In recent months, it has been abundantly evident that harnessing artificial intelligence (A.I.) to produce original ideas is not some harmless amusement. Regarding copyrights, it may soon become the focal point of many Intellectual Property (I.P.) conflicts.

AI-generated content and current I.P. protection requirements do not mix; they are like oil and water. The human mind must contribute, whether for copyrights or patents. However, as we shall see, this issue still needs to be resolved.

Can view This issue via the lens of patents for inventions produced by Artificial Intelligence software. The DABUS, or “Device for the Autonomous Bootstrapping of Unified Sentience,” was an example. Since Dr Stephen Thaler’s AI is neither a natural person nor cannot be an inventor or a patentee, most courts rejected his attempts to get patent grants. A.I.’s tireless efforts, however, cannot be overstated. Experts focus on patentability rather than getting computers. The exclusive rights flow to the system owner or programmer because they are natural persons.

But on October 20, 2022, the U.S. Court of Appeals for the Federal Circuit denied Dr Thaler’s most recent denial, rejecting his claims that “inventor” can encompass non-human entities when understood in the context of the Patent Act. The Court relied on precedent from the Federal Circuit and the U.S. Supreme Court to determine that an “inventor” must be a “person.”

“Inventors must be natural beings, and neither corporations nor sovereigns may be inventors.”

Since then, Dr Thaler has made it known that he plans to pursue his efforts by petitioning the Supreme Court.

Painting by numbers: the emergence of a new technique

ARON was developed by pioneering computer scientist Harold Cohen while employed at the University of California, San Diego. According to Cohen’s instructions, this programme created abstract paintings in 1973. The early AARON paintings were originally black-and-white prints that Cohen later coloured himself. However, Cohen soon showed Artificial Intelligence how to develop images independently.

Since then, AI-generated art has only gotten closer to the human ingenuity it imitates as the algorithms and underlying technology have improved. Examples included de-ageing and convincing face swaps of celebrities, as well as convolutional neural networks (CNNs) that produce art, such as Google’s Deep Dream project.

The DALL-E meme craze

Public users can now use their image results commercially after the beta release of DALL-E 2 in July 2022 — for a fee, of course. AI-generated imagery emerged online in a frenzy of creative activity that was simultaneously engaging, oddly humorous, and legally questionable.

Other AI-driven image generators were made public about the same time as OpenAI’s DALL-E 2, which was shoot. In the summer of 2022, Mid Journey, Stable Diffusion, and DreamStudio AI all released and experienced varied levels of success.

From the artwork featured on its Community Showcase website, it is clear that Mid journey is primarily interested in luring more serious (or, at the very least, more detail-oriented) artists to its tool. DALL-E 2, on the other hand, attracted a lot of interest from those looking to amuse themselves and others online by fusing the absurd with the everyday. Thus, we have well-known cartoon characters buying T.V.s from big-box retailers.

When given outlandish instructions like “Bjork dressed as a goth playing golf, about to tee off at the ninth hole on a sunny day in Iceland,” DALL-E 2’s competitors Mid Trip and the open-source Crayon may produce similarly bizarre drawings. That is because these systems’ consumer-facing models rely on specific language cues, and their algorithms were grown using billions of image-word combinations found on the internet.

Up until someone loses an I.P. Docketers, it’s all fun and games.

While the OpenAI developers removed graphic or violent information and images of famous people from their “training” materials, they did not wholly omit patented, copyrighted, or trademarked intellectual property. On a scale of billions of examples, it would be challenging to filter out copyrighted content.

A.I. image-generating software has a wide range of restrictions when basing images on sources protected by intellectual property. For illustration:

Pixelz.ai often accepts suggestions containing images from well-known video games and comic books (as well as of public persons), and the results can be startlingly accurate.

• On the other hand, Jump Story uses A.I. to discover “genuine” original photos for publication without permission.
• Crayon is in the middle of the spectrum. While trademarked character names are not out of line from appearing in prompts, the end products may not accurately reflect the registered intellectual property.

Uncertain legal situation

Before a flood of artificial intelligence (A.I.)-generated images took over the internet’s popular culture, I.P. regulators in the U.S. were already debating the legality of an A.I.’s output.

Dr Thaler (of the DABUS mentioned above) had requested that the United States Copyright Office (USCO) grant him the copyright to a piece of art that his computer programme, Creativity Machine, had “autonomously developed.” According to the USCO’s decision from February 14, “[Copyright] Office policy and practice consider human authorship a condition for copyright protection,” hence he was refused. This ruling and the outcomes of Dr Thaler’s other international initiatives demonstrate that in most states, intellectual property (I.P.) can only result from human inventiveness. South Africa is an exception to this pattern, having issued him patents with DUBAS as an inventor.

Similar problems emerged with developing several A.I. image-generators, but more enormous stakes were involved. The following are only a few of the many questions that original art creators, A.I. programmers, end users, I.P. regulatory bodies, copyright and trademark attorneys, and other I.P. professionals must take into account:

• Who, if anyone, “owns” the work produced using these programmes?
• If AI-generated art contains protected content, does it automatically infringe copyright or trademark law, or can fair use or dealing rules be applied?
• Does this violate the artist’s copyright when an artwork is second-hand to “teach” an A.I.?
• What constitutes actual harm to the original work of the copyright or trademark holder?

            At least in part, several of these queries have been addressed. For starters, it has been confident that an A.I. cannot claim any intellectual property right because it is not an average person. Therefore, only the business that created the A.I. tool, the individual who entered the text prompt, or those who already owned the copyrights and trademarks in question may assert ownership. This proper transfer differs depending on the software and licence:

• Based on a Creative Commons Attribution-Noncommercial 4.0 International Public License, Mid Journey permits authors who utilise its A.I. on a free or trial basis to use the imagery in any noncommercial fashion with attribution. The business can freely licence the song while it is still its property. Ownership of the work subscribers produce for Mid journey is contingent upon Mid journey’s approval.
• Although paying subscribers can use the imagery they create for reprinting, selling, merchandising, and other commercial purposes, DALL-E retains ownership. Intimating that ownership may reside with the firm rather than the system’s creator or operator, OpenAI has cautioned users that they risk losing their rights to use photographs created using DALL-E if they violate the company’s terms and content restrictions.
• Being open-source, Crayon places no limitations on what users can do with user-generated content. Still, this licence may soon be subject to legal review because the business is primarily ad-supported.

Persistent questions

The presence or absence of transformation is a standard indicator of copyright infringement. It can be difficult to prove infringement because images from DALL-E 2, Crayon, Mid Journey and even Pixelz might be so drastically changed from the originals that they are identical to derivative works. Other goods using the same tools are based on pre-existing I.P. and must be more innovative to avoid accusations of infringement. But do they seriously impair the I.P. owners’ ability to profit? Only a case-by-case analysis can be second-hand to evaluate this.

Has money been made off of the A.I. images? If the answer is yes, it is an infringement, though there may still be exceptions for fair use or dealing. Most of these tools have premium subscription models, so consider that (Pixelz requires regular payment after a brief trial). Also, when gathering photographs for the A.I. systems’ training data, did the corporations break any I.P. laws? The key determinants are the datasets used to train the A.I. models, some of which may be copyright protected, and whether the newly created images sufficiently replicate the source to sustain an infringement claim. There is no precedent supporting the idea that using publicly accessible Artificial Intelligence training data to create a work is considered fair use.

Some organisations have yet to raise concern flags about AI-generated photos, even if codified remedies to these problems may not be available very soon: In September of this year, renowned image source Getty Images prohibited their use. In contrast, Microsoft was confident enough in the potential of DALL-E 2 to collaborate with the Artificial Intelligence company to develop Microsoft Designer. The Microsoft Office suite will eventually include this next programme.

While some people might be concerned about violating someone else’s exclusive rights, artists and organisations, require clarification on what all this might mean for their intellectual property. There is no need to worry, though. Your partners in developing a thorough, efficient I.P. protection strategy that will withstand the shifting regulations and winds of technology are the I.P. law specialists at I.P. Docketers & Associates.

 On November 23, 2022, this story first appeared on the CITMA website. To the CITMA website.