As if ChatGPT and friends aren’t causing enough havoc, they are roiling the legal system. All that “fluent pomposity”? Who owns it? Can ChatGPT copyright it’s “hallucinations”?
This spring, Rahul Rao reports that, to date, most copyright and patent systems are only prepared assign IP to humans [1]. When a text, image, or design is created by an AI, this does not fit current models. One solution has been to declare the product not protected, i.e., free for use.
If and when an AI may be granted ownership of its output, this raises questions about the rights of contributors to training data. If ChatGPT is trained on images snarfed from the Internet, shouldn’t the creators of those images get a slice of the IP rights to what ChatGPT makes out of there data?
Rao discusses some of the implications. If an AI can own images or designs, it seems likely that the world will be flooded with “a widening torrent” of AI patents, competing with human inventors. Personally, I think this is nonsense, since the AI is incapable of transferring or enforcing a patent. Nor, for that matter, can an AI be held liable for misbehavior. But what do I know?
Alternatively, if machine intelligences are excluded, and their inventions are open for use, then there will be little incentive to develop this technology. Or, more likely, people will claim ownership of the IP in one way or another.
This would be an extension of long standing practices of large organizations, in which specific people are nominated to formally own specific works created by a nebulous group. It would be simple enough for AIs to operate under contracts which assign all IP to specific parties.
But the underlying issue, of course, is “What, if anything, separates a human creation from an AI creation?”
Personally, I have difficulty with the entire concept of intellectual property in any context. It’s one thing to argue about where ideas and expressions of them come from, and who contributed what to some artifact. But leaping to the elaborate fictional world in which certificates of ownership can be held and traded doesn’t make that much sense to me.
Extending this land of legalistic mysteries to large software programs that no one really understands only makes things worse. As far as I know, ChatGPT and friends don’t run themselves, and we know who created them. So how could they be assigned agency, let alone legal standing to “own” their output?
If an AI can “own” the IP for creative works, then, presumably, it can renew and otherwise extend those rights. Since AIs aren’t mortal, this means that some rights could be held forever. What could possibly go wrong?
My own suspicion is that assigning IP to Ais will be used as a method for the wealthy to control and profit from IP while avoiding responsibility. We will hear, “you can sue me, the AI owns it.”
We might well also see AI’s “licensing” IP from each other. I mean, if an AI can “own” created works, then it can buy and sell them, no? Patent collecting is already an annoying business, having computer programs mindlessly humming away in the corner, sucking up patents will be extremely problematic.
It’s all so absurd.
My advice that you didn’t ask for is to treat the products of AI as “work for hire”. Someone, somewhere, sponsored, created, and ordered the AI to run. The products of that AI should be considered “work for hire”, belonging to whoever “hired” the AI. Note that work for hire doesn’t imply that the employee has or doesn’t have autonomy or legal rights. It just implies that what the person creates in this context automatically belongs to someone else.
- Rahul Rao, Generative AI’s Intellectual Property Problem Heats Up in IEEE Spectrum – Artificial intelligence, June 13, 2023. https://spectrum.ieee.org/generative-ai-ip-problem
Robert McGrath's Blog 