AI art or generative art has taken the art world by storm over the last few months and has left everyone spinning. We take a look at AI, its uses and some of the issues surrounding it.
How Did AI Art Develope?
The concept of AI (Artifical Intelligence) has been around for decades and even prompted various sci-fi movies. AI art has technically been around since the 1960s but it’s only been since 2014 that generative adversarial networks or GANs for short has been running and today makes up most of AI image creation. GANs rely on scanning images or “scraping” and building images based on the scraped data. Both Dall-e and Midjourney for example use GANs when creating images. There were a few companies that played around with GANs for a few years (including Dall-e which released a series of images in 2021) and then in August 2022 Stable Diffusion was born and made generative art free and open to development. At first, generative art was rather rough and full of issues and had some pretty funny outcomes. However, a recent comparison done on a generative work created using Midjourney shows just how far the technology has come in one year and its hard, as an artist, not to be just a little bit terrified.
Image: A image depicting Obama and Trump playing basketball produced by Midjourney only one year apart. Image courtesy Zeneca on Twitter.
Is There A Use For GANs?
The argument for GANs is there potential use as a tool in many scenarios and professions. Examples put out there have included the use of GANs in healthcare. For example a company called terenz.ai are already using GANs to produce super resolution medical imagery to better assist healthcare professionals. GANs can also be used to make simulators such as a weather simulator to assist meteorologists. There has also been a suggestion to use GANs to improve image resolution for police identification and number plate reading etc. It’s important to emphasize that all the cases for the use of GANs has suggested it to be used as a tool and not as an end product. In all of these examples people are still required to do independent tests but GANs to some extent can be a useful tool to give professionals further information albeit with a pinch of salt. The use of AI as a tool in the art world also has the potential for good and can help usher in a new digital renaissance but we need to be mindful how we use it.
Image: AI and GANs being used in medicine. Image courtesy TS2 Space.
The Issues Surrounding AI Art And The Use OF GANs
Platforms such as Midjourney and Dall-e have allowed anyone to generate art. The result has been a deluge of AI creative content hitting the internet. Artists who previously relied on platforms such as Etsy or Redbubble have in some cases had their income wiped out by a grossly oversaturated market overnight. A simple search on Youtube can give you a tonne of instructional videos on how to generate AI art, target keywords and use bots to mass upload to internet platforms. One content creator claimed to be uploading over 100 images a day! As an artist producing original content it’s honestly hard to keep up and based on upload numbers even harder to get noticed. The impact on online art platforms is also only the tip of the iceberg. A photograph produced using generative AI recently won the World Photography Organization’s Sony World Photography Awards for a piece titled The Electrician. The work was subsequently removed but it is sadly not the first time an AI created work has won an art competition. In September 2022 a person won a fine art competition in Colorado using Midjourney. A few weeks ago a cover contest announced its winner for a book called “Bob The Wizard”. On the day the winner was announced people online started questioning the use of AI tools. Sean Mauss who created the artwork initially denied the use of AI tools and even shared photoshop files online. However users from Twitter scoured Midjourney data and were able to match images to the cover. The nail in the coffin was when the midjourney username was spotted in one of Mauss’ files. Within a day Mauss had withdrawn from the competition and taken both his social media and website offline. The saddest part to the whole story is in the wake of the controversy the organizer announced they would be abandoning any further competitions. The shear amount of admin monitoring AI generated art and the issues around it have also caused similar art competitions and marketplaces all over the world to consider their positions and has no doubt had a very real impact on creators everywhere.
Several art and design industries have experienced similar issues. In February this year Netflix made a short anime film called “Dog and Boy” where the background artwork was created using generative art. To make things even worse anime has recently sky rocketed to an industry high of 18.4 billion USD yet low anime illustrators can make as little as 200 USD a month. Netflix Japan commented that the experiment to do AI background art was in response to an apparent labor shortage in the anime industry. A claim that was quickly disputed by many online. It’s also not just illustrators that are being affected. For example, there are already a handful of AI interior programs that can redesign a space in many ways and you can even select and filter styles and aesthetics. Some of these programs even learn based on existing structure and mechanical systems. When one considers the impact of AI in text as well as images, its easy to see how jobs could be threatened and possibly made redundant in the near future. According to a report published by the South China Morning Post, AI will put a quarter of people in Hong Kong out of a job by 2028.
Image: "The Electrician" created using AI. Image courtesy Boris Eldagsen.
Besides AI art taking away individual jobs and having very real implications on creativity and shaping our visual future, there are also wider concerns on the impact of AI art on fake news and society. Earlier this year AI generated photos of Donald Trump in handcuffs went viral leading to a huge spread of misinformation. Fortunately the photos were spotted as fakes but as AI art rapidly improves (the finger problem in AI photos for example is now no longer an issue) it will only get harder to discern fact from fiction. The CEO for Stability AI has already warned that the technology could “wipe out humanity”. The CEO for OpenAI, Sam Altman has even appeared before the United States Senate to warn AI can “cause significant harm to the world.”
There are other concerns as well; in particular regarding copyright and authorship in AI generated content. Due to the nature of GANs working by “scraping” the internet they have been found to have stolen millions of references that are not open source and are now offering this stolen material for free. One youtuber compared it to the beginning of napster, the theft of music and the impact on the music industry. The revelation of artists being scraped for AI has led to several class action lawsuits and a few months ago Getty Images filed a lawsuit against Stability AI claiming copyright infringement of more than 12 million of their stock images.
What Next for AI And The Art World?
While lawsuits for image theft may turn out to be successful there is some thought that it still won’t be enough to keep AI art at bay. As mentioned by a few tech experts “the genie is out of the bottle”. Adobe recently released it’s latest program “Firefly” with it’s very own take on art generating using licensed stock images and happily avoiding any copyright issues. As Adobe’s director puts it, “If I ask Firefly to draw Mickey Mouse it won’t understand what I mean”. While Firefly would likely generate an image of a mouse, it won’t be able to reproduce Disney’s intellectual property. So while particular artists and images would be protected, people would still be able to generate images with no real technical ability or understanding. So where does that leave us? As it turns out the salvation for artists is again in copyright. No, not the artist’s copyright, but AI art’s copyright or rather the lack thereof. Within the US, for example, copyright can only be granted to a human author and has been for years. For example, a painting produced by an elephant technically has no copyright. The same therefore applies to AI generated art. Earlier this year the US copyright office declared the comic book “Zarya of the Dawn” to have only limited copyright protection. While the words and arrangement of images is protected by copyright law, the images themselves (created using Midjourney) were denied of copyright.
Image: "Zarya Of The Dawn" which was denied copyright over its images. Image courtesy Kris Kashtanova.
The issue surrounding copyright has caused some companies to take a stand to protect their content and ensure copyright over their products by contractually demanding against the use of AI by their employees and sub-contractors. For example Trent Kaniuga, who counts Blizzard, Riot Games, Capcom and Epic Games as his clients, has announced that many video game companies are now contractually demanding artists don't use AI in their work. There is however some wiggle room in the laws over copyright in AI artwork. The US Copyright Office earlier this year announced that it will consider an AI-generated work copyrightable if a human can prove they themselves put a meaningful amount of creative effort into the final content. This is a similar standpoint to copyright when creating a collage from magazine clippings for example. Ultimately this is probably where AI image generating can be best used - as a tool and reference for artists rather than as an end product. It is important to note that copyright laws differ in various countries and many are still deciding where they stand in the fight over AI art. In China for example copyright law states that a “work” can have copyright. Unfortunately as of yet the definition of the term “work” does not exclude work created by AI. Reportedly, the rapid use of AI in China has led to a staggering 70% drop in jobs in the game development sector alone. In Hong Kong the stance on copyright is the author must have exercised the requisite “labour, skill or effort” in producing the work.
Where Does That Leave Us?
As artists there are already a few things we can do. “Have I Been Trained” is a website that allows you to scan your artwork to see if your work has been used to train AI. The website aims to establish norms of consent in the use of artwork in AI and gives the option to opt in or opt out of future AI training. There are also a few programmes you can use to protect your artwork against AI scraping in the future. Examples include “Glaze” and “ArtShield” which both work by placing an invisible watermark over your work that confuses AI scrapers. Finally we can also try to avoid using AI output as artwork. We need to move away from the idea of AI art as art and instead move towards the idea of AI art as a tool. While the ethical argument doesn’t always resonate with individuals, a stronger case for many will be the laws regarding copyright and the fact that, at least for now, any AI created content is devoid of copyright. So be aware that anything you create using AI could theoretically be used and sold by anyone else. Laws regarding the use of AI tools will undoubtedly change and adapt in the coming months and years. Like it or not though AI art is most likely here to stay. In what capacity AI art will be able to be used and what the future ultimately holds for AI in the art industry we will have to wait and see.