DALL-E Vs Getty Images: How AI Is Cleverly Going Around Copyright Laws
The era of AI-image generation tools is here and DALL-E is leading in terms of producing an image based on natural language input. However, concerns regarding the system using copyright images for training and image creation have popped up. A user posted on Hacker News about the tool producing images with the Getty Images logo, which is a service that supplies stock images and more.
The user inputted the text “king of Belgium giving a speech to an audience, but the audience members are cucumbers” and the AI tool produced images. But to his surprise, one of the images contained a bold Getty Images watermark. The watermark in the image may be because the tools used stock photos from Getty Images, and didn’t remove the watermark in the final output.
Why is watermark an issue in images produced by DALL-E?
All the images available on Getty Images are licensed. It means that you have to pay for them in order to use them on your website/app or anywhere on the internet. But the watermark in the DALL-E image raises numerous concerns.
Did the training model use stock images from sources without their permission? Or did it enter a contract with the company? Does it scrape images from almost everywhere for training without giving due credit or following copyright procedures?
The users also raised an issue if the images with the watermark are used or uploaded somewhere, will it invite a copyright claim from the company? AI training needs lots of data and various companies resort to aggressive data scraping to obtain loads of data and train their AI. The line between fair use and misuse is often blurred while developing and training AI tools.
However, this is a genuine concern for users who want to use DALL-E to generate images and use them. If they use something which invites a copyright claim, it will make their lives worse. DALL-E will surely aim to monetize the product in the future and if it wants to do so, it must clear all the air regarding the watermark and image usage. It must also come clear on how the tool was trained and if it used images from Getty Images without permission.
