Skip to main content

There’s a lion in London’s Trafalgar Square that eats words and roars AI-generated poetry

Share this story

The four bronze lions that surround Nelson’s Column in London’s Trafalgar Square are pretty passive-looking creatures. They sit on all fours and gaze blankly ahead, more sphinx than lion. Legend has it that their sculptor originally planned for the animals to be posed in more active stances, stood up on their hind legs and roaring at the square. But Queen Victoria reportedly vetoed the decision as too shocking. Now, 151 years after they were originally unveiled, the lions have a new colleague, and he is definitely turning heads. 

Lion number five is the creation of Es Devlin, a designer best known for her stage sculptures that are used in opera and theater as well as the world-spanning tours of Beyoncé and Kanye West. Devlin says she was inspired by the anecdote above to make a sculpture that could participate actively in the life of the square, which has long been a site for both celebration and protest.

“The thought lodged in my mind,” writes Devlin for Google Arts & Culture, which sponsored the work for London Design Week. “What if we could invest the lion with a diversely crowd-sourced single collective poetic voice?”

Visitors “feed” the lion a word using a tablet, and it uses AI to generate a poem in response.
Visitors “feed” the lion a word using a tablet, and it uses AI to generate a poem in response.

The end result is Please Feed The Lions. Members of the public can approach the sculpture, and, using a tablet, enter a word of their choice for it to “eat.” The lion then roars at the crowd (a sound which is more ethereal than animal, full of swelling strings and distant screams) while a screen inside its mouth displays a snippet of AI-generated poetry based on their input. At night, these words are also projected onto the lion’s body, moving like a torrent of fireflies, and up the side of Nelson’s 169-foot column.

While Devlin is responsible for the overall concept of Please Feed The Lions, technologist Ross Goodwin is the man who created the poetry. His AI generator is built from some common machine learning components; in this case, it’s what’s known as long short-term memory recurrent neural network, or LSTM network. Like many machine learning systems, this generates its output by finding and replicating patterns in a dataset.

This dataset is a huge corpus of 19th century poetry, some 25 million words of it, that was compiled by Goodwin. The neural network scans this information, and it learns to predict what letters tend to follow one another. Then, when someone enters a word, this acts a “seed” for the network, which predicts what letters might follow and churns out some poetry in response.

At night, the AI-generated poetry is projected onto its body.
At night, the AI-generated poetry is projected onto its body.
During the day, the poems can be seen in a screen in the lion’s mouth.
During the day, the poems can be seen in a screen in the lion’s mouth.

The resulting verse is readable, although not exactly coherent. When I visited the sculpture, I fed the lion the word “bone,” and in return, I got: “That bone to shadow of the corner smoke / Of calm and splendor and new moon and sky.” Which, you know, fine.

Leaving aside questions of poetic quality, the lion is certainly entrancing. It’s an exact replica of the original four, which were scanned using LIDAR (the laser-bouncing technology that guides self-driving cars) and then re-created in a resin cast. More noticeably, it’s painted in a shade of vermillion so bright that it warps its surroundings.

The color certainly pops in pictures, but it’s just as effective in real life. It obliterates detail, and from a distance, it looks like the sculpture has been edited out of reality, leaving behind a lion-shaped hole where a lion ought to be. This is perhaps the most effective part of the installation. Despite the wonders of AI-generated poetry, we’re still more interested in very bright objects, especially when that color clashes with the classicism of its surroundings.

The AI poetry is projected onto Nelson’s Column.
The AI poetry is projected onto Nelson’s Column.
You “feed” the lion words using a tablet.
You “feed” the lion words using a tablet.

Both times I visited the sculpture, at lunch and again in the evening, there was a curious crowd waiting to enter a word and see what poetry the sculpture would generate.

Watching people come and go, it reminded me of many of our interactions with AI systems, which we sometimes treat as digital Ouija boards, asking them to retrieve gnostic wisdom for us from beyond the digital veil. Think about those auto-predict keyboard games that sometimes go viral that ask you to type a phrase like “In the future, I will be” in iOS and then hit the suggested word button until you’ve filled out a sentence.

With Ouija boards, we used to ask the spirit world for guidance, half tongue-in-cheek, half serious. Now it’s algorithms that seem like the keepers of a similar, obscure knowledge. People would feed a word to the lion, read the results (“Love like a light unbidden shone the wave / and the still sunshine of the world would glide”), and then walk away looking slightly puzzled, as if they’d hiked up a mountain to ask a guru for guidance and now had to go away and ponder the mysteries of what they’d been told.

Not everyone was impressed, though. As I was taking photos of the sculpture, I saw a smartly dressed man walk up to the lion and rap it on its flank. The surface echoed in a rather wet and disappointing fashion, like plywood, and the man shook his head.

“It’s an obscenity,” he said when he noticed me looking. I asked him to explain. “Because it’s a lion, and a lion is a carnivore, and here it is eating poetry. It’s denigrating art.” And what did he think of how it looks? The man sniffed: “Ocular stimulation.”

No arguments there.

From a distance, the sculpture looks like it’s been edited out of reality.
From a distance, the sculpture looks like it’s been edited out of reality.

Photography by James Vincent / The Verge