In June 2022, the internet was taken by storm due to the introduction of an Open AI system: DALL-E 2, which can produce any image which the user wants in a span of few seconds. However, in doing so, the AI has generated a debate: isn’t the AI infringing on the copyrights of the original cartoonists or photographers? Further, it has added fuel to the debate: “Is AI sentient?” and what this means for the Art industry.
This article aims to shed light upon the workings of the DALL E 2 AI, the significance of its presence in a financially distraught art industry, understanding the notion of an AI Author in the Indian context and the legalities of whether the work created by the AI is borrowed or stolen.
WHAT IS DALL E 2?
Dall E 2 is a Generative Artificial Intelligent (GAIs) system that generates works of art that can compete with conventional artists in technique, if not intent. Dall E can perform this feat by relying on photographs that match the major aspects of the description that a user provides it. This exciting development by an AI is bound to be revolutionary.
DALL-E 2 will be made available to up to one million individuals as part of a large-scale beta test, according to Open AI. Each user will be able to create 50 generations for free during their first month of service, followed by 15 each month thereafter. (A generation is the creation of four images from a single prompt, or three more if you want to change or vary anything that has already been created.) Additional 115-credit bundles can be purchased for $15, and more specific pricing is expected as the programme matures. Importantly, users have the right to monetize the photos created with DALL-E, allowing them to print, sell, or otherwise licence the images.
It has “trained” on around 650 million photos collected from the internet, learning the correlations between images and the words used to describe them from that dataset. However, while Open AI screened out photos for content (e.g., pornography and duplication) and applied additional filters at the API level, such as for famous public personalities, the firm concedes that the system might occasionally generate works that include trademarked logos or characters.
The Dall-E 2 is a higher-resolution, lower-latency version of the first system that generates pictures from user-written descriptions. Compared to its predecessor, it can produce pictures in a variety of formats, including photorealistic photos, paintings, and emojis. It can also “manipulate and organise” elements in its photographs.
TAKING ADVANTAGE OF LACUNAS
A. Can an AI become the Author?
An AI cannot become an author, and if so, there cannot be an infringement. Many countries such as Spain and Germany still consider that a work is bound to have only one author, who is to be a human being. Therefore, the creation of an AI is considered authorless. The US Copyright Office took the same stance in 2019, stating that “works generated by a machine or a purely mechanical mechanism that operates randomly or mechanically without any input or creative participation from a human author would not be registered.”
According to the US Copyright Office, it will register an “original work of authorship, provided the work was created by a human being.” This is due to the long-held belief that the only things worthy of copyrighting are “the fruits of intellectual labour”, as generated by the “creative forces of the mind.” The doctrine was created in 1991 by an instance of a phone book business stealing listings from another. The Supreme Court ruled that, while a phone book was compiled with care, the information contained within it was not an original work created by a human being and hence could not be copyrighted. It will be interesting to observe if any challenges are issued to people attempting to licence or sell a DALL-E work for this reason.
Taking advantage of this legal environment, huge technology firms are training their robots with vast databases without fear of intellectual property problems. A creative issue such as this, demands innovation by the lawmakers in upending the legal status of “author” and including AI within it as well.
In India, the position of law is different. In August 2021, the copyright office had recognised authorship of an Indian AI tool- Raghav. This has come after years of ambiguity under the copyright law regarding the authorship of an AI in India. This means that when GAI’s like DALL E 2 are able to create original work, the AI will receive the authorship of the work. Further, this would tantamount to holding the AI liable in case of infringement of others copyright. It will be interesting to witness an AI defending itself in the courts!
B. Text Data Mining and its Legality
The legal position on AI systems like the DALL E 2 is unclear, particularly in India, where there have been few instances using Text and Data Mining, or TDM. This is the technical term for training an AI by sifting through a massive amount of source data looking for patterns. TDM is largely protected under Fair Use in India, which allows different types of copying and scanning for the purpose of giving access.
As of now Open AI has committed to evaluate various approaches to deal with potential copyright and trademark issues, which could include allowing such generations as part of ‘fair use’ or similar concepts, filtering specific types of content, and working directly with copyright [and] trademark owners on these issues.
However, It isn’t only a DALL-E 2 issue. As the AI community develops open-source implementations of DALL-E 2 and its predecessor, DALL-E, free and commercial services emerge on top of models trained on less-carefully filtered datasets. For Example Pixelz.ai, which launched an image-generation software driven by a bespoke DALL-E model this week, makes it trivially easy to produce photographs with numerous Pokémon and Disney characters from films such as Guardians of the Galaxy and Frozen. In the US, in Authors Guild v. Google the author-representation organisation, claimed Google of digitising printed works that were still protected by copyright. The primary goal of the project was to categorise and database the texts in collaboration with numerous libraries to facilitate a study. Authors, on the other hand, were concerned that Google was infringing on their copyright, and even if it wasn’t, it was making the text of a still-copyrighted work publicly available which was forbidden from scanning and storing it in the first place. The Second Circuit eventually decided in favour of Google, stating that digitising copyright-protected material did not constitute copyright infringement.
For an AI to produce copyrightable work, the work must be original. In the case of DALL E 2, one must look at how the AI operates, how it reformulates or changes the existing work, how it learns and generates new pieces, and how it is based on patterns that it observes working. All these must be considered case by case, with the necessary exceptions.
The issue is that because these AIs’ code cannot be audited, determining how they operate is significantly more difficult.
C. Protection under Creative Common Licence
Creative Commons licences do not limit reuse to any specific sort of reuse or technology as long as the terms attribution (BY), no-derivatives (ND), share-alike (SA) and non-commercial (NC) are followed. As a result, to the degree that copyright license is necessary, no specific or explicit permission is required from the licensor to utilise CC-licensed content to train AI systems. Furthermore, creative commons licences do not have constraints and exceptions like fair use. If the usage does not require authorization under copyright or sui generis database rights (e.g., text and data mining authorised under an exemption), the AI training activity may be conducted without consideration to the CC licence.
Furthermore, these licences work within the framework of the copyright system. The licences do not cover privacy, personality, publicity, or other forms of rights or ethical issues.
The training of programmes such as DALL-E Mini is based on free datasets, one of which is YFCC100M, which comprises 99.2 million Flickr photographs distributed under various Creative Commons licences.
However, the widespread and indiscriminate usage of these photos is incompatible with the CC licence. One can moot that the AI does not abide by the Creative Commons’ “restrictions stipulations”, thereby infringing an artist’s copyright.
CONCLUSION
The Google developer who engaged with LaMDA was certain that AI has a conscience. AI systems like DALL E 2 have once again proven why that discourse has weightage. As for the laws governing these systems at the moment, experts recommend that for specified instances, a newly generated copyright be developed for widespread usage by AI. An update that should be included in future regulations that would have a significant impact on copyright management.
It would also be required to determine whether the countries in the European Union and the United States share the same viewpoint and how India finds its way out to its advantage. This argument does not appear to be coming anytime soon, since work is still being done today to appropriately govern technology that emerged years ago, such as social networks.
Source: indiaai.gov.in
