A Neural Network Generated a Bunch of Mutated-Looking New Animals
"Straying even further into the unknown, the model produced weird abstract patterns and unidentifiable entities, all with a vaguely biological, 'life-ish' feel to them."
You probably remember the neural network that generates photos of people who don’t actually exist. You might even remember the one that spits out photos of nonexistent cats, or the one that whips up fake résumés, or the one that dreams up listings for imaginary rental properties.
Now, a programmer named Aldo Cortesi has created an even stranger algorithm — one that draws silhouettes for nonexistent animals, some of which look plausible and others which look like nothing you’ve ever seen before.
In a post about the project, Cortesi wrote that he was indeed inspired by algorithms that generate human likenesses.
“Looking at these images, it seems like the neural net would have to learn a vast number of things to be able to do what these networks were doing. Some of this seems relatively simple and factual — say, that eye colours should match,” he wrote. “But other aspects are fantastically complex and hard to articulate. For instance, what nuances are needed to link the configuration of eyes, mouth and skin creases into a coherent facial expression?”
“Of course, I’m anthropomorphising a statistical machine here, and we may be fooled by our intuition — it could turn out that there are relatively few working variations, and that the solution space is more constrained than we imagine,” he added. “Maybe the most interesting thing is not the images themselves, but rather the uncanny effect they have on us.”
While he was mulling the implications of that tech, he heard about a database called PhyloPic, which collects silhouettes of animals. He wondered what would happen if he used that cache of data to train a new neural network, designed to create zoological designs.
Curiosity got the better of him — and, he wrote, he was “pretty sure I would get a few good prints for my study wall out of it” — so he adapted existing machine learning code and trained the model using the PhyloPic database. In the end, he generated 50,000 images and sorted through them by hand to find the best examples (and, he admits, sometimes adding filters or flipping them for aesthetic purposes).
The result is a menagerie of organisms that range from the plausible to the nightmarish — or like uncanny hybrids from Alex Garland’s 2018 sci-fi flick “Annihilation.”
He also produced a range of alarming-looking hominids:
As well as winged insects:
And other types of bugs:
As well as things that, well, defy easy categorization:
The whole thing is worth a read — and Cortesi’s post contains even more examples of the grotesque creatures his algorithm generated.
“Straying even further into the unknown, the model produced weird abstract patterns and unidentifiable entities, all with a vaguely biological, ‘life-ish’ feel to them,” he wrote.
READ MORE: Generative zoology with neural networks [Aldo Cortesi]
More on neural networks: Watch an AI Die, Neuron by Neuron