With the rise of Artificial Intelligence, models have been created to generate images that did not exist before. From the trend of having ChatGPT create Studio Ghibli images to Coca-Cola sending out an AI advertisement, these machine-created images have been taking over the internet. For many, these image generators are a fun way to create art. However, ‘AI art’ is increasingly becoming a larger controversy.
How is ‘AI Art’ made?
AI image generators are generative adversarial networks or GANs. They consist of two deep neural networks. Neural networks are a method used by AI to teach it to process data in a similar way to the human brain. The first network is the generator, and it starts with noise (random inputs) and then refines the image pixel by pixel based on probability distributions from the given data set of real images. The second network is the discriminator network, which produces a number between 0 and 1. If the data looks real, then the score will be closer to 1, and if it seems fake, the score will be closer to 0. The generator keeps changing the pixels of the image until it receives a 1. As the image generation model is used, the discriminator improves its ability to detect fake data, which leads to the generator creating more realistic images.
What is Controversial About it?
One of the main reasons ‘AI art’ is controversial is that it steals from artists. To create these images, the AI must use existing data; they cannot exist on their own. While some real images and artworks are gathered ethically, many are not. The work of photographers and artists who post their work online is often stolen and fed to these generators. Lots of this work is copyrighted, and there have been many lawsuits filed against AI companies profiting from the stolen work.
Another popular controversy is the lack of creativity in ‘AI art’. Unlike a human artist who carries intent through every stroke, ‘AI art’ is created through refining random inputs. Any originality from the piece happens by chance and without purpose. Meir, a researcher for Brown University, argues that art is made to move us, provide new perspectives, and give the viewer emotions; it is not made simply for entertainment. While AI can create pretty pictures, they lack the depth from a human perspective.
How is it Directly Harmful?
According to a survey conducted by Upwind in 2024, 87% of Americans are concerned about scammers using AI. More than half of these respondents reported they felt more worried in 2024 than in the years prior. By using AI to create deepfakes, which are fake videos of real people, scammers can pretend to have kidnapped loved ones or impersonate them by replicating their voice and asking for some money. AI can also be used in phishing by crafting more believable emails and fake websites quickly without any misspellings.
In addition, AI damages the environment through its freshwater usage. While AI servers use a lot of electricity, the models have significant cooling costs. This water is consumed in two ways: onsite water consumption to cool down the servers, and offsite water consumption to cool down the thermal power and nuclear power plants generating the electricity. 500 milliliters of water are consumed when ChatGPT-3 is run for 10-50 queries. 500 milliliters is around half a bottle of water. In 2025, ChatGPT processes over 1 million queries a day. This translates to around 2641 gallons to 13208 gallons of fresh water consumed a day from ChatGPT alone. Since AI images use more energy than text queries, they have higher cooling demands as well. Most of this water stays in circulation inside closed cooling systems. However, in evaporative cooling centers, the water is lost to the atmosphere. The overall water intake depends on the weather, season, and location of the data centers. Companies that own these centers are not required to report on how much water they use, so there is currently no definitive data showing the water consumption.
Google, Meta, Microsoft, and Amazon pledged to put more water back into the environment than they use. However, as there is increasing demand for AI computer power, scientists predict that this goal of being water positive will become much harder to reach. The exact environmental footprint of AI is difficult to measure since there are no federal or state regulations requiring companies to disclose their water and energy consumption.
Artists Respond to AI
The Glaze Project works to protect online artists from art theft by providing tools to prohibit AI models from using their artworks. AI models like Midjourney and Stable Diffusion have learned to mimic the art styles of many people in the professional artist community. This directly hurts artists’ livelihoods by replicating their work without any compensation. Glaze allows artists to apply “style cloaks” to their images that mislead AI when the images are included in their training data. Glaze has been shown to be highly effective with a 92% success rate under normal conditions. Nightshade, another project created to protect artists, poisons generative models with prompt-specific attacks that disrupt a model’s capability to generate images from certain prompts.
There have been many court cases around AI. One of the most prevalent ones is Thaler v. Perlmutter in 2023. As shown in the U.S. Supreme Court documents, the case presented the question: “Can a non-human machine be an author under the Copyright Act of 1976?” Dr. Stephen Thaler created an AI model, which he named the “Creativity Machine,” and submitted an application for an image generated by it to be copyrighted with the AI as the creator. The copyright office declined, and Dr. Thaler brought the matter to court. The court ruled against him since the Creativity Machine was not a human being and could not be the author of a copyrighted work.
Another ongoing court case, Andersen v. Stability AI, asks the question of whether AI models can be trained on copyrighted material. Sarah Andersen, a popular cartoonist on the internet, complained about AI copying her art style. In January 2023, Anderson and other artists filed a class-action copyright infringement lawsuit against Midjourney, DeviantArt, and Stability AI. This inspired other court cases, like the Authors Guild class action lawsuit against OpenAI. At the August 12th hearing, U.S. District Judge William Orrick sided with the artists and denied the AI companies’ motion to dismiss the copyright infringement claims. The trial is set to begin on September 8, 2026.
