologists can infer the broader principles at play at a much larger scale in real-world ecosystems. Most microbial lab models consist of two to six species, a sweet spot “sufficiently complex to answer the question at hand, but not more complicated than required.” Some models are entirely synthetic, consisting of microbes that would never ordinarily interact in nature. Others are drawn from natural ecosystems as diverse as the human gut, soil and cheese rinds.
In fact, recent research suggests that traditional fermented foods, like cheese and kombucha, are the perfect place to look for communities of microbes to develop into ‘experimentally tractable’ model ecosystems. One of the most robust microbial models in current use, Yeast-LAB, is a three-species community consisting of Brewer’s yeast and two lactic acid bacteria that are commonly found in sauerkraut and cheese. The microbes at work in fermented foods play well together, their kitchen-friendly growth substrates – milk, tea, soybeans – are readily available, and the conditions necessary for them to thrive have already been tested through centuries of human culinary ingenuity. They are complex, but self-contained. Using fermented foods as a starting point, researchers can move quickly from observing microbial diversity in the wild to building ‘highly manipulable in-vitro systems.’ This makes fermented foods ideal candidates for studying the mechanisms of microbial community formation, still a great unknown in microbiology.
My miso, then, is already a kind of model – one good enough to eat. It is a semi-natural interface between the artificial simplicity of a lab system and the dizzyingly diverse communities of microbes flourishing outside my kitchen door. Fermentation is a way of capturing the complexity of microbial ecology, literally drawing it from the air, rendering it knowable, and ultimately domesticating it. Japanese farmhouse miso, whose production dates back to at least 600 AD, is traditionally prepared by leaving cooked soybeans out to capture wild mould spores from the environment, which makes each pot of miso a sampling of a region’s microbial terroir. Today, although 90% of Japanese miso is produced in factories using a commercial strain of Aspergillus oryzae mould and pure starter cultures of the lactic acid bacteria Tetragenococcus halophilus and the yeast Zygosaccharomyces rouxii, recent surveys of commercial miso and koji – the mould-inoculated grain used in the first phase of miso production – still reveal complex, diverse microbial communities under the surface. Even the most standardised, plastic-sealed tub of grocery-store miso is host to dozens of species of bacteria and yeast. We may sterilise, streamline and contain, but complexity follows of its own accord. This, too, is a likeness of
Some biologists question the usefulness of model systems, wondering if they’re too artificial to provide real insight into the dynamic architectures observed in surveys of natural communities. Ecological patterns span spatial and temporal scales; the more-than-human world unfolds at the scale of individuals and entire populations, across ecological and evolutionary time. Each microbe, plant, vertebrate, and insect has its own vantage on the system; each constitutes the others’ environment; each responds to disturbances in others. Each species exists in real time and across deep time. Into this morass wanders the field ecologist, whose perspective is narrowed by the publishing demands of short-term research grants. Microbial models offer a rare opportunity to transcend these issues of pattern and scale. In the lab, elements can be isolated, noise winnowed away, the flow of time condensed. Reality itself turns experimental. Like Tom Ray in his digital jungle watching evolution proceed at a clip, an experimenter studying a model community can observe phenomena like succession, coevolution, invasion and even the effects of climate change over generations.
The appeal of a model system lies not in reproducing nature precisely, but in putting it together, bit by bit, until broader mechanisms begin to emerge. In this sense, the use of model systems in biology is not unlike the use of computational systems like Tierra in Artificial Life. As the computer scientist Christopher Langton writes, Artificial Life is “simply the synthetic approach to biology: rather than take things apart, Artificial Life attempts to put living things together.” Life is a nonlinear phenomenon, he argues, and although traditional biology has given us a solid understanding of the mechanics of life on Earth, it has largely missed what emerges from the dynamic interactions between life’s constituent parts. To address this, Artificial Life simulates and then observes the behaviour of those parts in each other’s presence. It’s colloquially understood as a bottom-up approach – that is to say, synthetic rather than analytical. As the historian of science Evelyn Fox Keller observes, the word ‘synthetic’ does double duty here, meaning both constructed – put together – and artificial – simulated.
Of course, there are degrees of artificiality. A microbial model system does not exist in nature, but its constituent microbes are nevertheless real – they are alive, after all. Conversely, the organisms in an Artificial Life system are not alive, but their interactions, and the processes emerging from those interactions, are “every bit as genuine as the natural processes they imitate,” writes Langton (or as Tom Ray put it to me: not a model, but an instance). In between microbial models and Artificial Life systems stretches a wider mathematical expanse: numerical simulations of the human microbiome, industrial models of fermentation kinetics, individual-based simulations of the 3D dynamics of microbial communities, genome-scale metabolic models and so on. Like microbial models, these computer models trade fidelity for function. Which is to say, they aren’t real either – but through them, it’s possible to reproduce real phenomena observed in surveys of microbial biodiversity like the Earth Microbiome project and the Human Microbiome Project. And perhaps understand them.
There will always be something a model misses or ignores. This is by design. Making a model requires elision: choosing between what is meaningful and what is merely noise. This is complicated by the fact that what appears to be noise at one scale can look like a signal at another. And so, the model must exist suspended like a pearl in time, dynamic and static at once. Model-makers talk of ‘sweet spots’ and ‘Goldilocks points’, those just-right balances between complexity and simplicity. They must decide what matters; what they are looking for dictates how they look. As such, there can never be one model for any given system; there must be multiple, complementary models, each with its own focus. Many misos with their own flavours. It’s in the overlaps between these models that truth is found, if at all. As the biologist Richard Levins wrote in a highly influential 1966 paper on modelling in population biology, when it comes to forming any kind of robust theorem based on mathematical models, “truth is the intersection of independent lies.”
That is to say: modelling is a practice of constructing the false in pursuit of the real. Building a completely faithful reflection of a living system is impossible. You wouldn’t want to, anyway, since the system is already a perfect model of itself – a map of its own territory. A true duplicate would consume enormous computational resources, and in the end it would be as difficult to interpret as the thing it was modelling. Instead, models shave away unnecessary detail in order to produce a more general picture of a system’s mechanics, and in turn expand our understanding of how it forms, stabilises and works. We still do not know if microbial communities adhere to fundamental ecological laws, or how patterns observed at the microscopic scale influence ecology at the planetary scale. This understanding will only become more vital as we face the challenges of the coming century: microbes, interacting around and within us at all times, quite literally sustain life on Earth. If we fail to understand them, they’ll outlive us.
To return to my original question: yes, one could model miso, although the server fans would be whirring before long. Complexity begets complexity. Recent research in ‘unconventional computing’ suggests that we might do well to skip the model entirely and turn to the natural world itself as a computing substrate, building computers from slime mould, fungal mycelium, seedlings, and perhaps entire ecosystems. Until then, miso, like all fermented foods, is a bridge between the Petri dish and the microbial wilderness. Someday soon an enterprising biologist will draw from their own homemade miso a hardy model system. In sterilised isolation in the lab, that model will tell its own stories. They will not be exact reflections of what microbial communities do in the wild, but they will contain echoes. As those echoes reverberate and overlap, the model will reveal – briefly, ecstatically – the shape and song of the more-than-human world.
After I threw out my misery miso, I tried to clean out the stoneware crock. First, I washed it with soap and hot water, but the smell lingered. I soaked it in vinegar, but when it dried, the smell remained. I boiled the crock and rinsed it with bleach. Nothing changed. Dark mould stains crawled across the porous stone. Finally, I gave up. The mould remained, another echo. I think of it as a reminder of all that is still unknown, and unknowable: that which exists beyond the model’s frame. Of course, what the model cannot – or will not – represent is the stuff of life itself. It is its ungovernable mess, its noise and stench. It is the very mystery the model is built to clarify. It is a wilderness that cannot help but creep into even the most contained system. In traditional foodways around the world, fermentation vessels are used and reused, carrying communities of microbes across human generations. Along the way they become inoculated, too, with stories. This is only one.
Claire L. Evans is a writer and musician. She is co-author of the Grammy-nominated group YACHT, author of Broad Band: The Untold Story of the Women Who Made the Internet, and co-editor of the speculative fiction anthology Terraform: Watch Worlds Burn. She lives in Los Angeles, where she is an advisor to graduate design students at Art Center College of Design.
 Sandor Ellix Katz, Fermentation as Metaphor, Chelsea Green Publishing, 2020.
 Rice Brewing Sisters Club, “Seokkeodungdung: Doing “Social Fermentation”, Hackers, Makers, Thinkers: Collective Experiments in Social Fermenting, (conference), 2022.
 Johan Bengtsson-Palme, “Microbial Model Communities: To Understand Complexity, Harness the Power of Simplicity”, Computational and Structural Biotechnology Journal volume 18, 2020.
 Benjamin E Wolfe and Rachel J Dutton, “Fermented foods as experimentally tractable microbial ecosystems”, Cell 161.1, 2015.
 Joanne G Allwood, Lara T Wakeling, David C Bean. “Fermentation and the microbial community of Japanese koji and miso: A review”, Journal of Food Science 86.6, 2021.
 Christine M Jessup, Rees Kassen, Samantha E Forde, Ben Kerr, Angus Buckling, Paul B Rainey, Brendan JM Bohannan. “Big questions, small worlds: microbial model systems in ecology”, Trends in Ecology & Evolution 19.4, 2004.
 Evelyn Fox Keller, Making Sense of Life: Explaining Biological Development with Models, Metaphors, and Machines, Harvard University Press, 2003.
 Robert Marsland III, Wenping Cui and Pankaj Mehta. “A minimal model for microbial biodiversity can reproduce experimentally observed ecological patterns”, Scientific Reports 10, 2020.
 Richard Levins, “The Strategy of Model Building in Population Biology”, American Scientist 54.4, 1966.