In a recent opinion out of the District of Delaware, Judge Stephanos Bibas revised an earlier decision and, in this instance, granted summary judgment in favor of copyright owner Thomson Reuters and against ROSS Intelligence in connection with ROSS Intelligence’s use of a number of copyrighted headnotes from the Westlaw legal research database to train an AI legal research tool to retrieve judicial opinions responsive the questions provided to the AI tool. The tool did not use copyrighted material in its output and was not considered “generative” AI, which may provide a distinction for purposes of the fair use analysis.
To determine if copyright infringement occurred, the Court determined that the text of the Westlaw headnotes is original, although not overly creative, and that over 2,000 of the headnotes were copied and used to train the defendant’s AI tool. So, the originality of the copyrighted work and the copying elements of the copyright infringement test were satisfied.
Among other defenses asserted by ROSS Intelligence, the Court assessed the fair use defense, which can serve to limit the exclusive rights granted to copyright owners. In copyright law, fair use is statutory. Title 17 United States Code Section 107 provides the four elements should be considered to determine if fair use applies: (1) the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes; (2) the nature of the copyrighted work; (3) the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and (4) the effect of the use upon the potential market for or value of the copyrighted work.
The Court analyzed all four factors and said that factors (1) and (4) weighed most heavily. Notably, the Court found that factors (1) and (4) weighed in favor of the plaintiff/copyright owner and that less weighty factors (2) and (3) weighed in favor of the defendant.
Briefly summarizing the Court’s analysis of the factors:
First, the use made by the defendant was undoubtedly commercial. Defendant ROSS Intelligence sought to commercialize the AI tool and benefit from Westlaw’s copyrighted headnotes without paying a recurring fee for Westlaw. The Court also determined that ROSS Intelligence’s use was not transformative because the purpose and character of the copyrighted work were not changed by the AI tool (if it were, that would likely support a finding of fair use). And, because ROSS Intelligence was not copying source code that had an underlying functionality but the actual content itself to train its AI system, the “intermediate copying” line of cases that might support fair use was distinguished.
Second, the Court determined that while Weatlaw’s headnotes have sufficient creativity to support copyright protection, they are not as highly creative works as a painting or poem might be. Thus, the nature of the copyrighted work factor weighed in favor of finding a fair use.
Third, the Court determined that because ROSS Intelligence did not use the copyrighted material in the AI tool’s output, the substantiality of the public-facing output relative to the amount of copyrighted material was limited. This factor was also found to favor a finding of fair use. An extreme, contrary example would be if a copyrighted photograph were removed from a person’s social media page and reused in its entirety without permission on another’s social media page. There, the amount of the work used would be 100%.
Fourth, the Court determined that the infringement would have a substantial effect on the market for the copyrighted work. Westlaw charges to access its content, and the ROSS Intelligence AI tool is a potentially competing product. This would directly affect Thomson Reuters’ bottom line if and when people switched to ROSS’s tool to avoid paying Westlaw charges.
In sum, the Court rejected the fair use defense and granted summary judgment of copyright infringement to Thomson Reuters. In all likelihood, this will be appealed to the Third Circuit Court of Appeals.
This is one of the first decisions to assess the fair use defense in connection with the use of copyrighted material to train an AI research system. Other courts handling similar cases, and there are several of them, will likely look to this decision for guidance on how to analyze the law in this novel area.
For companies building artificial intelligence systems, this decision should lead business owners to exercise caution as well as due diligence in terms of the content that may be used to train an AI system. Using copyrighted content to train a system could constitute copyright infringement and not be protected as fair use. Consult a qualified copyright attorney to assess your specific situation. And, remember, just because something is on the internet does not mean it is fair game to copy.
