In the summer of 2022, a ground-breaking AI tool named DALL-E took the internet by storm. DALL-E is an image generator that creates images based on text inputs. The system is driven by Generative AI (GenAI), a form of artificial intelligence that absorbs vast datasets of images, which it then uses to generate new content. For example, you could ask DALL-E to create a picture of an astronaut riding a horse, and it would! The images it created were often surreal or hilarious. For a while, people had a lot of fun coming up with ideas for AI to generate.
In the few years since DALL-E became accessible to the public, GenAI platforms have become far more powerful. The systems generate clearer and more detailed images in significantly less time. Despite its progress, only some are happy with this new technology, and there are good reasons to question its quick rise to prominence.
Many artists, especially those who do commercial work, have been critical of GenAI. Their livelihoods depend on creating engaging art, so it may seem natural to think of this as a welcome new tool, but questions arise when artists consider its use.
Is creativity a good indicator of intelligence?
GenAI tools like DALL-E, Midjourney, and Stable Diffusion are producing increasingly polished and accurate images based on user prompts. However, some argue that its “intelligence” stems primarily from pattern recognition within vast datasets. While human creativity is fueled by a complex interplay of emotional depth, personal experiences, and unique perspectives.
Image: freepik (Generated by Midjourney 6)
AI art or authorless art?
Created by the French collective Obvious, Portrait of Edmond de Belamy (2018) is an AI-generated portrait produced using a Generative Adversarial Network (GAN) and trained on WikiArt’s vast database. Its sale at Christie’s
auction in 2018 for a staggering USD 432,500 ignited debates on artistic authorship and the value of art in the AI era.
Source: Obvious
When people think of art, they usually think of pencils, sketchbooks, paint, and canvases. As cameras became available in the 19th century, there was a general fear that photography would replace painting. The famous French painter Paul Delaroche famously proclaimed, “From today, painting is dead!” While many artists added photography to their toolkits, painting continued to flourish. Photography became a medium in its own right; another tool to express the artist’s vision.
Lesson from CEO
Upon the launch of DALL-E 2, OpenAI CEO Sam Altman addressed its societal impacts. He acknowledged the potential disruption to the job market for illustrators, urging those most negatively impacted to use the technology and take ownership by providing data for AI training. He also emphasised the need to educate the public that images online might be fake and should not be blindly trusted.
Source: Sam Altman / X
WHILE THE “SCRAPED” DATA IS “PUBLICLY AVAILABLE,” THAT DOESN’T MEAN IT IS IN THE PUBLIC DOMAIN.
More recently, digital art programmes have encountered similar scepticism. These programmes enable people to paint and draw on a computer, offering advantages over traditional art, such as greater control and the ability to edit freely. There is a misconception that digital art is created by pushing a button and the computer does all the work. However, using a computer programme to paint is essentially the same as traditional painting, albeit with a different medium. You paint on the computer screen instead of a palette. It’s like switching from a grand piano to an electric keyboard; the technology creating the sound is different, but the hands creating the sounds are the same. Artistic skills transfer from traditional to digital mediums, and the choice of medium is up to the artist.
While GenAI might seem like another tool, the difference is in the automation. There’s no skill involved. You type in a prompt, push a button, and receive an image. Users have little to no ability to edit or control the output. While artists might be best suited to harness these capabilities, they tend not to use them because these systems are trained on artwork sourced from the internet. This practice raises serious concerns because while the “scraped” data is “publicly available,” that doesn’t mean it is in the public domain. This leaves the platforms and their users at risk of copyright claims. Moreover, concerns go beyond copyright concerns, as the datasets used to train GenAI systems can also contain private images, pornography, and personal medical records. All this has left many artists hesitant to adopt a tool developed that uses resources without permission, which may include their own work and that of their peers.
Salmon fiasco
Social media erupted in laughter when a GenAI image generator, tasked with creating a photo of “salmon in the river,” delivered images of salmon fillets swimming instead. This hilarious blunder highlights potential quality issues with GenAI’s image generation, raising concerns about its reliability for professional applications.
Source: Bored Panda
The nature of these systems has fostered a legal debate as courts are being asked to determine whether this data scraping violates copyright laws and if AI developers should have to pay to license this data. GenAI is dependent on this data. It needs an incredible amount of data to work and the only way to improve it is to continue feeding it more. Despite the critical nature of this data to their systems, AI companies appear reluctant to pay for the valuable data they profit from.
It’s crucial to understand that GenAI is not intelligent. Instead, it predicts the most likely answer based on its dataset, like an advanced form of auto-correct. It cannot learn or draw conclusions. It can only use the information in its dataset to create facsimiles, but doing so can cause Generative AI to produce erroneous outputs that can lead to potentially damaging inaccuracies.
There’s also potential for harm from the use of these platforms. This can come from people using them to replace artists, and they can also be used to spread misinformation.
GenAI cannot effectively replace a skilled artist. Some may disagree, but the people who typically make such claims tend to lack experience with professional artwork. Worse still are those in positions of power who would normally hire artists but see GenAI as a cost-saving opportunity.
Professional artists collaborate with clients to fulfil their visions. Artists need to communicate that vision effectively to the intended audience, which necessitates the ability to adjust and edit the final artwork. GenAI cannot do this due to its limited ability to edit and the variability of its outputs — if you repeat a prompt, you will receive a completely different image. The inability to consistently deliver on a vision with quality work makes these platforms unsuitable for professional work.
THE INABILITY TO CONSISTENTLY DELIVER ON A VISION WITH QUALITY WORK MAKES THESE PLATFORMS UNSUITABLE FOR PROFESSIONAL WORK
On top of that, the lack of intelligence allows GenAI to generate images with logical errors, such as humans with incorrect numbers of fingers or strange background elements. While artists can edit these images, the final image is often dissatisfying. Rather than working towards the client’s vision, they’re working to remove problems. Worse yet, correcting AI images can be more difficult than just hiring an artist to do the work. To create high-quality, professional work, why not just hire an artist?
GenAI is already having adverse effects on artists’ careers. Greg Rutkowski, a Polish digital artist who has worked on properties like Dungeons & Dragons and Magic: The Gathering, is a good example. His work is in a dataset used to train many GenAI models. By September 2022, Rutkowski’s name was used as a prompt over 93,000 times — on a single platform — generating a flood of images that resemble his paintings. Because of this, online searches for Rutkowski’s name yield these AI-generated images alongside his own work. This makes it difficult for fans and potential clients to distinguish Rutkowski’s genuine creations, which affects his livelihood. It’s a disheartening scenario where a program trained on the artist’s work benefits from his creative contributions while simultaneously being pitched as a method to replace him. Still, proponents of GenAI prefer its use over hiring artists because the results are passable at a glance, and it is relatively cheap. Paradoxically, GenAI needs to consume more art while leaving less work for the artists it is designed to replace.
Prompt: In style of Greg Rutkowski
In response to artist Greg Rutkowski’s opposition, Stability AI, creators of the popular image generator Stable Diffusion, removed Rutkowski’s work from their training data. However, the opensource nature of Stable Diffusion allowed the online community to develop a tool that mimics his style, effectively reintroducing his “work” into the AI art scene — regardless of his wishes.
Source: Greg Rutkowski / X
PARADOXICALLY, GENERATIVE AI NEEDS TO CONSUME MORE ART WHILE LEAVING LESS WORK FOR THE ARTISTS IT IS DESIGNED TO REPLACE.
Another troubling concern arises from the potential to use GenAI to create harmful images, including ones attributed to an artist without their involvement or endorsement. Some of these systems allow the creation of explicit content that can use a person’s likeness to create revenge porn. Disturbing evidence exists of a training dataset with links to child sexual abuse material and an AI creation of an actor that was used to read pornographic material. Also, safeguards to prevent people from using GenAI to create and spread misinformation are up to the system’s creators, and many have none in place. Proponents of GenAI seem to be labouring under the philosophy of “ask for forgiveness, not permission”, but what will happen if the damage they inflict is irreversible? These questions and concerns highlight the need for efforts to mitigate the misuse of this technology.
Uncanny valley in Cannes
In February 2024, Air Canada was ordered to pay damages to a passenger misled by its virtual assistant into buying a full-price ticket. The airline initially refused a refund, arguing the chatbot was “a separate legal entity responsible for its own actions.” The case highlights the liability implications and risks of businesses relying heavily on AI.
Image: Shutterstock
Artists’ concerns go beyond the potential for unethical development, adverse effects on their livelihoods, and the creation of misinformation. There is also a growing realisation of how many people view artists and their work. This can be seen in those who claim they want to create a tool to help artists while actively seeking to replace them. They claim to admire art but don’t hesitate to feed it into a machine meant to strip artists of their livelihoods. It appears that a vocal portion of society desires quick and cheap art while lacking concern for those who make art for a living.
A striking example came in the form of an image generator that imitated the art of Kim Jung Gi, a respected artist who tragically died of a heart attack in October 2022. Just three days after his passing, a developer created a machine to churn out copies of Kim Jung Gi’s art. Although it was ostensibly done to pay homage to a great artist, many in the artistic community saw it as insensitive and heartless as Kim Jung Gi could not consent to his work being used like this. Behaviour like this makes artists question how they can continue creating if their work risks being misappropriated, fed into a machine, and used to generate an everincreasing wave of content without their consent or compensation.
PROPONENTS OF GENERATIVE AI SEEM TO BE LABOURING UNDER THE PHILOSOPHY OF “ASK FOR FORGIVENESS, NOT PERMISSION”, BUT WHAT WILL HAPPEN IF THE DAMAGE THEY INFLICT IS IRREVERSIBLE?
Generative AI offers incredible potential. We should be hopeful about its future, but the current approach leaves artists distrustful of GenAI and its proponents. These platforms are unlike any artistic tool that has come before, and they foster many important questions for artists. How can they be expected to use something that takes their work and seeks to replace their jobs? How do they communicate their concerns to the AI developers who have chosen expediency over morality while refusing to compensate the artists whose work they have appropriated? How can artists avoid the use of their work to promote injustice, violence, or wars? How can they make a living in these hostile circumstances? And above all else, how do they contend with a society that may not even care? We must answer these questions if we want GenAI to be ethical and fair. ∞
Guardians of creativity
Developed by a team of computer science researchers at the University of Chicago's SAND Lab, Nightshade is
a free tool designed to protect artists from copyright infringement by GenAI programmes like Midjourney and
Stable Diffusion. The tool poisons AI training data by introducing subtle errors into images that cause AI models to misinterpret the data, making it harder for them to scrape images without consent.
Source: University of Chicago (An example of “poisoned” data image generations in Stable Diffusion.)
REFERENCES
- @cadmusfly. “Trying To Make A Horse Riding An Astronaut In Midjourney.” Neural Networks Dream in AI, Tumblr, 5 Jan 2023. https://teuthisdreams.tumblr.com/post/705571249852694528/tryingto-make-a-horse-riding-an-astronaut-in.
- David, Emilia. “AI Image Training Dataset Found to Include Child Sexual Abuse Imagery.” The Verge, 20 Dec 2023. https://www.theverge. com/2023/12/20/24009418/generative-ai-image-laion-csam-googlestability-stanford.
- Davis, Wes. “AI Companies Have All Kinds of Arguments against Paying for Copyrighted Content.” The Verge, 5 Nov 2023. https://www. theverge.com/2023/11/4/23946353/generative-ai-copyright-trainingdata-openai-microsoft-google-meta-stabilityai.
- Deck, Andrew. “AI-Generated Art Sparks Furious Backlash from Japan’s Anime Community.” Rest of World, 27 Oct 2022. https://restofworld. org/2022/ai-backlash-anime-artists/.
- Edwards, Benj. “Artist Finds Private Medical Record Photos in Popular AI Training Data Set.” Ars Technica, 21 Sept 2022. https://arstechnica. com/information-technology/2022/09/artist-finds-private-medicalrecord-photos-in-popular-ai-training-data-set/.
- Ellery, Simon. “Fake Photos of Pope Francis in a Puffer Jacket Go Viral, Highlighting the Power and Peril of AI.” CBS News, 28 Mar 2023. https://www.cbsnews.com/news/pope-francis-puffer-jacket-fakephotos-deepfake-power-peril-of-ai/.
- Felin, Teppo, and Matthias Holweg, Theory Is All You Need: AI, Human Cognition, and Decision Making. 24 Feb 2024. https://papers.ssrn.com/ sol3/papers.cfm?abstract_id=4737265.
- “From Today, Painting Is Dead: Early Photography in Britain and France.” Barnes Foundation, 2019. https://www.barnesfoundation.org/ whats-on/early-photography#:~:text=On%20first%20seeing%20a%20 photograph,in%20the%20mid%2D19th%20century.
- “Greg Rutkowski: Portfolio”. ArtStation. https://www.artstation.com/ rutkowski. Accessed 6 Jun 2024.
- Heikkilä, Melissa. “This Artist Is Dominating AI-Generated Art. And He’s Not Happy About It.” MIT Technology Review, 16 Sept 2022. https://www.technologyreview.com/2022/09/16/1059598/this-artistis-dominating-ai-generated-art-and-hes-not-happy-about-it/.
- Kelleher, Kara. “Revenge Porn and Deep Fake Technology: The Latest Iteration of Online Abuse.” Boston University School of Law, 10 Aug 2023. https://sites.bu.edu/dome/2023/08/10/revenge-porn-and-deepfake-technology-the-latest-iteration-of-online-abuse/#:~:text=And%20 the%20rise%20of%20Artificial,create%20pornographic%20images%20 and%20videos.
- Litchfield, Ted. “Baldur’s Gate 3 Actors Tear into AI Voice Cloning: ‘That Is Stealing Not Just My Job but My Identity.’” PC Gamer, 14 Apr 2024. https://www.pcgamer.com/gaming-industry/baldurs-gate-3- actors-tear-into-ai-voice-cloning-that-is-stealing-not-just-my-job-butmy-identity/?utm_source=twitter.com&utm_medium=social&utm_ campaign=socialflow.
MELISSA CHIU
Melissa Chiu is a 3D artist and Illustrator currently working in the entertainment industry. She graduated from the Rochester Institute of Technology with a degree in 3D Digital Design, and has worked in both advertising and video games.
JULY 2024 | ISSUE 12
NAVIGATING THE AI TERRAIN
