Tom Carey's article, "Examining the Possibility of Compulsory Copyright Licensing for LLM Training" was published on February 25, 2024 in IP Watchdog.

“Rather than rely on a court’s coin flip as to whether the use of the copyrighted work to train LLMs is fair use, it may be better to develop a licensing mechanism that rewards publishers without subjecting AI developers to massive statutory damages.”

ChatGPT and similar generative artificial intelligence (AI) tools rely on large language models (LLMs). LLMs are fed massive amounts of content, such as text, music, photographs and film, which they analyze to discover statistical relationships among these inputs. This process, describe as “training” the LLMs, gives them the ability to generate similar content and to answer questions with seeming authority.

The business community, and society at large, seems convinced that AI powered by LLMs holds great promise for increases in efficiency. But multiple lawsuits alleging copyright infringement could create a drag on development of LLMs, or worse, tip the competitive balance towards offshore enterprises that enjoy the benefits of legislation authorizing text and data mining. A lot seems to hang on the question of whether LLM training involves copyright infringement or instead is a fair use of copyrighted content.

Read the full article on IP Watchdog's website.