A natural-history plate asks viewers to trust a strange-looking creature. Its label and tidy composition imply that someone found, classified and recorded a specimen. Argentine artist Sofia Crespo uses that familiar format to present organisms that were never collected. In Artificial Natural History, neural networks generate the images; the book-like presentation lends them the visual authority of science. The tension is the point. A convincing image can borrow the grammar of evidence without carrying evidence of a living thing.
Crespo describes the project, dated 2020–2024 on its work page, as a speculative natural-history book. She lists convolutional neural networks among its media, alongside print processes including cyanotype and giclée. The images appear as specimens with names that look catalogued, such as a numbered fish, coral, butterfly or bird. Nothing in the project asks us to believe that those labels describe discovered species. Their power comes from imitating how natural history has taught viewers to recognize a claim.
This is a more exact argument about artificial intelligence than the worn debate over whether a computer can be creative. AI image systems can make patterns from training material and present a plausible output. They cannot establish biological existence simply by producing a picture that looks taxonomic. Crespo stages the gap between seeing and knowing. The result asks what habits of classification are already built into our eye, before a generated image arrives to exploit them.
Her artist biography identifies her as Argentine and situates her practice at the intersection of AI and biological systems. That matters here because the work is neither a detached software demonstration nor an exhibit of exoticized nature. It is an artist’s sustained examination of how technology rearranges our relationship to living forms. A print medium makes the generated image feel durable; a museum or gallery setting can make it feel authoritative. Crespo uses both effects while keeping the creatures hypothetical.
The project also points back at older human technologies. Scientific illustration was never a neutral window; people chose what to collect, how to draw it, and which traits merited a label. Machine generation intensifies those choices and hides some of them in data and model design. Crespo’s invented specimens make the act of categorizing visible again. They are most useful when their seductive detail makes a reader pause: What was observed? What was inferred? What was merely made? The answer cannot come from the image alone. The work teaches a form of visual skepticism without asking us to abandon visual wonder.