Wednesday, January 18th, 2023
By now, you’ve probably heard of ChatGPT. It’s an artificial intelligence (“AI”) site offered by a company called OpenAI. ChatGPT and its ilk will dramatically affect content creators.
It can be used to write shockingly good essays and computer code. Businesses could use it to compose explanations and summaries of things. But it has limited knowledge of events after 2021, so it’s weak on current information.
What does it mean for businesses and the people who work there? Today, I’ll focus just on copyright issues.
First, some background: ChatGPT is a large-scale generative AI model. It’s optimized to generate content rather than answer questions. A Google search is usually better at the latter.
It works by having studied a massive database and then learning what is good or bad output through self-learning and human feedback.
You use it by inputting a prompt, which is a request for it to compose something, such as “write a 700-word essay on the copyright implications for businesses using ChatGPT.” (I did that. I give it a B.) Users can rate ChatGPT’s output, which helps it perform better.
OpenAI, in which Microsoft is an investor, will soon release a better version of ChatGPT, GPT-4. Google is competing with a rival AI chatbot, LaMDA, which is in beta testing.
Because ChatGPT is a content generator, understanding copyright law is critical.
Copyright is the exclusive right of an author of an expressive work (such as an article or computer program) to control the copying, distribution, performance, and display of the work and the making of new works based on the original. Copyright infringement occurs if someone else commits one of those acts, such as copying, without permission and if the act isn’t fair use.
Copyright doesn’t protect ideas. It protects only how they are expressed. Thus, generally speaking, if you entirely rewrite something produced by someone else, you can own the copyright to what you create and would avoid infringing upon any copyright in what you rewrote.
An author owns a copyright to what he or she writes, even if the copyright isn’t federally registered and no copyright notice is used. Thus, presume anything you see in writing online or elsewhere is someone’s copyright property.
Usually, an employer owns the copyright to what an employee creates within the scope of employment but does not own what an independent contractor creates unless the employer gets a written ownership assignment from the contractor.
OpenAI’s terms of service for ChatGPT give the user all ownership of output generated by that AI, to the extent any such rights exist. But the U.S. Copyright Office held that AI-generated material can’t be anybody’s copyright property because, to be copyrightable, a human must author it.
That doesn’t mean you’re out of the woods regarding possibly committing copyright infringement by ChatGPT’s output as-is. If your prompt targets a specific work or author or asks for content on a niche subject, ChatGPT’s output might be too similar to a single work in its database that is protected by copyright. It’s remotely possible that infringement could happen even if you use an untargeted prompt.
Consider these tips for protecting yourself against a copyright infringement claim:
• Rewrite what ChatGPT generates as much as possible.
• Use prompts that don’t reference a specific copyrighted work or author and, if possible, avoid prompts that are super narrow in subject matter. Don’t reference in your prompt any author who might still own copyright rights. This generally means any living person or anyone who died in the last 70 years.
• Internet search the output to try to ensure it isn’t too similar to any one thing.
• Save a copy of the prompt so you can later show you weren’t targeting any specific copyrighted work.
• If you don’t entirely rewrite the ChatGPT output, disclose that you initially generated your work with an AI and say you have not knowingly copied anyone else’s work.
Doing the last two things won’t insulate you from copyright infringement liability, but they might reduce your damages exposure.
Overall, it appears the risk of committing copyright infringement using ChatGPT is small. It claims to be engineered to avoid generating an output identical to any single work in its database. And if you use output with only a tiny portion of someone else’s copyright property, that copyright owner may have a weak case.
In 1960, computer scientist J.C.R. Licklider wrote of a coming “man-computer symbiosis.” The upshot was the day would come when AI becomes powerful, but humans are creative, so the best results could be had by close work between computers and human brains. We’ve been getting there for years, but publicly available generative AIs are a giant leap forward. Prepare yourself accordingly.
Written on January 18, 2023
by John B. Farmer
© 2023 Leading-Edge Law Group, PLC. All rights reserved.