A SMALL SAMPLE FROM ~200,000 NEW LEPIDOPTERAN GENUS NAMES GENERATED BY A NEURAL NETWORK
Zenophnua, Traglosteryx, Cortix, Urunopallis, Zencolastres, Cronoides, Corpema, Braostera, Merganemodes, Cacoschter, Colecra, Careopita, Trotonis, Adoneara, Zangycomoelus, Abriba, Sceneperanthesia, Atyrhageoris, Zacama, Yacostes, Bagda, Chroxys, Aurhalea, Aphanope, Argymophdona, Zethys, Xenobrix, Boroltiga, Odograpta, Ctenoides, Calinodes, Anioptristis, Zarniame, Basnia, Arliometrina, Barachropitia, Adurosis, Aletis, Bomblena, Yalea, Callopisia, Durgedileta, Ctenchizomelia, Cathriglopha, Chrysolypsis, Amnothoe, Catoschlorena, Caristila, Chostodes, Chidava, Chloromuca, Chymenolophis, Cartia, Marthoides, Meschoarmia, Derina, Exofilla, Epithoerochlora, Cygonna, Dixea, Micaletra, Diecomia, Brobletis, Connopus, Gnophora, Crycrenis, Dixyla, Leudoplera, Hethychomia, Gromaphrapna, Lowekiodes, Heabesis, Lanca, Dalysiodia, Danopsis, Hyparda, Hyplolotos, Callosteca, Listocledia, Heterobacta…
Species are weird. Biologists guess that somewhere between 10 and 30 million species exist on earth; somewhat fewer than 2 million are “known to science.” Species exist across a skewy and wildly-balanced range of temporalities—they stretch in densely-braided diversity across millions of years, and yet they make their presence obvious to one another in the instant of encounter. And as the microbiologist Lynn Margulies showed, species can’t be said properly to exist for microbes, which transfer genomes among lineages with extraordinary ease; the species is an artifact of the emergence of complex cells, flourishing as a phenomenon of multicellular organisms. In the career of life on earth, being-species* arises as a peculiar kind of weather.
We moderns feel our being-species keenly. The genus Homo has only one species today, the sapiens, in all its ubiquity and variety. By contrast, the genus Agrilus—the so-called “jewel beetles” of the family Coleoptera—contains some 3,000 species (of which the Emerald Ash Borer, A. planipennis, is likely the most familiar).
Looking for ways to get closer to this dark abundance of biodiversity, I took a list of ca. 1800 genus names (drawing from a list of the Coleopteran family Geometridae, commonly known in English as the inchworms) and fed them into a recurrent neural network written by Andrej Karpathy (@karpathy). A neural network is a computer program that approximates the computational work done by neurons, creating a virtual network of synaptic pathways seeking to analyze a set of target data. The network doesn’t carry any a priori rules for what makes a genus or a species, nor does it work from a model of evolution or a theory of biodiversity. It processes the target data, tries to recreate it incrementally, and then compares the results to the target, choosing the analytic pathways—the connected “neurons”—which produces the best results. In a very rough and merely metaphoric way, the process could be called evolutionary, although genera and species arise in very peculiar ways not captured by this rough metaphor of neural selection.
The neural network produced more than 200,000 new “genera.” The block of vaguely-latinate morphemes at the top of this post is a small selection from the output. Each of the extant, scientifically-recognized binomials in the Geometridae is an artifact of human attention and craft; each is represented somewhere by a butterfly or moth pinned to a board, indexed with a tag which ties it to a place, a date, a discoverer. But the sheer estimated number of genera and species which have yet to be named prompts me to wonder, not only whether we’ll ever name them all, but whether naming them all is phenomenologically possible.
Chrysolypsis, Urunopallis, Zenophnua—the neural-net genera have the flavor of taxonomic nomenclature, even though the neural network doesn’t know anything about Darwin, Linneaus, or DNA. (neither, of course, do the inchworms.) As climate change induced by the last remaining species of the genus Homo advances inexorably, perhaps we could make use of such undiscovered genera, to help us get close to the enormity of biodiversity that might be lost before it can be named.
*: I mean something other than, but related to, the Marxian concept of “species being”—something at once fraught and affect-laden and also embodied and ineluctably ecological. I’m intrigued by Thomas van Dooren’s concept of “flight ways,” in the book of the same name, which takes species as life forms with forms of life: wherein habits, migratory patterns, displays, modes of communication, all that is entrained in any conception of what a given species “is,” are understood as bound together in a living matrix embedded in history, laden with affect, and laced with storyful richness.