Footnote 189: A Copyright Story
On the first page of my book, Artificially Human, you’ll find an attestation:
These words were written by me, a human. Machines helped with the footnotes.
I thought it was a funny way to begin a book about automation and AI. The U.S. Copyright Office found it less amusing. This is a story about how aging bureaucracies are running headlong into an AI-powered world for which they are woefully unprepared.
Before diving into the details, I want to make one thing clear. The people at the U.S. Copyright Office have been nothing but professional, responsive, friendly, and reasonable in my interactions with them. This is not a rant about how the organization is run. It’s about how we’ve placed the office in an impossible position by failing to invest in government services.
With that clarification, let me take you on a journey into the world of copyright administration.
Welcome to 1999
On July 8, 2023, I published my first book. I didn’t write it to make money or become famous. I wrote it because I’d spent years immersed in the field of automation and wanted to organize my thoughts. I also wanted to prove to my now-deceased 10th-grade English teacher that he was a poor judge of writing talent.
With my book published, it was time to secure copyright protection for my work. Registering with the U.S. Copyright Office isn’t technically necessary, but who doesn’t love a government-issued certificate?
Upon visiting copyright.gov, I was pleasantly surprised. The landing page is modern and intuitive. Heck, there was even a “Copyright and Artificial Intelligence” feature on the homepage.
Source: https://www.copyright.gov
So far, so good. I clicked the “Registration” menu and then the “Register Your Work: Registration Portal” link. That took me to a page where I could log in to the Electronic Copyright Office (eCO) registration system. This is where things went downhill.
The eCO system is straight out of 1999. Confusing layout — check. Random fonts — check. Four separate links for help and information — check. This was more along the lines of what I was expecting.
Source: Electronic Copyright Office (eCO) Registration System
Fortunately, the website worked as intended. I submitted my application into the void and awaited a response.
Machines Helped with the Footnotes
I didn’t have to wait long. Less than two weeks after submitting my copyright application, I received the following message:
I am working on the copyright application you submitted to register “Artificially Human: Making Sense of Automation and AI.”
Information in the deposit copy indicates that some of the text in this work (footnotes) was generated by artificial intelligence (“AI”).
To be protected by copyright, a work must have been created by a human author. Material that is generated by an AI program does not satisfy this requirement, even if the material was generated in response to numerous and/or complex prompts entered by a human being. The Copyright Office will not register material generated by an AI program.
Apparently, the U.S. Copyright Office didn’t find my attestation amusing. In full transparency, I used ChatGPT to format the footnotes in my book. I provided the author, title, publisher, and other source information and asked the AI to format an APA-style footnote.
I explained my process to the Copyright Office examiner, clarifying that no content in the book was generated by AI. Treating the ChatGPT formatted footnotes as AI-generated content seemed like a slippery slope. Did that mean wording changes suggested by Grammarly or auto-insertions done by Scrivener were also AI-generated content?
I sent my reply and waited for a verdict.
Footnote 189
The examiner responded within 24 hours. I don’t know what I was expecting. I’ve become so accustomed to automated messages and apathetic customer service that this seemed like a losing battle. I was surprised to see a human on the other end of the email chain.
Here was the examiner’s response:
Thank you for your reply. You indicated that AI was only used for footnote formatting.
I have noted that footnote 189 says the following: This information was sourced from ChatGPT. Here are the relevant parts of its response: “The training process took several months to complete and consumed an estimated 285,000 processor-hours…The actual computing power required to run GPT-3 will depend on the specific use case and the size of the input data, but it can be run on a single GPU or CPU, albeit with slower performance.”
It appears that this footnote includes some text created by ChatGPT. Is this correct?
If the situation is otherwise, please explain.
Are you kidding me??? For context, footnote 189 appears on the last page of Artificially Human. Had the examiner reviewed my entire book? Had he simply searched for “ChatGPT?” Was the review process actually automated? I was both frustrated and kind of impressed.
At this point, I was emboldened to argue my case. I knew a human was reviewing my application. Perhaps a compelling argument could win him over.
Here is my reply.
Yes, footnote 189 was sourced from ChatGPT. However, I would point out that the information is specific to the model that underpins ChatGPT (GPT-3). The model did not provide any original content. It simply provided data on behalf of OpenAI, which is both the owner of GPT-3 and ChatGPT. This would be no different than if I sent an email request to OpenAI asking for this information and received an automated response from the company.
That makes sense, right?
Apparently not…
Thank you for your help. Based on your response I have amended the application to exclude “text generated by artificial intelligence.” The registration certificate should arrive in several weeks.
As noted in my first email, the Copyright Office will not register material generated by an AI program. To be protected by copyright, a work must have been created by a human author. Material that is generated by an AI program does not satisfy this requirement, even if the material was generated in response to numerous and/or complex prompts entered by a human being.
I received my certificate this week. It’s a beautiful piece of government-issued wall art. Near the bottom of the certificate appear the following words:
Material excluded by this claim: text generated by artificial intelligence.
Footnote 189. That’s the text excluded from my copyright claim. I wrote nearly 40,000 words. My copyright includes a limitation due to 50 words generated by an AI describing itself. Feel free to reproduce footnote 189 without my permission.
Old Models, New Challenges
This adventure wasn’t entirely unexpected. I reference copyright law in the “SETTING THE RULES | POLICYMAKERS” chapter of Artificially Human. I claim we have an administrative problem, not a policy problem.
I’m thankful for the time and care the U.S. Copyright Office invested in my application. Many companies could learn a thing or two about customer experience from them.
Unfortunately, the office’s operating model is unsustainable. As AI reduces the production cost of text and visual content, the U.S. Copyright Office and our legal system will struggle to keep pace with demand. I didn’t use AI to generate content for Artificially Human, but many other authors will for their books. Most will not include an attestation like mine.
We need to overhaul how technology works in government. This is not a matter of the U.S. Copyright Office implementing an algorithm that spots AI-generated content. It requires substantial investments and changes like those described in Jennifer Pahlka’s book, Recoding America.
Footnote 189 is a sign of things to come. Talking about policy is interesting. Talking about administration is boring. However, I believe the latter is far more important in a society rapidly adopting AI. Machines won’t be relegated to helping with the footnotes for long.