The rich history of human painting unfolds across tens of thousands of years, with its origins traced back to the Paleolithic Age.

During this epoch, people adorned rocks with figurative designs, imbuing them with possible mystical meanings.

Over time, painting evolved into a multifaceted art form, serving as a visual language to depict human life, historical events, myths, legends, and the wonders of natural landscapes.

AI painting, although having roots in early computer technology, took on a distinctive form in a relatively niche field. A pivotal moment occurred in 2012 when Chinese scientist Wu Enda spearheaded a project investing one million dollars, employing 1,000 computers and 16,000 CPU resources.

This initiative aimed to train one of the world's largest deep learning networks at the time, guiding computers to generate images of cat faces. After an exhaustive three-day training period, a fuzzy cat head emerged, marking a milestone in AI-generated art.

However, the origins of AI painting reach further back than commonly perceived. In the 1960s, as computers made their debut, artist Harold Cohen embarked on an innovative venture.

In the 1970s, Cohen, a professor at the University of California, San Diego, initiated the development of the AARON computer program designed for painting. Unlike contemporary AI paintings producing digital works, AARON controlled a robotic arm, translating coded instructions into tangible artworks.

Cohen's dedication to refining AARON continued for decades until his passing. By the 1980s, AARON demonstrated the capability to draw three-dimensional objects. Progressing into the 1990s, AARON expanded its repertoire to include a vibrant spectrum of colors.

Today, it is claimed that AARON continues to produce art, although the specifics of its creations remain veiled due to the proprietary nature of its code. Cohen's artistic style, characterized by colorful abstract paintings, remains embedded in AARON's output, a testament to decades of iterative learning.

While AARON's intelligence remains a subject of debate, its historical significance as the pioneering program in automatic painting using a physical canvas earns it the title of the originator of AI painting.

Teaching AI to paint is a complex process involving the construction of training data derived from existing paintings. The challenge lies in the vast amount of information encoded in a painting, measured by the length and width of RGB pixels. The initial approach involves obtaining an AI model that outputs a regular combination of pixels.

However, distinguishing between genuine paintings and mere noise is intricate. Richly textured and naturally stroke-infused paintings involve numerous parameters such as position, shape, and color, exponentially increasing the computational complexity of deep model training.

Following Wu Enda and Jeff Dean's pioneering cat-face generation model in 2012, AI scientists intensified their efforts in this burgeoning field. In 2014, a significant breakthrough emerged with the introduction of the Generative Adversarial Network (GAN).

This deep learning model operates on the premise of an internal balancing act between two programs, the "generator" and "discriminator," resulting in transformative AI art creation.

The evolution of AI painting traverses epochs, from the Paleolithic Age to the present day, marking milestones with each stroke of innovation. As technology continues to redefine the boundaries of artistic expression, the fusion of human creativity and artificial intelligence promises a fascinating future for the world of visual arts.