The People vs. AI: Who Owns Ideas in the Era of Generative Artificial Intelligence?

Kate White — Mon, 22 Apr 2024 19:53:30 +0000

Key Takeaways:

The rise of impressive generative AI models like OpenAI’s Sora is sparking questions about how intellectual property law will need to adapt. The main areas of disruption include determining ownership of IP created by AI systems, as current laws only recognize human “authors” and “inventors.” While Sora and similar AI models produce high-quality outputs, they are still limited in their true understanding and may just be highly sophisticated imitations.

But how will the real transformative impact of AI come about, and what implications might this have for intellectual property law?

Watch our video: Sora and AI: Who Really Owns AI Creations?

The Disruptive Potential of Generative AI in Intellectual Property Law

OpenAI’s Sora software, a text-to-video generative AI model, recently made headlines showing a series of 1-paragraph prompts and some high-quality footage generated based on those prompts. It is a rightfully impressive showing, but it’s still too early to identify how IP law will be disrupted by generative AI.

The first area of possible disruption is who (if anyone) owns the rights to a work or an invention created by AI? The US Constitution reserves copyrights and patents to “Authors and Inventors,” which thus far has been interpreted as limited only to human beings. There are similar provisions in, for example, the European Patent Convention. As recently as the “monkey selfie” case (finally resolved in 2018), where a photographer left a camera out for a monkey to play with and then advocacy groups tried to claim copyright of the photograph on behalf of the monkey (the photographer also unsuccessfully tried to claim copyright), limiting authorship and inventorship to humans was not particularly in dispute in the legal world. There simply had not been any reason to alter the fundamental rationale of IP protection: in exchange for the hard work of creating the invention or the work and to incentivize this hard work, the creator receives a limited monopoly over monetizing it.

With generative AI like Sora, however, it is now possible to reward a person’s (rather, peoples’) hard work in setting up the model that created the invention or the work. For example, GPT-3 required 45 terabytes of training data, which cost several million dollars and uncounted person-hours of time and expertise. This is a significant investment, which if it can create a sufficient benefit to society, should be rewarded by changing IP law to incentivize it.

This is not an open-and-shut case. Sora, and other generative AI, is an expensive tool for creating limited outputs. Creating and operating a generative AI like Sora is a non-trivial exercise, with millions of dollars and person-hours devoted to collecting, curating, and then supervising any AI’s interaction with its training data. AI-assisted creation is and will be the provenance of large corporations that can foot that bill, or else services like Patsnap which specifically offer AI functionality to subscribers. The classic garage inventor or independent filmmaker will not have access to AI. A modern George Lucas could not use AI to make Star Wars. Also, garbage in-garbage out is still true, and the quality of the AI’s training and interaction can be variable (as shown recently by Google), even with the best intentions.

The Limits of Generative AI and the True Disruptive Potential

My own experiments with generating prompts shows that the AI is always returning something that doesn’t—quite—match what I had in mind when I wrote the prompt. The technology is still in its infancy and may end like 1950s-era predictions of nuclear-powered cars and frequent commercial space travel by, to pick one example, the year 2001. It is not a sure thing that AI will reach the level of quality that its proponents are suggesting.

More importantly, even AIs that “understand” their field really don’t.

Just as computers swiftly execute repetitive tasks to produce complex outputs, AI engages in iterative pattern-matching and sorting to determine—admittedly, complex–associations. In the terms of the classic syllogism, AI does not understand that Socrates is mortal because mortality is an inherent attribute of humans, it has merely identified that there is an association between data labelled as a “human,” and data labelled “mortal.” This is how AIs frequently suffer “hallucinations” in which they present things which simply are not true. Data labelled “cat” is also associated with “mortal,” even where CAT cable or Navy catapults or other, non-feline uses of the word are encountered.

In addition, sometimes there is not an association to reproduce so that a normal run of an image-generating AI like Stable Diffusion produces many images that are not kept after human review—figures have misshapen faces or anatomical impossibilities created by the AI. In the world of litigation, multiple attorneys have been sanctioned after filing ChatGPT-drafted legal briefs which cited cases which simply don’t exist.

In the case of Sora, despite some commentators being impressed by the AI “understanding” cinematic language, it really does not. AIs like Sora are inextricably tied to their training data and they more-or-less reproduce a “best fit” imitation of the associations detected in that data. Assuming Open AI trained Sora on competently shot and directed video (not on the entire corpus of YouTube or TikTok), it would be surprising if it didn’t show competent camera work. Should IP law upend its underlying assumptions to incentivize mere imitation? While it is tempting to imagine that AI will continue advancing based on its “understanding” of the world, it would be a mistake to attribute abilities that AIs, by design, simply do not have.

The second, and most long-lasting, area of disruption will come as inventors and practitioners leverage AI to enhance their own processes. This is where the true disruption will occur, and it will organically arise from the practitioners as new use cases for the pattern-matching and associations are found. In the 1940s, a “computer” was a person, often a woman with significant math skills but few job prospects to make use of them. By the end of World War 2, primitive electronic computers were in use to, for example, decode German Enigma messages. By 1976, the Cray-1 supercomputer was the size of a closet and used an 80 MHz processor and a little more than 8 MB of RAM (simplified for comparison’s sake). By 2024, most people carry around in their pockets a device with multiple gigahertz processors and RAM, 1000x improvements. A user looking at a Cray might think through miniaturization and Moore’s law and determine some of the use cases of modern smartphones, but simply could not imagine all the intersections of cameras, video, sensors, etc. that have come to pass.

Adapting Intellectual Property Laws to the Era of AI-Driven Innovation

But a few things are clear: AI’s ability to sort and classify, and then extrapolate based on those classifications to generate new material, will give new inventors and patent-seekers the ability to iterate on hundreds or even thousands of alternatives to find novelty, or to invent around competitors. Adding AI to existing use cases (vehicle sensors, for one), will also be areas of invention. It’s entirely possible that patent law will need to revisit the Alice decision in light of the integration of AI into business.

“Disruption” will be the word of the day. The unfolding spectacle of its impact across industries and where exactly that disruption will take place, will be fascinating to watch.

Closing Comments:

The path forward may not be straightforward, but the journey promises to be a captivating one, with the full impact of AI’s disruption on the IP landscape yet to be fully realized. As stakeholders from all sectors grapple with these emerging issues, the legal and business communities will play a pivotal role in shaping the future of innovation and creativity.

Follow us on socials:

Author Bio

Christopher Klimovski is a dynamic professional, leveraging his diverse educational background encompassing a Bachelor’s degree in Medical Science and a Juris Doctorate specializing in Intellectual Property law to delve into the realms of futurism and technological disruption. With a passion for exploring the intersections of science, law, and technology, he crafts insightful analyses that illuminate the implications of cutting-edge advancements on society, industry, and beyond. Christopher’s work is characterized by its depth, clarity, and foresight, as he navigates complex ethical and legal landscapes to offer readers a glimpse into the possibilities and challenges of an ever-evolving future.

The post The People vs. AI: Who Owns Ideas in the Era of Generative Artificial Intelligence? appeared first on Patsnap.

who owns AI Archives | Patsnap