Ethical GEN-AI in Art & Design: Solutions for Artists and Creators

The foundation of any GEN-AI system lies in the vast datasets it's trained on. These datasets, comprising images, videos, and text, are the fuel that powers AI models. A critical ethical issue arises from the fact that many of these datasets are built upon the work of living artists who have not given their explicit consent for their creations to be used in training algorithms. Artists have historically built their careers and livelihoods on the unique styles and bodies of work they've developed over decades. GEN-AI tools can now replicate these styles with remarkable accuracy, often generating hundreds of images in the likeness of a specific artist. The economic benefits of these generated works predominantly flow to the software companies, leaving the original creators with little to no compensation. This situation is ethically problematic, especially when the generated art directly competes with or threatens an artist's ability to earn a living, or even damages their reputation. As with consumers of fast fashion who unknowingly fund exploitative labor and environmental damage, users of GEN-AI must recognize the implications of the technology they employ. While legality may lag behind ethical considerations, informed action is paramount.

Addressing the ethical concerns of GEN-AI requires developing models that are trained responsibly. Several approaches are emerging: 1.1 **Adobe Firefly:** While Adobe's broader business practices are sometimes debated, Firefly represents a step towards ethical GEN-AI. Its models are trained exclusively on Adobe Stock images, openly licensed content, and public domain works. This ensures that the output is clear for commercial use. Although its quality might not match that of models trained on broader datasets, it serves specific purposes effectively. 1.2 **Shutterstock TRUTH:** After a legal dispute, Shutterstock reached an agreement with Stability AI, allowing the use of its images for training. However, the output from Shutterstock's models often resembles stock imagery, indicating a potential limitation in creative flexibility. 1.3 **Exactly:** This platform offers a unique solution by enabling users to build and resell their own AI models, providing greater control and potential for monetization. 1.4 **Bria.ai:** Bria.ai stands out by using licensed materials from major stock image providers and collaborating with artists who opt in. Crucially, they implement a revenue-sharing model that compensates creators and rights-holders. Their technology can even trace the influences of specific artists in generated images, allowing for proportional benefit sharing. This approach represents a promising future for GEN-AI, where artists are recognized and rewarded for their contributions. The challenge with Bria.ai lies in its deployment, as it primarily builds models rather than distributing them directly to individual artists.

An advanced method for achieving unique and ethically sourced AI-generated art is through fine-tuning foundational AI models. This process involves retraining a pre-existing model, such as Stable Diffusion, with a curated dataset of images that reflect a specific style or aesthetic. By doing so, you effectively overwrite parts of the original model, imprinting your unique artistic vision onto the AI's output. An illustrative example is the collaboration between Domestic Data Streamers and artist Marta Ribas. They fine-tuned an AI model using Ribas's original collage work, aiming to capture her distinctive textures, color palettes, and figurative style. The resulting generations closely mirrored Ribas's artistic output, demonstrating the potential for AI to act as a powerful tool for artists to explore and expand their creative boundaries while maintaining stylistic integrity. This method offers a high degree of control and originality, allowing artists to develop AI models that are extensions of their own creative practice.

Currently, the onus of protecting artwork from unauthorized AI training falls primarily on individual creators. While systemic changes are advocated for, artists can take immediate steps: 1. **Opt-Out:** Actively opt out of all GEN-AI platforms where possible and register with services like Spawning to prevent future scraping. Ideally, an opt-in system should be the standard. 2. **Poisoning:** When uploading new material online, use tools like Glaze and Nightshade. These tools subtly alter image pixels in ways imperceptible to the human eye but disruptive to AI training processes, making replication impossible. 3. **Encryption:** Utilize platforms like KinArt that employ image segmentation and random tag swapping to interfere with AI model training. These methods aim to obscure or scramble artwork data, making it unusable for training purposes.

The rapid advancement of Generative AI presents both unprecedented opportunities and significant ethical challenges for the art and design world. This article has moved beyond simply highlighting the problems to proposing concrete solutions. By advocating for fair training models, ethical prompting, custom model development, and innovative revenue streams, we can begin to reshape the landscape of AI-generated art. The responsibility lies with all stakeholders – artists, developers, legal experts, and educators – to collaborate and build a future where AI serves as a tool for artistic empowerment and fair compensation. This collective effort is essential to ensure that technological progress in AI art is built on a foundation of respect for creators and their invaluable contributions.