Connect with us

AI News

An AI invented a bunch of new paint colors that are hilariously wrong



Amidst the current trends in artificial intelligence (AI), many of us have found ourselves intrigued by the unique names given to paint colors. Janelle Shane, a forward-thinking research scientist known for her playful explorations with neural networks, took this curiosity to a new level. Shane embarked on a project to train a neural network to generate fresh paint colors along with fitting names, contributing to the ongoing conversation about AI’s capabilities.

In a post on her Tumblr blog, Shane shared, “In this experiment, I fed a dataset of around 7,700 Sherwin-Williams paint colors along with their corresponding RGB values to a neural network. (RGB represents the values for red, green, and blue colors.) Could the neural network learn to devise new paint colors and label them with appealing names?” This aligns with the growing trend of utilizing neural networks for creative tasks, showcasing AI’s potential for artistic ventures.

Shane informed Ars Technica, a reputable source for technology insights, that she chose to work with a neural network algorithm called char-rnn. This algorithm, gaining popularity in the AI field, predicts the next character in a sequence, making it particularly effective for text generation tasks. The algorithm was essentially tackling two tasks: creating sequences of letters to form color names and producing sequences of numbers corresponding to RGB values. As she closely monitored the algorithm’s progress, she observed its ability to generate colors even before it became proficient at consistently naming them—a characteristic commonly seen in AI learning processes.


Over time, as the algorithm processed the dataset, it progressively improved its capacity to generate appropriate color names. This iterative learning is indicative of the ongoing advancement in AI techniques. While the generated color names such as “Soreer Gray” (a slightly greenish shade) and “Sane Green” (a purplish-blue hue) still held a surreal quality, they highlighted the algorithm’s intriguing outputs. This mirrors the ongoing exploration into AI’s creative potential and its potential to spark unconventional outcomes.

When Shane adjusted the algorithm’s “creativity” setting, the results became even more fascinating. For instance, the algorithm produced a violet shade named “Dondarf” and a vibrant Kelly green referred to as “Bylfgoam Glosd.” This experimentation aligns with the contemporary discourse around AI’s imaginative abilities and the balance between structured and unstructured outputs.

Through numerous iterations, Shane eventually guided the algorithm to recognize basic colors like red and gray, albeit with some inconsistency. Notable instances include a sky-blue shade labeled “Gray Pubic” and a deep green hue referred to as “Stoomy Brown.” These variations underscore the dynamic nature of AI outputs.

In her conclusion, she humorously notes, “1. The neural network seems to have a strong affinity for brown, beige, and gray; 2. The neural network’s knack for inventing paint names is amusingly off-kilter.” These observations contribute to the ongoing dialogue about AI’s decision-making processes, its biases, and its potential impact on creative domains. While refining the algorithm’s parameters might have led to more accurate results, the quirkiness of its current outputs adds a layer of fascination to the broader discussions surrounding AI’s role in creativity and daily life.


Click to comment

Leave a Reply

Your email address will not be published. Required fields are marked *