Recently due to expanded beta access to DALL-E, Legalese was able to utilise the mind-blowing and powerful text-to-image generator. If you do not know what I am talking about then I would highly recommend stopping what you are doing and watching a YouTube clip of DALL-E in action. It’s pretty impressive.
“red panda on holiday eating sushi in watercolour style”
The question that has been on everyone’s mind is, who owns the images that DALL-E creates? Or who is the owner of other AI-powered text-to-image generators?
Would it be me because I prompted the programme to create the above red panda piece of art? Or could it be the owner of the AI generator that trains the model to create these images?
In a statement from OpenAI’s (the owner of DALL-E) spokesperson said “OpenAI retains ownership of the original image primarily so that we can better enforce our content policy.” However, the problem with this statement is that it does not simply answer how it is owned by OpenAI nor how users are able to utilise the image.
There are multiple ways of looking at this legal conundrum. OpenAI has, on face value, a legitimate claim to the ownership of the images because they created the programme which creates the images upon relatively simple commands from third parties. However, let’s look at previous examples where an image is created by using a camera. The camera is the method for generating an image. However, the manufacturers of the camera do not have any claim to the image.
Copyright protection is one that arises automatically upon the creation of the work. By default, the creator of the work is automatically given this protection. However, there are basic exceptions to the default, such as those of work that has been commissioned. In this instance, the individual who commissioned the work, not the creator, becomes the owner of the copyright.
In the United States, for a work to be entitled to copyright protection, such work must have a human element involved in its creation. If the work is produced solely by mechanical processes including random selection without the contribution of a natural creator (i.e. a human author) then the work will not be registrable.
The legal questions begin to formulate around whether a human’s input was necessary for the creation of the image. Then there is the question of which human’s input. Was it the creators of the AI programme? Are the individuals whose images were consumed by the AI programme in order to create AI-generated works of art part owners? Is it the company that sells the use of the programme to third-party users? Or are the users who pay for the service to utilise the AI-generated images the owners?
These questions have yet to be answered. There will likely be a case which delves into the mechanisms of copyright law to unpack what existing mechanisms for intellectual property rights can apply in these instances and, in all likelihood, how existing mechanisms for intellectual property rights may need to evolve and expand to accommodate this new frontier.
This area of law will continue to develop over time. AI-generated works are not limited to images. They include literary and musical works as well. It is only a matter of time before case law and/or legislation begin to provide answers to who the owner is of my watercolour red panda on holiday.
– Christian Tabor-Raeside
