In Defense of Generative AI ("AI Art")

Introduction

For simplicity, I refer to generative AI images as "AI art." Whether these qualify as "art" in the traditional sense is discussed below.

Ethical Considerations in AI Art ("AI Art Steals From Real Artists")

This objection hinges on two key issues: the use of unauthorized art in training data and the replication of human artists' styles by AI.

The Ethics of Using Unauthorized Training Data:

Understanding generative AI's fundamentals is crucial. It's not a mere compilation of its training data. Instead of simply cutting and pasting fragments of images like a scrapbook, a generative AI, particularly one utilizing diffusion algorithms, synthesizes new images by learning patterns and principles. Effective training demands an understanding of artistic elements such as perspective, color theory, light and shadow, lighting, and human anatomical proportions.

Our grasp of how diffusion models encapsulate these elements is limited, but it's evident they develop algorithms reflecting key artistic principles. Thus, both humans and AI derive their image-creation techniques from observational learning. (Although humans also learn in a variety of other ways.)

In essence, generative AI is a transformative process. It doesn’t replicate source material directly, barring training bugs. Consequently, consuming unauthorized art through AI shouldn't be deemed unethical, paralleling the freedom humans have to learn from existing art without needing explicit permission.

The Ethics of Imitating Human Artists' Styles:

Ethical evaluation of derivative works necessitates clear criteria. Setting aside legal aspects, let's consider the ethics:

Claiming a direct copy or a photo of someone’s work as your own constitutes fraud. The same applies to copying a work manually or conceptually without acknowledgment.

If you copy a work by hand and present it as your own, this is also fraud. If you take the concept of the work and present it as your own, that is also fraud. But what about mimicking a style without claiming original creation? This act is ethically sound; it neither deceives nor harms the original artist. Imitation isn’t theft. No moral or legal right exists to prevent others from learning and drawing inspiration from an artist’s style.

Distinguishing between imitation of a specific work (unethical without credit) and a style (ethical with attribution) is crucial.

Aesthetic Critique of AI Art ("AI Art Is Ugly")

If AI art was irredeemably bad, there would be no need for debate. But the reason this is a controversial topic is that many people do like AI art, or at least find it apt for commercial use. Criticizing these tools based on current limitations overlooks their potential and evolutionary trajectory.

AI art has evolved from rudimentary beginnings — at first, it was a wonder that software could synthesize from prompts at all. Its development is uneven, with certain visual aspects proving more challenging than others. Perhaps only an AI with human-level intelligence can fully explore the depth of visual narrative, and more primitive tools will always be derivative. However, within their competencies, tools like DALL-E and MidJourney have demonstrated remarkable capabilities:

Take a look at https://midjourney.com/showcase & https://mid-journey.ai/showcase/

Since tools like DALL-E are now public and free, anyone can try creating images, and many non-artists do. Those creating images for marketing purposes often combine a lack of understanding of the tools' limitations, low standards, and low quality control and distribute "ugly" images. This shouldn't be a reflection on the tools themselves, just as poor drawings don't diminish the value of pencils.

The most criticized limitations of AI art, such as inadequate representations of physics or anatomy, are rapidly improving. For instance, focusing on "long tail" problems like hand positioning, which require intricate models, has already enhanced AI capabilities in these areas.

AI Art and the Future of Human Artists (“AI Art Squeezes Out Real Artists”)

The rise of AI art does threaten traditional artist roles, a trend likely to intensify. This is unfortunate, but on the whole, I believe these tools will have a positive impact on both creators and content consumers.

As these tools improve, they will likely become integral to most content creation within a decade, not replacing but augmenting human creativity. Artists must choose to adapt or risk obsolescence.

I anticipate a division of labor emerging in creative roles. The majority of creators will become storytellers, utilizing generative AI to craft works such as movies, television shows, games, and music, drawing from a diverse catalog of styles. As these tools evolve, their productivity is projected to increase by 10x, 100x, and eventually 1000x compared to the pre-AI manual processes. These creators will transcend the role of mere prompt engineers, evolving into true editors and artists. They will transform scripts and concepts into polished narratives, directing generative tools in a manner akin to the combined duties of a director, cinematographer, and editor, with natural language as their primary instrument.

Additionally, a continued economic demand for "manual" artists—those who create by hand—will persist. However, this group will constitute a small, exceptional minority, focusing not on storytelling but on the development of new visual styles. These novel styles will then be integrated into new models for storytellers to utilize.

In summary, whereas pre-AI artists were tasked with crafting both the visual style and the story, the post-AI era heralds a shift: a minority will concentrate on pioneering new visual styles and narrative techniques, while the majority will leverage an extensive array of styles to weave compelling stories.

Originality in AI-Generated Art ("All AI Art Is Derivative")

Current AI tools inherently remix existing ideas from their training data, limiting their capacity for original visual language or narrative innovation. Full narrative and artistic originality from AI remains a future prospect, contingent on achieving human-equivalent AI intellect.

Despite this, AI's value in creative processes is immense. It facilitates a division of labor between story creation and style innovation. With the bulk of demand leaning towards new stories rather than styles, AI's role in increasing creative output is significant.

Moreover, with Generative AI, artists can train the tool on new visual language they invent, then apply it at scale. I expect an explosion in films and other mediums featuring radical new storytelling techniques.

Furthermore, the contemporary world is experiencing a critical deficit of artistry in our everyday surroundings. When I draw comparisons between today's household objects, architecture, and interior designs to those from the era around the turn of the 20th century, a stark contrast becomes evident. Modern designs often appear sterile and lack the intricate aesthetics that characterized earlier periods. This blandness is particularly noticeable in our homes and offices, which seem to suffer from an acute absence of beauty and artistic expression.

The potential of AI in this context is significant, even without the necessity for wholly original works. AI-generated art can infuse our environments with the much-needed aesthetic richness and diversity that current design trends often overlook. By leveraging generative AI, we could reintroduce detailed, personalized, and diverse artistic elements into everyday items - from furniture and home decor to the broader architectural landscape. This revitalization would reconnect us with the art-infused ethos of previous centuries, where beauty and function intertwined seamlessly in everyday objects. AI, in this way, holds the promise of democratizing access to artful living.

The Essence of Art (“Only Humans Can Create Art”)

Questioning whether AI-generated images qualify as "art" misses the point. The crux of the matter lies in attribution—balancing credit between the AI's input and the originality of the prompt writer and source material. The inability to trace specific contributions within an AI model doesn't detract from the artistic value of its output. Art's impact, not the labor invested, defines its worth. Thus, if AI-generated work resonates powerfully and beautifully with its audience, it merits classification as art.

I love the images I create. I don't dispute that I deserve a minuscule fraction of the credit for these images. But there's no doubt that that there is beauty and meaning in the work -- at it will only get better from here.

Conclusion: Attempts To Ban AI Art Are Counter-productive

Avoiding or banning AI tools on any level (personal, corporate, or national) is unlikely to change the outcomes -- but it will determine the winners.

It's a given that tools boosting creators' productivity by 1000x will inevitably take the lead in content creation. Should any group or nation decide to ban or avoid these tools, they'll find themselves overshadowed by content produced elsewhere. In the pre-generative AI era, the world was abundant in untold stories but faced a scarcity of artistic talent to bring them to life. AI art addresses this by amplifying artistic capabilities, thus enabling the narration of many more stories. Any resistance to this trend will only accelerate obsolescence.

AI-empowered creators are set to far exceed the productivity of traditional methods, shifting the balance of cultural influence. The pre-AI landscape, rich in narratives but poor in artistic resources, is poised for a dramatic transformation. AI's ability to scale up artistic talent will unlock countless untold stories. Societies and groups that resist these advancements risk lagging in the global narrative sphere. Resistance is futile—it only quickens assimilation.