I don't know if you've ever tried to paint a room – but if you have, I'm sure you've come across some bizarre paint names.
How does that relate to AI?
Well, Janelle Shane is a research scientist, and neural network aficionado – as any good data scientist should, she spends a lot of her free time playing with neural networks (… to varying results).
Recently, she attempted to train an algorithm (specifically char-rnn, a neural network that attempts to anticipate the next character in a sequence) to create new paint colors and accompanying names.
After it was fully trained, here are the results:
As you can see, it knows there are colors, but it struggles to keep the names associated with the appropriate color.
"Turdly" seems pretty close though … but I always wanted to paint my room "Stanky Bean."
As you might know, that was not the first time artificial intelligence has created something artificially stupid.
For example, how about when Microsoft attempted to create an AI chatbot named "Tay" by analyzing tweets?
Tay quickly became very …. very racist.
Clearly, mistakes were made (and Tay isn't as dangerous as a human racist) – but, it does raise questions.
AI is cool … But, artificial stupidity is still scary.
For now, that is the way it is.
As we explore new avenues and smarter algorithms for AI to be trained with, it will be interesting to see what Frankenstein's monsters we create.