What is AI image scraping, and how can artists fight back?

AI-generated art has been around for some time, but in the past year it has really taken over online. Despite concerns that artificial intelligence will outpace humans in other forms of “creativity” (see: ChatGPT’s Uninspired prose and “hideous” Songwriting in the style of Nick CaveVisual culture has largely borne the brunt of the bot uprising thanks to the widespread popularity and accessibility of text-to-image generators such as from E2or apps like Lensa, that can Transform your personal photos To the dreams of artificial intelligence at the click of a button.

Even virtual artists have to start somewhere, though. Before they can produce their own bizarre artwork, AI-powered models like DALL-E, Midjourney, Lensa, and Stable Diffusion must be “trained” on billions of images, just like a human artist drawing inspiration from art history. Where do these images come from? It’s been taken — or “scraped” — from the Internet, of course.

In other words, technical AI tools rely on man-made images of training data, which is collected by millions of different sources across the internet. Not surprisingly, people aren’t always happy with their data being harvested, and now they’re starting to back off.

Last week, Meta filed a complaint against surveillance startup Voyager Labs for scraping its user data, and Getty Images similarly announced that it is suing the creators of Stable Diffusion Stability AI for illegal scraping of its content. Then, there are the artists taking the fight into their own hands, with a class action lawsuit filed against Stability AI, Midjourney, and DeviantArt for using their work to train corporate image production companies.

But why are so many artists such bad news, and why are multi-billion dollar companies like Meta involved? First, let’s cover some basics…

What is scraping exactly?

Internet scraping basically involves creating software that automatically collects data from various sources, including social media, stock photo sites, and (possibly The most controversial) sites where human artists display their work, such as DeviantArt. In the case of AI image generators, this software generally searches for pairs of images and text, which are compiled into huge datasets.

Some companies are completely transparent about the data sets they use. Stable prevalence, for example, uses a dataset compiled by the German charity Lion. “LAION datasets are simply indexes of the Internet, i.e. lists of original image URLs along with the ALT scripts found associated with those images,” the company explains in a blog post. website.

Other owners of image generators, such as OpenAI (DALL-E) or Midjourney, haven’t made their datasets public, so we don’t know exactly what images the AI ​​was trained on. However, given the quality of the output, it is believed to be quite extensive.

How is the data used to train image generators?

The billions of text-image pairs stored in these massive data sets essentially form a knowledge base for teaching image generators how to “create” images for themselves. This teaching process involves having the AI ​​associate composition with the visual data of the image and accompanying text.

In a process called “diffusion,” the AI ​​is shown increasingly blurry or “noisy” images, and taught to reconstruct the original image from the visible noise. Eventually, using this method, he will be able to create images that weren’t there before. However, he can only do that if she goes through the process of copying the billions of images already floating around the internet.

What does that mean for artists?

Because artists’ original work—shared on social media, art-hosting websites, or elsewhere online—often falls into huge datasets that are used to train artificial intelligence such as text-to-image generators, they often fear That their work be plundered. These fears are unfounded.

On the Stable Diffusion website, it explicitly states that artists are not given a choice as to whether or not their work is scraped. “There was no opt-in or opt-out of the LAION 5b model data,” referring to the trained data. “It is intended to be a generic representation of language and image communication on the Internet.”

For the most part, criticism of this appropriation revolves around the theft of artists’ work, and the fact that AI image generators could gradually replace them in professional roles. After all, why would a company commission an artist when it can type in their name and get AI to produce similar artwork for free? On the other hand, some artists suggest that the ability to completely scrape Internet content will lead to more creative freedom, or even help develop New forms of creative expression.

Who is fighting again?

In some cases, companies — or even entire countries — are trying to crack down on the indiscriminate scrapping of laws and regulations, though the exact rules for this relatively new practice remain murky.

On January 17, for example, Getty Images launched legal action against Stability AI, claiming that its machine learning model “Copying and illegally processing millions of images “protected by copyright. In statmentGetty Images goes on to say that it believes “artificial intelligence has the potential to stimulate creative endeavours” but that the AI ​​in stability has not sought a license to scrape the Getty collection for its own commercial use.

Meanwhile last week, meta File a complaint v. surveillance startup Voyager Labs, alleging that it improperly collected data from social networking sites Facebook and Instagram, as well as other sites such as Twitter, YouTube, Twitter and Telegram. To expose the data, Voyager Labs has apparently created more than 38,000 fake profiles, and extracted public information from more than 600,000 other users without their consent. Meta is asking the company to stop, as well as forfeit its earnings and payout.

What can artists do?

At the same time as high-profile cases from the likes of Meta and Getty Images, there’s a coalition of artists taking legal action against some of the art industry’s biggest giants. in complaint filed in United States District Court for the Northern District of California On January 13, artists Carla Ortiz, Kelly McKiernan, and Sarah Anderson allege that Stability AI, Mdjourney, and DeviantArt have violated copyright laws with their imagery — as well as the art of tens of thousands of others. Artists – to feed their image generators.

“Although Stable Diffusion’s rapid success was based in part on a great leap forward in computer science, it was more dependent on a great leap forward in image customization,” the complaint says.

Besides legal action and advocacy for legislation to toughen repeal laws, however, there’s not much artists can do to protect their work right now, other than shutting it down entirely. For many artists, of course, this simply isn’t an option.