AI Generated Art and Computer Ethics

With AI constantly growing more accessible, more apps and lenses are being used by the public to create exciting art. There are many generators that will create an image based on a few keywords and selected art styles, some TikTok lenses and apps will create a portrait of you in different artistic styles. It has become trendy on social media to use these types of apps and camera filters to alter photos. It's exciting to see what else we can do with this new technology. However, this has caused outrage from artists whose work seems disturbingly similar to some of the AI-generated art, even some of their signatures and watermarks surviving the AI training process and making it into the final image. Is this theft? How do we determine what is a new, original piece of work and what is theft? Where do we draw the line? Who do we blame? One of the side effects of the rapidly expanding field that is AI is that it is hard to maintain our laws and ethics in this uncharted territory.

AI generated image with warped signature from original artist

Why does a machine need to copy human work in the first place? Surprisingly, AI cannot generate new information for itself (yet). In this instance, the AI is fed thousands of example images to "learn" from in order to create unique images. In order to make really diverse images, the AI needs diverse images for input. Even if the algorithm is robust, the AI will be biased toward one art style if there are not enough examples to draw from. What may have occurred was the engineers in charge of training this algorithm did not use enough examples from different sources. If they only drew examples from a handful of artists, the AI would only learn their style, and even more embarrassingly, their signatures.

Can this really be considered stealing? Isn't the ability to copy art with a few small changes the goal for AI? Can anyone really be blamed for something a machine created? These are conversations regarding computer ethics that are now becoming more and more necessary with the rapid expansion of AI technology. Laws protecting intellectual property become blurred when AI is factored in because there are no preceeding cases to reference in these evaluations. It is crucial to analyze these cases as they happen in order to set a standard for the future as the capabilities of technology continue to grow.

In this case, the engineers are likely at fault. Regardless of their training techniques, they should have communicated with the artists and paid them for their work. The only time an engineer does not need to pay for imagery for their AI is if the image data is free and open source, if they are given the imagery for free to use, or if they collect the images themselves. There is also no room for a mistake in this case. Training an algorithm takes hours, days, or sometimes weeks. During this development, the algorithm is tested multiple times. The people working on them must have realized how closely the AI-crafted images looked like the reference art. There is no ignoring the warped signature mark in the corners. It was sloppy of them to release an app that blatantly copies artwork.

In the future, cases may not be as clear cut as this one. Especially as algorithms rely less on given data and lean toward more unsupervised learning. That's why it is important to implement guidelines where we can to prevent theft of intellectual and creative property. It's unfair to the artists that spend hours of hard work on a painting to have their artwork stolen, replicated, and reposted all over social media for free. Setting these laws will require long discussions about who to hold responsible in these events and how to protect artists' work. It will also require case-by-case evaluations of what constitutes original work. Humans also take inspiration from other people's work, but we have an intuitive moral sense of where the line is crossed into theft. We have to be prepared to evaluate machines with the same scrutiny.