Over 30 lawsuits challenging the training of Generative AI on copyrighted materials are pending, most in federal courts across the country. The copyrighted materials range from news stories to photographs to music. The law is unsettled whether such training violates copyright law.
However, the uncertainty means that training can continue to until we get final guidance from the courts (or the legislature), both of which take time. For example, Thomson Reuters, which provides Westlaw, sued a competitor for copyright infringement in May of 2020. Last month, the Court partially granted summary judgment finding that the headnotes and numbering were copied. The case is still pending trial.
Of course, it will be impossible to “untrain” the GenAI engines, which are and will be in use. This may leave the courts to grapple with the difficult question of what remedy is appropriate, if and when it is determined that such training does not constitute “fair use” or fall within an exception for data mining.
In decision issued Tuesday, Judge Eumi K. Lee ruled that it remained an “open question” whether using copyrighted materials to train AI is illegal – meaning UMG and other music companies could not show that they faced the kind of “irreparable harm” necessary to win such a drastic remedy.