ChatGPT, an artificial intelligence (AI) language model developed by OpenAI, has revolutionized content creation and information synthesization in just a few months. Give ChatGPT a prompt, even a fairly complicated one, and it will spit out a human-like response in seconds. OpenAI recently launched GPT-4, the even more capable update to ChatGPT’s previous version 3.5. The artificial intelligence company behind ChatGPT claims “GPT-4 is more reliable, creative and able to handle much more nuanced instructions than GPT-3.5.”
Parkway has started to integrate ChatGPT into some of our processes, so we’re excited about the update. So far, we’ve been liking the paid version of ChatGPT for things like ideation, finding synonyms and writing the first drafts of internal documents. The possibilities of using ChatGPT are seemingly limitless. But as we begin to figure out how ChatGPT can best help Parkway (and our clients!), we wanted to make sure we understood what we can and can’t do with the tool from a legal perspective.
We’re pretty good at a lot of things, but we aren’t lawyers, so we chatted with an expert: Vincent LoTempio, a registered patent attorney who handles patent, trademark and copyright details for inventors, businesses and entrepreneurs. Vincent helped us understand two reasons ChatGPT won’t be replacing human writers any time soon:
- ChatGPT is trained via unsupervised learning.
- There are unique copyright considerations.
Here’s what Vincent had to say about these limitations, and what we (and other businesses) should keep in mind when adding ChatGPT to our workflows.
OpenAI’s chatbot uses a process called “unsupervised learning” to generate unique responses. Vincent told us about his own experience with ChatGPT, saying, “While most of the responses I found from using ChatGPT were relevant and accurate, inherently the process of unsupervised learning can lead to misleading responses. When I prompted the chatbot to provide reasons why it may produce such responses, it pointed to the possibility of topics “outside the scope of its training data,” “ambiguity in language” such as homophones and nuances, biased or inaccurate source material, missing context and technical issues.”
OpenAI acknowledges these limitations, even within GPT-4. The company writes in its release announcement, “GPT-4 generally lacks knowledge of events that have occurred after the vast majority of its data cuts off (September 2021), and does not learn from its experience. It can sometimes make simple reasoning errors…or be overly gullible in accepting obvious false statements from a user.”
These limitations reinforce the importance of not relying too heavily on the responses provided by ChatGPT where inaccuracies may lead to real-life consequences such as medical or legal advice, Vincent reminds us. There are many industries that rely on subject matter experts to create content. In these spaces, ChatGPT probably isn’t a great partner. Even in less technical industries, businesses should consider AI-generated content a helpful tool but avoid copying it verbatim. Depending on the use case, an industry-specific or business-centric AI bot may be more helpful.
It’s important to understand where ChatGPT sources the information in its responses. Vincent asked the chatbot to explain: ChatGPT is trained on “massive datasets of human language.” These include both “books, articles and other texts that are in the public domain” as well as “copyrighted material, such as news articles or website content” that “may infringe on the copyright holder’s intellectual property rights.”
As an intellectual property lawyer, Vincent knows how vital it is to attribute information to copyright and trademark holders, including copyrighted source code, which a casual ChatGPT user might not recognize as protected IP. When Vincent asked ChatGPT for its sources following random history questions, it would not provide specifics but reiterated it was trained on a “vast corpus” of information. It did, however, provide useful books related to the given topics for further research.
The US Copyright Office will register “an original work of authorship, provided that the work was created by a human being” and refuse claim registrations if “it determines that a human being did not create the work.” Vincent says, “With this in mind, output from ChatGPT would not be copyrightable. However, the degree to which human oversight or modification is made in the process could alter that distinction. It will be interesting to see cases in the future weigh on that question as AI-generated content becomes more prevalent in our lives.”
You may provide input to the Services (“Input”), and receive output generated and returned by the Services based on the Input (“Output”). Input and Output are collectively “Content.” As between the parties and to the extent permitted by applicable law, you own all Input. Subject to your compliance with these Terms, OpenAI hereby assigns to you all its right, title and interest in and to Output. This means you can use Content for any purpose, including commercial purposes such as sale or publication, if you comply with these Terms. OpenAI may use Content to provide and maintain the Services, comply with applicable law, and enforce our policies. You are responsible for Content, including for ensuring that it does not violate any applicable law or these Terms.
This allows users to use the content as they see fit so long as it “does not violate any applicable law or these Terms.” That said, under Section 3(b), OpenAI does stress that some outputs may be the same for other users. In this case, the output wouldn’t be considered “your content.” Vincent leaves us with this: “As the infinite monkey theorem implies, if you give a million monkeys a typewriter and infinite time, they eventually produce Shakespeare. With 100 million active users in January 2023, ChatGPT may be no different.”