Who Owns What: How to Classify Ownership of a Chatbot’s Generated Content

Editor’s Note:  Please check back next week for Ann’s related article “Academic Theft – Is ChatGPT an Offender?”  These two pieces provide a great perspective into the emerging world of AI, and Ann has worked diligently in her legal research and writing.

Human invention breeds automation.  When a remedial, repetitive task can be done by a machine, almost immediately will it become an industry standard.  Yet, as society progresses, the perception of what is considered “remedial” is constantly evolving.  For example, one significant advancement of automation was the printing press.  Once, newspapers and novels had to be hand-pressed or handwritten, but with the printing press, suddenly written materials could be mass-produced.  Although it could be said the art of scribing was lost, what was gained from the printing press was far more significant.  From there, the Industrial Revolution changed the very landscape of the world.

With modern production automation at its peak, people have shifted to developing automation for other aspects of life.  Specifically, artificial intelligence has entered the conversation.  Artificial intelligence is distinct in its automation, however, as it seeks to mimic the human mind.

There are a variety of artificial intelligence programs on the market but most recently “chatbots” have gained popularity.  Chatbots can simulate human conversation.  This is done by utilizing Large Language Models, which are essentially probability analyses.  By comparing the prompt with documents either uploaded by the user, or documents already existing online, the model can “learn” and predict the next words in the sequence.  The more information inputted into the machine, the more robust the output will be.

Yet, questions emerge.  When a person inputs a custom response into the chatbot, the chatbot outputs a unique response.  If the person then passes that output off as his own, should this even be penalized?

Taking it a step further, would ChatGPT have any legal claim against someone who used its program, passed the work off as his own, and profited from the work?  Can a company copyright the material its program is creating?

ChatGPT and Copyright

No – or at least not under any existing laws.  OpenAI, the parent company of ChatGPT, does not have any legal claim for the simple reason that to register for copyright, it must be a human who created the work.  However, this reasoning goes both ways.  As OpenAI has no claim to the material generated, the person typing the prompts into ChatGPT may not file for copyright protection either.  The U.S. Copyright Office released a policy statement comparing a person inputting prompts into ChatGPT to a person commissioning an artist.  A person would have no copyright to a commissioned art piece, just as a person has no claim to computer-generated code.

However, the U.S. Copyright Office’s statement does find ChatGPT as a resource permissible.  The Copyright Office will not register a pure “mechanical reproduction,” but if a person modifies the material to such a degree that the work is “independent of” the AI-generated material, then that modified material may be eligible for copyright protection.

Though the U.S. Copyright Office has clearly defined ChatGPT to be a machine, could this make generated outputs fair use?

What Kind of Machine?

To best answer this question, it would help to categorize exactly what chatbots are.  Are they search engines?  Or are they databases?  If ChatGPT can be categorized as either, it may make use of generated outputs in academic and legal practice more palatable.

A search engine quickly and comprehensively retrieves data from the web, compiling this data into a listed format.  The engine will then rank this list in what it determines to be most relevant to the user.  This list is likely extremely broad, with the user being left to search for what he needs.  Yet, a research tool, such as a database, acts much the same by narrowing the output more finely than a search engine might.  This is due to a database drawing from a more finite data set.

Tracing ChatGPT back to its conception, the core of the programming was always the Language Learning Model.  By being able to access and compress a large number of data points into a uniquely generated output mimicking speech patterns, chatbots seem to be a supreme combination of search engines and research tools.  With that combination, chatbots cannot fit squarely under either definition of a search engine or research tool.

Though it may be difficult to categorize ChatGPT, its similarities to search engines do reveal its hidden danger.  ChatGPT may not have a copyright claim to its outputs but the authors of the materials it “learns” from could.  If a user requested ChatGPT to recite a novelist’s work, and that user passed the work off as his own, the novelist would have a claim against the user.  More likely, ChatGPT may incorporate copyrighted work into its outputs unbeknownst to the user.  With a search engine or a database, a user can trace answers to the source.  However, with ChatGPT, the user has no idea how ChatGPT generated its output.  If you request a citation, the unpaid version of ChatGPT reminds its user that it cannot access the internet, so it is unable to provide that output. With such a response, a user has no discernable way of knowing what part of the output is fair use.

In the never-ending battle to automate all aspects of human life, ChatGPT comes close to automating the human mind. Yet, while ChatGPT and similar chatbots are revolutionary, they are nothing more than shallow mimics.  Chatbots may be able to process an input and generate an applicable output in seconds, but the chatbot adds nothing of substance.  It is for this very reason chatbot responses are ineligible for copyright: they are just machines.