Years after AlphaGo’s victories over Go masters Ke Jie and Lee Sedol between 2016 and 2017, artificial intelligence has once again captured global attention. In September 2022, at an art competition in Colorado, USA, game designer Jason Allen’s AI-generated work “Space Opera Theater” won first prize in the digital category — marking the first time in art history that an award was given to an AI-generated work. Just months later, ChatGPT swept across the globe, becoming the most popular AI software and sparking a global “AI race.”

Jason Allen’s award-winning work “Space Opera Theater” (Source: The Washington Post)

As artificial intelligence moves from the cutting edge into the periphery of daily life, people inevitably begin to reconsider and discuss questions about art and science: What is art under the influence of AI? What kind of art do humans need? How should we evaluate AI-generated art? Perhaps this presents an opportunity to more deeply understand the logic behind AI evolution and significant changes in the artistic domain.

From the Turing Test to ChatGPT

On February 15, at the 2023 World Government Summit in the UAE, Elon Musk said in a remote interview: “Artificial intelligence has been developing for quite some time; it just didn’t have a user interface that most people could use. What ChatGPT did was add a usable interface to AI technology that had existed for several years, and more advanced versions will emerge.”

In the 1950s, AI technology (hereafter AIGC) first emerged with Alan Turing’s famous “Turing Test.” By the late 20th and early 21st centuries, AIGC technology shifted from experimental to practical applications, though it still couldn’t directly generate content. Today, with the advent of Generative Adversarial Networks (GANs) and continuous iterations of deep learning algorithms, AI-generated content has flourished — from GPT-1 in 2018 to GPT-3 and InstructGPT in early 2022. ChatGPT emerged from this foundation, incorporating human feedback reinforcement learning and human supervision fine-tuning, thus acquiring capabilities for understanding context, maintaining continuous dialogue, and unlocking numerous application scenarios.

This powerful AI dialogue model with strong learning and interactive capabilities belongs to the text production domain within AIGC modalities. Other areas include audio generation, image generation, video generation, game generation, and 3D generation, all of which have achieved substantial results. Among these, image generation technology has the closest relationship with art. In the latter half of 2022, AI creation tools emerged in abundance, and while creation quality varied, they all maintained close relationships with natural language-to-image transformation.

The Emergence of AI Image Generation Platforms

In early 2022, OpenAI, the company behind ChatGPT, released the image generation tool DALL·E 2. In this model, natural language (text) can automatically generate images, and creators can continuously optimize based on results. It has precise applications in both artistic and design fields, such as illustration creation, product design, and graphic design.

Additionally, AI company Stability AI launched Stable Diffusion. Compared to OpenAI, Stable Diffusion adopted an open-source and decentralized model, allowing anyone to develop and use it through open-source code and online tools, aligning with their platform philosophy of “AI by the people, for the people.” This model emphasizes artistic and stylistic image generation, excelling in artistic image creation with adjustable parameters.

Images generated by Stable Diffusion (Source: Data Park)

Unlike DALL-E 2 with Microsoft’s backing and Stable Diffusion with substantial funding, Midjourney remains an independent research institution with minimal staff. Its most distinctive feature is community-based operation, with the model built into Discord chat software where users can publish works in real-time and learn from other users’ works and prompts. By simply entering the “/imagine” prompt, like invoking a magic spell, the system generates four new images within 60 seconds, allowing users to improve or enhance visual quality based on these results.

Chinese tech companies have also joined AI development. Besides Baidu Group’s ERNIE-ViLG, Alibaba DAMO Academy launched the M6 model, primarily applied within Alibaba’s e-commerce ecosystem. Designers need only input seasonal design requirements such as color, style, and fabric, and M6 designs numerous clothing pieces for designers to select and expand upon. Thus, M6 serves as an assistant to designers or junior designers, allowing them to focus on creativity and decision-making, improving work efficiency while reducing operational costs.

Beyond these image generation tools, there are AI video generation tools like Imagen Video, Make-A-Video, and RunwayML; game generation tools from Tencent AI Lab and Hyperparameter; and 3D generation models like Magic3D and DreamFusion.

Overall, the integration and collaboration between intelligent technology and artists has become inevitable. Low-barrier creation tools have made art more “democratic,” with painting techniques and software technology no longer serving as barriers. However, the resulting issues of artistic copyright and regulation have become more severe, and society increasingly needs designers with genuine creativity, design strategy, and user insight.

The Return of “Art is Dead” Concerns

Notably, AI creation has already entered multiple fields including the art auction market and art restoration, generating broader attention.

In October 2018, Christie’s New York auction house sold an algorithm-generated work resembling a 19th-century European portrait, “Edmond de Belamy,” for $432,500. This print, created by the French art collective Obvious using Midjourney, sold for 45 times its highest estimate, shocking the art world.

“Edmond De Belamy,” GAN algorithm, inkjet print (Source: obvious-art.com)

In December 2022, “Unfinished·To Be Continued,” a painting combining AI continuation of Republican-era talented woman Lu Xiaoman’s unfinished work with famous Shanghai School painter Le Zhenwen’s completion, sold for 1.1 million yuan at Yunduo Xuan’s 30th anniversary auction, marking the world’s first successful auction of an AI landscape painting.

Left: Le Zhenwen’s completed work; Middle: Lu Xiaoman’s unfinished work; Right: Baidu ERNIE-ViLG’s completed work (Source: Zhejiang Literature and Art Official Account)

As various AI creation tools emerge and AI artworks gain recognition, who exactly is the artist? Can these still be considered art? How should we view the relationship between technology and art, and how should we position ourselves?

Despite human doubts, technological development driven by capital makes AI’s intervention and significant position in the art field inevitable. While “the death of art” remains a concern for a minority, more people worry about whether humans will ultimately be replaced by artificial intelligence. Furthermore, will humans lose imagination before AI, and is art moving away from humanity?

The author believes that those who proclaim “art is dead” or reject AI are remarkably similar to their predecessors who opposed photography 150 years ago — 19th-century French poet and art critic Charles Baudelaire considered photography “art’s most fatal enemy.” In the 20th century, many artists similarly refused to use digital editing tools and computers. They deeply believed that humans were the essential source of art, and technology would deprive humans of control over artistic expression. However, even brushes, paints, and paper are products of technological progress; humans have never developed without technology.

Art exists even in creation completely without human participation. The progression of art history differs from scientific history — it doesn’t present a single progression but moves toward the richness of “no first place in literature,” which is the significance of Duchamp’s “Fountain” — as an artwork, “Fountain” marked the beginning of traditional art moving toward broader richness. Subsequently, art became determined not only by the work’s form but also by the creator’s intention and audience understanding.

Left: “Fountain,” Marcel Duchamp, 1917; Right: Face-shaped water-ground stone, circa 3 million years ago (Source: Wikipedia)

Artist Sol LeWitt said, “The idea becomes a machine that makes the art.” Perhaps ideas have always been machines making art, and modern art has merely reawakened people’s consciousness. A face-shaped water-ground stone unearthed from Makapansgat, South Africa, dating back about 3 million years and currently collected in London’s Natural History Museum, is a naturally water-shaped, uncarved reddish-brown jasper about 6 centimeters wide. One can imagine that when our ancestors picked up such a stone from the riverside, carefully appreciated it, and brought it home to collect, art quietly emerged. This means humans had begun to recognize their group’s facial features and could observe and associate with the images of other things. Artworks don’t necessarily need to be created by humans, but they must record human will and activity that words cannot convey.

While artificial intelligence sacrifices intelligence — originally considered its goal — in pursuit of computational efficiency, it will increasingly become an effective tool rather than a subject with will and thinking ability.

In the “Space Opera Theater” series, although the materials come from the vast database learned by artificial intelligence, the inspiration belongs to Allen — he spent 80 hours and underwent over 900 iterations, relying on editing an enormous vocabulary to complete the painting. In some ways, he should be called a poet rather than a painter. In fact, even using the same vocabulary, it’s unlikely to generate an identical “Space Opera Theater,” as AI receiving the same description can generate different results.

Baidu’s ERNIE-ViLG, under the operation of numerous technical personnel and art experts, went through multiple steps including AI learning, AI continuation, AI coloring, and AI poetry generation, addressing three major challenges: human-machine painting process integration, controllability, and high-resolution. The improvement of artificial intelligence aims to help AI better understand human instructions to depict the images people envision when giving commands.

Creativity Inspired by Fragility and Limitation

Writer Walter Kirn wrote in an article titled “There’s No Such Thing as AI Art”: “Artificial intelligence knows nothing of writers’ tortuous, painful inner drama. It merely writes, filters, analyzes, and then executes. It doesn’t forge ahead; it takes no risks. Only humans, we fragile species, can do that. If it’s just pressing some buttons, that’s not art, is it?”

Although artificial intelligence has greatly accelerated the production and elimination speed of artworks, the values of people growing up in the digital age are also rapidly changing. We view material scarcity, ownership, and possession differently. In the future, AI art will become commonplace, and people will pay more attention to the setbacks and hardships humans encounter in the creative process, and the inspiration and wishes they express.

However, people using AI for creation are no longer equivalent to those using brushes or pressing shutters. Artists can no longer focus solely on the scenes before them and the thoughts in their minds but need to find and learn ways to communicate and cooperate with AI.

Jason Allen, in a media interview, stated: “People need to ‘overcome their denial and fear of technology.’ Artificial intelligence is a tool, like a brush — without people, there is no creativity.”

“Space Opera Theater” creator Jason Allen

The continuous advancement of artificial intelligence to satisfy human imagination about art and design needs is itself the most valuable aspect of AI art. Just as the art evolved from the statue of Akhenaten in ancient Egypt to the Classical Greek Spear-bearer, humans’ ever-changing expressive desires and creative wisdom are manifested in artistic evolution. If we cannot imagine a future without technology yet worry about the so-called “tragic spring” it brings, how can we demand that art isolate itself from technology?

Perceptual Enhancement is AI’s Opportunity for Art

Why should we use artificial intelligence for art? Is artificial intelligence merely a tool? Against the backdrop of artificial intelligence development and metaverse scenarios, human nervous systems and sensory organs have been greatly extended. In daily life, people usually follow Aristotle’s “Five Senses” as a criterion, but recent neurological research points out that balance, thermal sense, and pain sense are also extremely important.

In the creation and appreciation of metaverse art, the main relied-upon senses are currently vision and hearing. Although over 80% of information is processed through vision, in perceiving external stimuli, visual response sensitivity is far lower than auditory. Among the five senses, smell deserves extra attention. Its working system is very special — it’s the only one of the five senses that doesn’t pass through thalamic neurons but directly enters the primary olfactory center of the brain. This center called the olfactory bulb, is located behind the frontal bone between the eyebrows and directly connects with the amygdala and hippocampus. The former evolved from brain blocks specifically responsible for monitoring chemical substances, while the latter mainly stores long-term memory — this explains why people always recall scenes related to a smell when encountering familiar odors. Since smell is not humans’ preferred communication method, it hasn’t been widely applied in human-computer interaction and immersive environments, but this also means it has broad development prospects.

In artificial intelligence and digital media art creation, various media stimulate all senses of the experience. Although the experiencer’s various senses will play their respective roles, in actual situations, they don’t operate independently but often complement and cooperate. In biology, this situation is called cross-modality (also translated as multimodality), meaning the integration or fusion of two or more sensory technologies, typically a characteristic of immersive environments. In layman’s terms, this is what we commonly call “synesthesia.” The subject’s multi-sensory experience becomes a super-sensory experience because their vision is enhanced by media that serve as extensions and transcendences of human senses supporting entry into the digital art world.

Artificial intelligence art will continuously use visualized scenes to question the essence of existence, perception, and the world, from macro to micro and back to macro. In the constant switching of dimensions, human perceptual systems continuously improve and construct the foundation of the artificial intelligence art era.

Artificial intelligence art is human “self-duration,” the existence and development characteristics of phenomena such as self, consciousness, and life. The changes and development of these things are continuous and indivisible processes — in this process, later states contain earlier states. As Bergson’s “duration” theory presents, “In duration, past and present become one, and continue to create… new things together.”

◼ View this article in Medium —>