Neural networks have been responsible for some of the most intriguing creations, from candy heart messages to insane recipes. Now, though, someone has tried to use the power of neural networks to make Dungeons & Dragons creatures, and the results are bizarre.
The popular tabletop game is already filled with weird and unique monsters as it is, like owlbears, for example — a combination of a bear and an owl, for those who didn't get that. But neural networks take it to the next level.
Janelle Shane, a research scientist who dabbles in neural networks, has created a whole roster of new D&D creatures based on the creature names of D&D 2nd Edition, which was released in 1989.
Neural Networks And 'Dungeons & Dragons'
To generate the monsters, Shane used what's called a recurrent neural network, she explains in a blog post. This type involves a highly complex algorithm, but fundamentally, it is fed text and uses that to learn a particular pattern. Once it does, it then creates new words based on the basic rules of the original words that were given to it.
Below are just some of the results:
• Lycanthrope, Wereladoo
• Pat, Great, Space
• Spectral Woof Greepy
• Animal, Dome
• Dwarf, Giant
• Dragon, Death Seep
• Dragon, Purple Fang
• Dragon, Curple Lard
• Dragon, Dead
Shane even has more creature names that couldn't fit into her post, and she'll give them away provided users leave their email in a signup sheet on her Tumblr page.
What's great about Shane's neural network-generated monsters is that it complements the inherent imaginative quality of D&D. Oftentimes the tabletop game is all about imagination, it should be noted, and visualizing something like a Shadowstaffer or a Burglestar incorporated into the game would probably be really stimulating.
Next Neural Network Project
Shane is already developing her next neural network project, in which she'll use the technology to create D&D character names. She needs help for this new undertaking, however. She's calling for character submissions in order to help train the neural network algorithm. Again, the more information it's fed, the higher the chances it'll generate more compelling results, so if you're a D&D diehard, consider helping Shane's project.
Also, if you just can't get enough of D&D stuff, Shane also has this other project where she used neural networks to generate spells, which includes a move called "Summon ass." Yep.