The rapid emergence of AI tools throughout 2022 has captivated many, including myself. Platforms like MidJourney, Stable Diffusion, DALL·E, and ChatGPT utilize machine learning models—which I'll refer to as "AI" for simplicity—to produce distinctive images or text based on user-defined prompts and parameters.

When successful, these tools evoke a sense of wonder, making the process seem deceptively simple.

A Cultural Divide Since DALL·E and MidJourney gained popular recognition, a heated debate has erupted at the crossroads of art and technology. Central to this discourse is the concern over the adverse effects of AI-generated images on professional artists. The discussion has become highly polarized, resembling a clash between "Artists and Tech Enthusiasts," reminiscent of last year's NFT discussions, which featured many overlapping voices.

While I don’t wish to take a neutral stance, I find myself in a position of cautious optimism, albeit laden with numerous concerns. I have personally generated over 4,000 images using MidJourney—some of which I cherish, while many others fall short. However, I also critique several practices within the industry that seem troubling.

In this piece, I aim to outline several points worth deeper exploration, as well as highlight some arguments I believe to be unfounded. This won't cover every nuance of the issue, and I may delve into it further in the future.

Defining Art This is perhaps the most challenging yet essential starting point. Attempting to define "art" can be an intriguing theoretical pursuit, but reaching a universally accepted definition seems futile and often exclusionary.

I think we can agree that art holds various meanings for different people and is fundamentally subjective. Disagreements over whether a banana affixed to a wall qualifies as "art" are perfectly acceptable.

Art can be created with a specific intention or none at all, and it can be appreciated from various perspectives. It might be visually appealing or unappealing. It can involve meticulous craftsmanship or simply capture a fleeting thought. It may even embody randomness. I believe that art need not be born from a clear purpose or require extensive effort to be valid.

Many, though not all, seem to accept that AI-generated images can be categorized as art in certain contexts, but the more contentious issue arises in determining who qualifies as the artist or author. Is it the human, the AI, both (and how), or neither?

Authorship and Attribution I don't identify as an artist; rather, I am a designer by trade. While I have created things I consider artistic, I refrain from labeling myself as a musician, photographer, or chef despite having dabbled in those areas.

I generate images using MidJourney, yet I do not see myself as an artist.

A compelling argument exists suggesting that AI-generated imagery warrants distinct treatment regarding authorship and attribution. There is no benefit in misleading others about the extent of one’s involvement in creating an image.

For some time, I've utilized AI-generated images as editorial visuals in my articles instead of relying on free stock photos. I find enjoyment in the process and appreciate the imperfect aesthetics.

I have explored various ways to credit these images transparently and fairly. Conversations on the topic have yielded a spectrum of opinions based on the terminology used: - "Some art I made" feels disingenuous. - "An image I generated" is an improvement but still lacks transparency without context. - "An image Midjourney generated based on my prompting and iteration" comes off as overly mechanical and unnatural.

As with any shift in language, such as changing from "guys" to "folks," the initial awkwardness will fade over time.

Fortunately, written credits don’t need to sound conversational. The phrasing I'm currently using is: > Image by Midjourney, directed by the author.

In previous iterations, I stated, “Image by the author, generated with Midjourney,” suggesting that Midjourney is merely a tool, akin to saying, “Image by the author, sculpted in Zbrush.” I moved away from this language because I believe these roles are not directly comparable. Art direction is a more fitting analogy, especially as tools evolve to allow for more nuanced image modifications.

The dialogue must persist, and individuals should experiment with what resonates with them while ensuring they don't diminish others' contributions. Regardless of the outcome, labeling a piece simply as “By [your name]” currently seems unethical and should be discouraged, regardless of one's enthusiasm or skepticism about the field.

Categorization Continuing with this line of thought, I believe that AI-generated imagery deserves specific categories, tags, or sections on platforms displaying art, especially in award contexts. The notion of pitting artworks against each other in a competitive format to deem one superior to another is perplexing, yet it ultimately supports artists. The artistic community significantly enriches society, yet this value often goes unrecognized in terms of compensation and visibility.

We should not have to revisit the debate ignited by the Midjourney piece that earned Jason Allen recognition in the Colorado State Fair art competition. Establishing an explicit "AI Art" category would be a logical starting point.

This practice is already common in photography and occasionally in digital art. By doing so, we allow those who care about AI art to engage with it while enabling others to bypass it if they choose.

Market Saturation Another major concern is that images created with prompts referencing artists’ names (like "in the style of Greg Rutkowski") can proliferate online, drawing SEO traffic away from the original artists and causing confusion for those searching for authentic works by Greg Rutkowski, a significant income source for many creators.

> “It’s been just a month. What about in a year? I probably won’t be able to find my work out there because [the internet] will be flooded with AI art. That’s concerning.” — Greg Rutkowski, interview with MIT Technology Review

Ethical Data Use In my view, the most pressing issue revolves around the ethics of data sourcing. Currently, numerous models either wholly or partially rely on datasets like LAION5B, which compiles vast arrays of images and corresponding text descriptions scraped from alt-text and web links by a non-profit entity known as Common Crawl. These datasets are intended for research purposes only, not for commercial use.

Given the scale of these datasets, they inevitably contain a significant number of copyrighted works, including those of living and active artists.

The images that Common Crawl has harvested, which LAION has made available, were originally published by their creators with a wide range of intended uses. However, it's safe to assert that "training machine learning models to eliminate jobs" is not a common goal.

Artists often express that AI-generated art constitutes plagiarism or theft of their original works, likening it to “searching Google Images” and misrepresenting the results as one’s own. While I find this comparison overly simplistic and inaccurate regarding how machine learning generates images, it doesn’t substantially alter the solution.

I believe the only truly ethical approach to AI-generated imagery is to develop an open training dataset to which individuals must actively consent. The financial model for this could include small compensation for initial submissions, alongside a royalties system for prompts that explicitly mention artists' names (with significantly higher royalties for commercial uses). This could be structured under a CC-BY type license specifically for AI training datasets, possibly indicated through image metadata.

Such a system might enhance a more curated open dataset that excludes data from platforms like Behance, Dribbble, ArtStation, and DeviantArt while still being able to learn general concepts like “what a human looks like” or “what a tree looks like.”

Note: ArtStation has faced criticism from artists recently for insufficiently protecting their interests. In response, ArtStation initiated a “NoAI” tagging system, allowing users to opt out of having their artwork used in training models. However, this opt-out mechanism alone is not sufficient, and the community is making that clear.

In Conclusion This domain is evolving rapidly. Technological progress can be exhilarating (especially when it feels almost magical), but it can also have detrimental effects on people and entire industries. In some cases, such as transitioning from fossil fuels to renewable energy, technological advancements can yield predominantly positive outcomes. The benefits of addressing climate change may outweigh the human toll of job losses.

However, this narrative differs significantly. The individuals suffering are primarily artists, and replacing them en masse with machines represents a net loss for humanity. We should seek ways to utilize this technology to augment creative processes, fostering new aesthetics while keeping humans—along with their imagination, intent, and complexity—at the forefront.

In summary, I encourage individuals to consider the following: - Advocate for the investment in an alternative, ethical dataset among companies like Midjourney, LAION, and StabilityAI. If you work in this field, help build it. It's a fascinating challenge. - Correctly credit and label AI-generated imagery in non-commercial contexts. Avoid exaggerating your role in the creation. - Avoid using artists' names as prompts. Unconditionally. - Practice active listening. Show respect when confronted with criticism and provide constructive feedback with good intentions. It’s insufficient to merely state, “I don’t see an ethical issue here.” If others express valid concerns, take the time to listen and educate yourself further. - Explore ways to help individuals transition from AI art to traditional art. Similar to how Guitar Hero inspired many to learn the guitar, there exists an opportunity to guide people from AI generation to mastering traditional techniques. Analyze the aesthetics of AI-generated images and consider how one might replicate similar outcomes using different methods. - Support artists whose work you appreciate. Purchase prints, share their creations, and commission their services. Back organizations like the Concept Art Association, which advocate for artists' rights in the face of AI technologies.

Thank you for reading. If you found this discussion engaging, please consider subscribing to Clip Content for weekly insights on design systems, technology, leadership, and more delivered directly to your inbox.