Exploring the Ethics of AI Art: A Modern Dilemma
Written on
Chapter 1: The Fascination and Fear of AI Art
The realm of artificial intelligence in art is both awe-inspiring and alarming. Within this paradox lies a realm of layered discontent, primarily due to its unpredictable nature.
For example, take a moment to examine the intriguing artwork displayed above. Its unusualness likely captivates your attention, not because of its beauty, but due to its bewildering subject matter. I aimed to generate an image of a Greek goddess opening a door, inspired by Banksy, but the result was unexpectedly a dog dressed in a gown, positioned on a bathroom door. Today's AI art applications allow users to input text prompts, which then generate images—sometimes resulting in the strangest of outcomes.
This oddity isn’t entirely new. In ancient Greece, individuals would journey to the Oracle of Delphi seeking answers to their queries. However, the term "Oracle" can be misleading; it referred to a person, specifically the priestess known as the Pythia. When asked a question, she would enter a trance and provide a response from the divine that often seemed nonsensical.
This experience can be likened to my own encounter with AI. It processes vast amounts of data to create images, but the logic behind its decisions often remains elusive, a phenomenon referred to as "the black box" problem.
So, how does AI determine what to create? Where does our unexpected canine in a gown originate? The answer to this question is quite intriguing.
Video: STOP THE LIES! - A.I. made art DOES NOT STEAL art! - Addressing the evidence
The data that drives AI art generation is sourced from an extensive collection of images and information gathered from the internet, often without explicit permission. Is this practice legally sound? The answer is murky, much like the oracular responses of the Pythia.
Chapter 2: Understanding the Data Behind AI Art
The foundation of AI-generated imagery hinges on the gathering and analysis of millions of categorized images—those accompanied by some form of descriptive text.
“AI image generation relies on the assembly and analysis of millions upon millions of tagged images; that is, images that come with some kind of description of their content already attached.”
— The Stupidity Of AI, James Bridle
The images available online typically come with "alt tags," which serve as textual descriptions that help search engines locate them. This process effectively translates the visual into a textual format, allowing for better processing. As noted by James Bridle in The Guardian, this tagging system aids AI art programs in comprehending images. Neural networks like those in DALL-E, Midjourney, or StarryAI analyze countless tagged images, enabling them to perceive nuanced qualities associated with both the visuals and their accompanying words.
Bridle elaborates:
“These qualities are then layered on top of one another to produce new arrangements of shape, color, and composition, based on the billions of differently weighted associations produced by a simple prompt.”
My attempts at using an AI art generator to depict Genghis Kahn in a church yielded diverse results—showcasing the unpredictable nature of this technology.
The data utilized by AI programs predominantly originates from two major non-profit organizations. The first is Common Crawl, which employs a web crawler to index the internet. For over a decade, they have amassed "petabytes" of data, currently indexing around three billion pages monthly.
The second organization, LAION, based in Germany, focuses specifically on images. They claim to have over five billion "CLIP-filtered image-text pairs," accessible to the public, with some data sourced from Common Crawl. Both entities assert that they do not collect or repackage images, but merely index them. While this may be technically accurate, it raises ethical questions about the fine line between public service and questionable practices.
Taking a Face Without Consent
A platform called Have I Been Trained exists to help individuals discover if their images are included in these databases, a service that has gained traction among artists.
In 2022, an artist known as Lapine discovered photos of her face indexed by LAION. According to Benj Edwards at Ars Technica, these images were not part of her art portfolio but were taken by her doctor for medical purposes. Lapine had previously filled out a form explicitly prohibiting the use of these photos beyond her medical treatment. However, following her doctor's passing, she suspects they ended up in the registry without her consent.
Edwards also uncovered other medical images in the database, raising concerns about the awareness of those involved. Photographer Robert Kneschke is preparing to sue LAION after discovering many of his images, including watermarked ones from Shutterstock, were in the registry. Dr. Andres Guadamuz, an intellectual property expert, noted that Kneschke sent a formal complaint to LAION, which responded by denying having any images, only links. This led Kneschke to issue a cease-and-desist letter, to which LAION replied with a legal notice asserting that Kneschke's claims were unfounded and that he owed them legal fees.
Even legal experts like Guadamuz find the situation perplexing, admitting that their initial assessments of Kneschke's claims were dismissive, but further investigation suggests there may be substantial merit to them.
As we navigate these murky waters, we are reminded of the ancient Greeks and their encounters with the Oracle.
Video: How does AI REALLY Steal Art?
Chapter 3: Our Modern-Day Oracles
I recall a peer ridiculing the ancient Greeks for their acceptance of the vague prophecies from the Oracle at Delphi, while we, in contemporary society, pride ourselves on relying on empirical data. However, the world of AI art seems to defy this logic. The same prompt can yield vastly different outputs, reminiscent of the enigmatic responses from the Pythia.
The legality surrounding the data used to train AI programs remains uncertain, even for experts like Dr. Guadamuz, who has authored numerous works on intellectual property rights and serves as Editor in Chief of the Journal of World Intellectual Property.
Despite her unsettling experience, Lapine continues to use AI art programs, and I find myself in a similar position. My collection of bizarre images likely stems from dubious sources, yet this does not deter my interest.
Perhaps the ancient Greeks would find amusement in our modern predicament. While they undertook journeys to Delphi for their cryptic insights, we remain seated, captivated by the enigma of AI art generation.
If you wish to explore more narratives like this, please consider subscribing to my mailing list. Your support on platforms like Medium is invaluable, allowing access to diverse voices like mine for a modest fee of $5 per month. I also appreciate any tips for referrals.