There is much discussion and debate about the ethical implications of AI in creating artwork, with some arguing that AI art steals from artists. Meta faced significant criticism when they updated their community guidelines to include using platform images, including artists' work, to train their models. Today, I will explore how an AI model trains using images and generates artwork, and where the line between inspiration and theft is drawn. Additionally, we will look back at the history of art and philosophize about whether AI-generated art can truly be considered art.

What does an AI do with these images?

Many people are confused about how AI uses your art, but it helps to think of it similarly to how humans learn to create art. We observe other artists, learn different patterns, understand composition, and see how colors interact. Based on these observations, we develop our own style and methods.

AI operates in a similar way: it uses a large dataset of images, categorizing them by styles, themes, semantics, and more. Unless a user employs a very specific and limited dataset of an artist, or prompts the AI to recreate something in a particular artist's style, the AI model uses your artwork much like an artist would use it as a reference.

However, the biggest difference is the perceived "lack of soul" in computer-generated images. While AI can generate art based on data, people claim it lacks the emotional depth and personal touch of human-created art. That said, those who put significant effort, creativity, and vision into their prompts can achieve results that challenge this notion. This critique isn't unique to AI art; modern and abstract art, like Rothko's works, also face criticism for seemingly requiring less skill. Yet, creating an image that provokes emotions and resonates deeply with viewers remains a challenge for all artists, regardless of the medium.

So how does it generate an image?

How does AI image generation actually work? Essentially, it involves training neural networks on vast datasets of images so they can learn patterns, styles, and relationships between different visual elements. Generally, there are two widely used types of neural networks for this purpose: Generative Adversarial Networks (GANs) and Transformer-based models, like DALL-E.

The former consists of two networks: a generator that creates images, and a discriminator that evaluates them. This back-and-forth improves the quality of the generated images. Transformer-based models, on the other hand, utilize a step-by-step approach: they predict and create images pixel by pixel, based on the input description.

Both these types of models don’t just spit out replicas of the images from their datasets. Instead, they learn to blend and combine various elements from a multitude of images to come up with something entirely new and unique.

‍

‍

Will AI change our view of art?

When I think about AI, I like to compare it to the invention of the camera in the 19th century. The camera had a significant impact on the art world, as art was primarily used to capture moments before its introduction. The arrival of the camera changed that dynamic: some artists embraced photography as part of their artistic work, while others were initially skeptical or resistant, viewing it as a mechanical process that lacked the creative soul of hand-crafted art.

One of the common concerns back in the day was the fear that the technical precision of photography might render the skill and craft of painting obsolete. Additionally, the rise of photography studios offering portraits at significantly lower prices, stirred up economic competition. This scenario bears a striking resemblance to the current debate over AI-generated art, and I think it's intriguing to observe these parallels.

Why? Because over time, photography carved out its own place as an art form. Artists began to explore the unique capabilities of photography, focusing on what it could do that painting couldn’t. This led to the emergence of artistic movements such as Impressionism, Cubism, and Abstract art, which emphasized abstract pieces and capturing movement.

Moreover, the invention of the camera made it easier for non-artist to capture moments. People who couldn't afford painted portraits could now have photographs of themselves and their loved ones. In a similar way, AI helps people create visuals for presentations or mockups. It doesn’t replace traditional art, as these individuals might never have the skill or resources to create such work themselves. Instead, it helps them better visualize their concepts. AI, much like the camera, democratizes the creation of visual content, making it accessible to a broader audience.

I strongly believe AI art will stay with us and grow as a tool that artists can use to create amazing pieces. It won’t wipe away creativity or artistic expression, but it will elevate those who struggle with the technical aspects of traditional art and challenge other artists to reinvent their styles and artwork. Artists like Refik Anadol have been pioneering AI art for years, showcasing how the technology can be used to produce stunning and innovative works.

‍

‍

AI as a tool for art theft

Of course, artwork does get stolen. When we look at stolen artwork or plagiarism, it’s often the user who abuses the system. There are two main categories: copying existing work and plagiarism. Copying involves taking someone else's work and claiming it as your own by removing watermarks, tracing, or other methods. Plagiarism, on the other hand, involves using a specific artwork or the style of an artist to create something very similar. This isn’t a new concept. Just as with Adobe tools, AI might make it easier to steal certain art, but the blame should be placed on those who misuse the tools, not on the tools themselves.

Still, we definitely need better ways to track and stop this kind of misuse. Platforms that host AI-generated content should put stricter rules in place and use more advanced algorithms to spot and flag instances of plagiarism. More transparency, like clear labels on AI-generated content and public access to AI training datasets, can help identify and prevent art theft.

We’re already seen some moves in this direction. For example, Adobe’s Content Authenticity Initiative can mark images that have been edited or created using AI tools. Similarly, social media platforms like Instagram allow users to label their content as AI-generated. However, these measures still rely heavily on users' goodwill, as AI-generated metadata can be easily removed, and user action is required to properly label the content.

Looking at other tech solutions, blockchain's transparency and immutability make it a valuable tool for establishing and verifying ownership records. This can really help resolve disputes and protect intellectual property. (Silicon Valley Innovation Center) (SpringerLink).

By boosting these tracking and transparency efforts, we can better protect artists and make sure AI stays a tool for creativity and not a means of exploitation.

Conclusion

In conclusion, AI art represents a revolutionary shift in the art industry, challenging our creativity and making artistic creation more accessible and mainstream for everyone. While there are still pitfalls, including the need for systems and laws to better protect artists, AI is rapidly progressing and will continue to evolve. Despite its current blind spots, such as biases (check out our blog about this). With the right ethical guidelines and protective measures in place, AI will enhance the artistic landscape, fostering innovation and expanding the boundaries of creativity.

Wrapping up, AI art marks a groundbreaking shift in the art industry, challenging our creativity and making artistic creation more accessible and mainstream for everyone. Despite some challenges, such as the need for better legal protections for artists, AI is rapidly evolving. And I’m curious to see what’s next.