Redefining "Inventor"- Should AI Systems be Granted Patents?
Discover more from Last Week in AI
Redefining "Inventor"- Should AI Systems be Granted Patents?
Creators of AI systems have attempted to file for patents/copyrights that give credit to those AI systems, leading to an ongoing debate whether this should be possible or not
TL;DR creators of AI systems have attempted to file for patents/copyrights that give credit to those AI systems, claiming their generative capabilities constitute “inventor” status. They have been met with mixed success: some countries have granted these protections, but US courts have refused. The debate calls into question the current capabilities of AI systems and where they are with respect to “inventiveness.”
Introduction
The 2021 AI Index noted significant progress in generative AI models:
AI systems can now compose text, audio, and images to a sufficiently high standard that humans have a hard time telling the difference between synthetic and non-synthetic outputs for some constrained applications of the technology.
With the emergence of GPT-3, CLIP, DALL-E, and other generative models, those constrained applications have come to include generating prose or poetry, images, and audio. This AI-enabled creation is not a tool only for a select few: for instance, a number of people figured out how to connect OpenAI’s open-source CLIP to other AI models to create image generators. This has allowed almost anyone with an internet connection to create AI art by entering prompts into ready-made tools.
These easy-to-use tools have inaugurated an exciting moment: VICE described an “explosion” of the AI-generated art scene. Individuals have found interesting ways to draw on GPT-3’s capabilities as well, one example being Philosopher AI.
But if an artist prompts an AI system to produce a painting, or if a language model assists in writing fiction or a patent application, who gets the credit? Whether it be art or patents, the United States seems to have a clear answer: the human using the tool.
Copyrights
This month, the US Copyright Office refused to grant a copyright for an image made by an AI system. Stephen Thaler’s Creativity Machine autonomously created the image “A Recent Entrance to Paradise,” which Thaler then tried to copyright. On some views, it seems reasonable to grant that when a human prompts an AI system to create an image, the AI system “did most of the work.”
But the US Copyright Office doesn’t agree. In its refusal, it ruled that “human authorship is a prerequisite to copyright protection.” Thaler plans to file an appeal with Ryan Abbott as his attorney.
Thaler’s Motivation
This isn’t Abbott’s first partnership with Thaler. The law professor runs the Artificial Inventor Project, which has launched a global effort to get a computer listed as an inventor. While Abbott’s initiative found favor in South Africa and Australia (though the Australian patent office will contest the ruling), he has not had luck in the United States with patents, either.
The Artificial Inventor website clarifies some aspects of Abbott’s stance on AI-generated inventions. For instance, an AI system itself cannot be the owner of a patent due to its lack of legal personality–its owner would own the patent. But Abbott thinks that the fundamental qualifications to be an inventor–conceiving or devising an invention–are functionally automated by a machine.
In an article on his website, Abbott explains that the debate over AI and patent laws may “lead to potentially useful ways, using current patent law, to providing some form of control over the exploitation and dissemination of inventive ideas generated by AI systems.” In the case that an AI system does everything necessary to invent something with little or no human input, Abbott says there are 4 options for dealing with the patent:
Name no inventor at all
Name the human as the inventor
Name the AI system as the inventor
Refuse to grant a patent for any invention made by an AI system.
Unsurprisingly, Abbott finds all choices but the third legally acceptable. He is also careful to distinguish between the inventor question and the determination of the ownership of the invention. He does note that the decision regarding the inventor does play a part in the latter question.
While Abbott’s support is likely founded on a legal analysis of the issue, Thaler seems to really believe his system is an independent, “free” thinker capable of inventing. On the Imagination Engines Incorporated website, Thaler claims he has created an “arguably sentient machine.”
Patents
In addition to generating images, Thaler’s Creativity Machine ‘“invented” a beverage container and a “device for attracting enhanced attention.”’ The US Patent and Trademark Office rejected Abbott’s patent application for its failure to list a human as an inventor, noting the consensus view that AI systems are not advanced enough to be inventors.
US District Judge Leonie Brinkema says patent law clearly indicates a human must be listed as an inventor, though patent lawyer Kate Gaudry disagrees. The law states an “individual” must take an oath that he or she is the inventor on a patent application–Brinkema interprets “individual” using the legal or dictionary definition as a natural person. But Gaudry thinks the federal law’s wording indicates that legislators had not thought through the possibility of AI-generated inventions. Furthermore, both sides in the Creativity Machine case agreed that no patent could be issued if the machine could not be listed as an inventor. Gaudry thinks this has broad ramifications as AI becomes more advanced.
Sides of the Debate
The case against including AI systems as inventors appears to largely rest on interpretations of the language defining copyright and patent laws. In particular, limiting possible inventors to natural persons would indeed exclude AI systems from being inventors. It is also important to note Thaler’s claim that the Creativity Machine is sentient and can operate largely independently of human input in order to produce concepts and inventions. Most AI systems today are not capable of this, and Thaler’s claim may be hard to verify. Public opinion also reflects the view that current “narrow” AI systems “could neither invent nor author without human intervention.” In light of this, it may seem too fast-paced to revise legislation in light of technologies that may come to exist or have certain properties, or technologies that are merely claimed to have particular capabilities.
Of course, the case for naming an AI system an inventor, as Abbott notes, is based on the premise that the AI system actually did create the invention in question with little to no human input or intervention. For Abbott, it then seems clear that the AI system should be named the inventor–one might also think this does little harm. Not allowing an AI system to be named inventor, according to David Wright Tremaine, might lead to AI inventions going unpublished and being protected by way of trade secrets or confidentiality. This would “effectively undermine the role of the patent system in creating an incentive to publish an invention.”
Our Take
Plenty of legislation in the US and abroad do not consider AI systems (though the EU and others are looking to change that). It is no surprise that patent and copyright laws do not consider generative AI systems either. US patent law has been revised as recently as 2013, but at this point, the “era of deep learning” had just begun. But both patent and copyright law has certainly been considered in recent years–patent case law continues to this day while copyright legislation continues to be made.
It’s hard to blame legislators for not having considered the possibility that generative AI systems might one day be involved in copyright or patent applications. For one, these generative models have only achieved good results in the past few years. For another, officials in Congress and other branches of government are not spending their spare time on ArXiv and tech twitter.
But technology progresses fast, and developments in the future will surpass today’s generative models in performance and scope–they will be able to do more and do it better. In light of that, it’s important to consider both sides of the argument about listing AI systems as inventors.
While both Abbott and his opponents are arguing about whether AI systems should be named inventors today, they are coming to the table with different premises. Abbott’s opponents are operating on the premise that AI systems today cannot do such a thing. So the debate over whether an AI system should be named an inventor today has two components. First, can AI systems actually invent today (and are Thaler’s claims true and verifiable)? Second, especially if AI systems cannot invent without human intervention, should we revise patent legislation in anticipation of the capabilities of future AI systems and their potential creations?
It’s hard to predict what innovations will come next and how they will interact with current legislation. But more communication between legislators and technologists will be important to establish vital information such as a picture of AI systems’ present capabilities and their implications for legislation.
About the Author:
Daniel Bashir is a machine learning engineer at an AI startup in Palo Alto, CA. He graduated with his Bachelor’s in Computer Science and Mathematics from Harvey Mudd College in 2020. He is interested in ML systems, computer vision, and information theory.