It all started when I typed a perfectly reasonable prompt (提示词)into one of several apps on the market that can create an image based on text. "Skull space laser dinosaur starship explosion," I wrote. The app processed for a few seconds, and returned four images, one of which was strangely accurate: a dinosaur-looking skull screamed out of an empty space, trailing fire. It looked like an illustration from the art magazine, and perhaps art from the magazine influenced its creation.
Text-to-image AIs identify images by looking at the text that people have used to describe those pictures online. When the app got my prompt, it studied images that random people had described as "dinosaur" or laser and soon then used what is called a diffusion model (扩散模型)to add a bunch of random chaos to those pictures. Once they were suitably completed, it "upscaled" them, removing noise and sharpening focus. Its work is so good that an artist using it recently won first place for digital images at the Colorado State Fair.
But there are major ethical (道德的)issues raised by the success of such AIs. The biggest has to do with those training data sets. Reporters recently discovered that the data set used by Text-to-image AIs contained images of violence. Some companies are working on ways to prevent the public from seeing images based on offensive and illegal pictures in the data set. A representative of the companies also noted that the images in its data set are "already available in the public internet on publicly available websites".
But even if this problem is fixed there is still the question of all the other pictures online that are being transformed into AI-generated masterpieces. As many artists have pointed out, their works are being used without payment. The image-generating algorithm (算法)creates illustrations and even movies by using data sets stocked with art stolen from artists who post their works online.
Some AI researchers argue that their algorithms aren't stealing from artists so much as learning from them just as human artists learn from each other. But a more ethical approach would be for companies to acknowledge their debt to artists and create a model of voluntary collective licensing, much like what radio stations first did in radio's early days. Back then, musicians created groups like BMI to collectively license their music to radio stations—then BMI would pay artists based on how often their songs were played. Perhaps artists and art institutions today could form a "collecting society" that would allow companies to license their artwork for data sets.
To create ethical AI systems, we need to acknowledge the people whose work makes those systems so magical. We can't simply snarf up every image online-we need humans to manage those data sets and we need to pay them to do it.
