As part of OpenAI’s DevDay, the company that makes ChatGPT and DALL-E announced that it will protect customers from legal claims surrounding copyright infringement.
Called Copyright Shield, OpenAI has vowed to “step in and defend our customers, and pay the costs incurred, if you face legal claims around copyright infringement.”
OpenAI did not specify exactly which of its products this relates to but added that it applies to “generally available features of ChatGPT Enterprise and our developer platform.” ChatGPT Enterprise currenty offers DALL-E 3.
This seems to suggest that this protection won’t extend to all of OpenAI’s products. ChatGPT Enterprise is the business tier of its Large Language Model (LLM) chatbot ChatGPT. Presumably, users of the free ChatGPT version and the Plus tier will not be afforded protection.
“OpenAI is committed to protecting our customers with built-in copyright safeguards in our systems,” OpenAI writes in a blog post.
All of these companies want to assuage customers’ fears over artificial intelligence tools as uncertainty hangs over the nascent industry.
The issue lies with how generative AI tools were made; AI image generators, for example, were trained on hundreds of millions of images, many of which were public domain but far more were not.
DALL-E, OpenAI’s AI image generator, is a black box when it comes to the training data — we simply don’t know how the impressive model was built. But in the past, OpenAI has said DALL-E was trained on “publically available” data.
An open-source training data set called LAION is searchable via the website Have I Been Trained. If a photographer has ever had their work anywhere online it is more than likely to be included in the data set and therefore more than likely to have been used to train various AI image generators.
However, the AI companies are keen to stress that they believe this type of training falls under the fair use doctrine as they use copyrighted materials in a “transformative” manner.
“When a model is exposed to a large array of images labeled with the word ‘cup’, it learns what visual elements constitute the concept of ‘cup-ness’, much like a human child does,” writes OpenAI in a recent public consultation by the U.S. Copyright Office.
“It does this not by compiling an internal database of training images, but rather by abstracting the factual metadata that correlates to the idea of ‘cup’. This enables it to then combine concepts and produce a new, entirely original image of a ‘coffee cup,’ or even ‘a coffee cup that is also a portal to another dimension’.”
Image credits: Header photo licensed via Depositphotos.