This is part of an ongoing series of posts about AI and how the law treats its outputs. You can find related posts here:
All of my maundering so far has focused on how existing law handles the output of machine learning systems. In particular, I have looked at how we treat inventions and works of art or literature that have been generated by computers and shown how patent and copyright law handles (or fails to handle) those creations.
But hey, if current law doesn’t cover AIs, why don’t we just change the law! For example, court decisions about machine (and monkey) ownership of copyright have touched on the fact that, if Congress had intended a non-human to own copyrights, it should have said so. But does that mean that Congress actually could have said so?
And now we’re getting into speculative laws. I unfortunately didn’t coin the term “Jurisfiction,” but I’m going with it, because this post is sitting at the intersection of law and science fiction. Put on your powdered white wigs and grab your electrogavels, we’re getting lawpunk up in here.
I promise, it’s going to be way less exciting than it sounds!
The State of the Art
Since we’re talking about two different kinds of law, patents and copyrights, it is useful to point out how they are treated.
In the world of patents, a human inventor is identified by their contribution to the conception of an invention. There is little else required, as a human could simply notice an existing natural process, realize it could be useful for something, and we would call that person an inventor. If a human looks at the output of a machine and realizes that output could be useful, we would call that a discovery just as surely as if they were peering through a microscope.
In the world of copyrights, a human author is identified by the exertion of their creativity to create a particular expression. A photographer doesn’t create the natural scene they photograph, but through the exertion of their creativity in framing the picture and configuring their camera, we recognize them as authors of the photograph. And when a human creates a digital image to reflect their creative vision, we call them authors of it just as surely as if they had painted it by hand.
But in both cases, we have machines that are generating works that are indistinguishable from what a human might have created, without any human intervention. At this stage, we have the human-centric ideas of “conception” and “creativity” that a machine cannot duplicate. In both IP regimes, we also have the idea that the law as-written only protects human beings.
But it is important to recognize that the law doesn’t have to require those features, and that the requirements could be changed by passing new laws. And further, since we’re getting all speculative here, what happens if a machine reaches the point where we have to acknowledge that it is conceiving and is creative. How can the law adapt to that possible future?
[EDIT ONE HOUR LATER]: I just learned that the Congressional Research Service has released a report on some of these issues. It reaches the same conclusions I have in my previous posts, but speaks with just a little more authority than my goofy blog, so I’ve attached it below. They recommend waiting to see how things shake out in the courts before Congress takes action. That’s a reasonably prudent, if spineless approach.
The Source of Power
Every federal law in the U.S. has its basis in the U.S. Constitution. Congress can only do things that the Constitution gives it the power to do. In the case of patents and copyrights, that is Article I, Section 8, Clause 8, commonly known as the Patent and Copyright Clause:
[Congress shall have the power] To promote the Progress of Science and useful Arts, by securing for limited Times to Authors and Inventors the exclusive Right to their respective Writings and Discoveries.
Two thirds of our intellectual property systems are founded on that short clause, which gives Congress the absolute authority (within the United States) to create patent law (covering the inventors and their discoveries) and copyright law (covering the authors and their writings).
The challenge is that those words, “Authors” and “Inventors,” can be interpreted as referring solely to human beings. Deriving from that, through the relevant strains of case law, both IP regimes have settled on the fact that the raw output of an AI system cannot be protected. In the context of patents, a human being’s “conception” of the invention needs to be involved, while copyrights require the human exertion of a “modicum of creativity.”
Meanwhile, the actual outputs of the machines are practically indistinguishable from what a human might create. Protecting these creations would serve the Constitutional goal of “promot[ing] the Progress of Science and useful Arts,” but without the involvement of the Authors and Inventors. So the question becomes, if Congress decides to grant copyrights and patents for things that are purely machine-made, without any human’s intervention, can they do so?
A Brief Trip Abroad
It is worth noting that the U.S. is singular in its focus on individual human inventors for patents. While Dr. Thaler has so far met little success in his worldwide crusade, the fact is that most countries do not put the same heavy emphasis on inventorship that we do. In many places, it may be a relatively straightforward change to adapt their patent systems to open the door for machine inventorship and authorship.
We also have international agreements to contend with. For example, the Berne Convention is a treaty that nearly every country in the world has signed onto, and it also refers to “Authors” as receiving copyright protection. The Patent Cooperation Treaty similarly refers to “Inventors.” At some point these issues will have to be addressed on an international level, regardless of how individual countries attempt to solve the problem.
Other Roads to Travel
The question of the scope of the Patent & Copyright Clause of the U.S. Constitution is ultimately one for the Supreme Court to decide. Given the conservative, originalist bent of the current Court, I think it is unlikely they would look kindly on a dramatic expansion of the clause to include machine inventors and authors which were clearly not contemplated by… well basically anyone.
However, Congress has other tools at their disposal for making laws. As I noted above, the Patent & Copyright Clause is the foundation for two thirds of our intellectual property systems, but what about the other third? Trademark law does not derive from the Patent & Copyright Clause.
Instead, Congress justifies trademark law on the basis of its power under Article 1, Section 8, Clause 3 of the Constitution, commonly known as the Commerce Clause:
[Congress shall have the power] To regulate commerce with foreign nations, and among the several states, and with the Indian tribes.
The power to regulate interstate (and international) commerce has been interpreted very broadly and justifies all sorts of meddling by Congress. It should be possible to shift most, if not all, of the existing patent and copyright laws over to deriving their authority from the Commerce clause by adding “in interstate commerce” in a few key locations. Other people have dabbled in this question. I include those links for the sake of completeness–I haven’t read them and don’t necessarily subscribe to their positions.
A hybrid approach could also be possible, where only machine-generated creations could be given patents and copyrights “in interstate commerce,” while human-generated creations could be protectable regardless.
One interesting quirk that would result from a commerce-clause–based approach to protecting machine-generated works is that anything that remains purely state-bound could not be covered. This happens frequently in trademark law, and each state has its own set of trademark laws as a result. In contrast, current federal law preempts state-based copyright and patent laws and prevents states from creating their own systems. Basing a new IP system on the Commerce Clause would break preemption.
What’s more, if it is determined that Congress cannot protect machine-generated works under the Patent and Copyright clause, then that seems to leave the door open for individual states to create their own patent-like and copyright-like protections for machines!
We could be looking at complete inter-jurisdictional chaos!
Why Did You Make Me Read All That?
Right now there is a serious gap in how machine-generated output is handled, and legislatures abhor a vacuum. Lobbying groups are already trying to figure out how best to protect their interests, but we’re so early in this that nobody even knows what their interests actually are! Meanwhile, the existing law is still getting sorted out through the courts, a process that will take years.
That gives us a chance to form our own ideas for what the future should look like. Here are some questions I am chewing on right now:
- How can we best create a fertile field for human creativity, when the commercial value of human works is shrinking?
- How can we contend with commercial interests that would happily fire every human artist if they could get away with it?
- If we are going to end up with a new IP regime for machine-generated works, how do we shape it into something that satisfies the above interests?
Right now, my instinct is to say that we should take a hard line on keeping AI-generated work unprotectable. If a company wants to own the inventions and art that it produces, let it hire human beings to do the work. Protectionist Luddism you say? Sure, but why not? I care a bit more about keeping human hands in the Progress of Science and useful Arts than I do about looms.
Anyway, here’s that report I mentioned from the Congressional Research Service:
Patenting the odd, the impossible, and the unnerving 