March 28: Generative AI, large language models, and fair use

     Generative artificial intelligence [GAI] poses novel copyright issues.  The Copyright Office has adopted the position that copyright protects only human authorship, and works created entirely by computer programs are not protected by copyright.  Works created by humans using computer programs as tools may be registered, but protection will be limited to the expression contributed by human authors.  We won't be looking at this question in class.  Most copyright experts believe that the outputs of GAI will be deemed infringing in any situation in which the same output would be infringing if it had been created by humans, and that GAI doesn't present novel legal problems in the context of determining whether its output is infringing.  We agree, and we won't be focusing on that question in class either.  The question we want to focus on this week is whether using copyrighted works to train  GAI is copyright infringement or is fair use.  That question is the subject of a dozen current lawsuits Links to an external site.. 

1.  Start by exploring Sora. Links to an external site.
2.  UM GPT is a large language model.  Ask UM GPT Links to an external site. a question.  Then ask it another question that you hope will generate an answer that infringes someone's copyright. We will ask you to report on the result in class.

3.  Read Timothy Lee's and Sean Trott's explainer for Ars Technica Links to an external site. of how GAI large language models work.
4.  The Copyright Office is conducting a policy study Links to an external site. examining copyright issues raised by artificial intelligence. The Register first held a series of hearings, or "listening sessions" so that stakeholders could express their views on how copyright law should treat GAI, and then invited members of the public to submit comments Links to an external site..  The Copyright Office has received more than 10,000 comments.  Here is a small selection.  Read one of these comments. Pay particular attention to the comment's answer to Question 8, which asks under what circumstances the unauthorized use of copyrighted works to train AI models would be fair use. We will ask you to share the views expressed by the comment you select with the class.

• • • Comment of University of California-Berkeley Library Download University of California-Berkeley Library
    • Comment of National Writers Union Download National Writers Union
    • Comment of copyright law professors Pamela Samuelson, Christopher Jon Sprigman, and Matthew Sag Download Pamela Samuelson, Christopher Jon Sprigman, and Matthew Sag
    • Comment of the News/Media Alliance Download News/Media Alliance
    • Comment of the Authors Alliance Download Authors Alliance
    • Comment of the Internet Archive Download Internet Archive
    • Comment of the Authors Guild Download Authors Guild

5.  The New York Times filed a copyright infringement suit against Open AI, claiming that the use of New York Times content to train Open AI's large language model infringed the Times's copyright. 

• • • Read the New York Times's Complaint Download Complaint
    • Read pages 1-17 of Open AI's Memorandum of Law in support of its motion to dismiss Download Memorandum of Law in support of its motion to dismiss
      • Optional:  Read pages 1-10 of Microsoft's Memorandum of Law in support of its motion to dismiss Download Microsoft's Memorandum of Law in support of its motion to dismiss

6.  Optional: Read Pamela Samuelson, How to Think About Remedies in the Generative AI Copyright Cases Links to an external site., Lawfare February 15, 2024