By Jill Hubbard Bowman
A copyright analysis under U.S. law for potential protection of AI technology or potential infringement is complex. The AI processing pipeline for development and deployment of AI technology involves many actors, actions, and types of technology. But any analysis should start with basic questions. It is essential to understand the genealogy of the AI technology and where it fits along the processing pipeline from data procurement and ingestion to training and fine-tuning to deployment and inference to the results interface and output. Surprising information may be revealed about third party rights or pre-existing licensing terms. Rights may change significantly as the technology morphs as it is processed in the pipeline. Actions taken earlier in the processing pipeline may affect potential copyright protection and liability downstream. The nexus of the actions for analysis to the actual model doing inference will also affect the analysis.
Before starting any copyright analysis, ask:
Who?
Who originally created, modified, and deployed the AI technology? Who is using it?
What?
What specifically is the AI technology? What is the format?
When?
When was the AI technology originally created? Modified?
Where?
Where was the AI technology created, modified, deployed, and used?
Why?
Why were the actions taken?
How?
How is the AI technology used or deployed? How is it licensed?
Who?
Understand the actors and actions each took in creating, modifying, processing, and distributing or deploying the AI technology at issue in the analysis. Understand who is performing the actions being assessed. Legal answers will vary depending on who is acting. It’s critical to understand whether it was a human or a machine. For example, did a human make creative choices in the selection and organization of the expressive elements of the work? Was the work generated automatically by a computer? Was a human in control of the creative expression in the work? Who allegedly violated an exclusive right of a copyright owner?
It is also important to understand the type of corporate structure for the key actors. Is the actor a non-profit entity, university, or commercial company? Who created any original copyrighted works in the database used in training and refining the model? Who collected and aggregated the data? Who authored the original model architecture? Who wrote the code? Who modified the code? Who deployed the technology? Who is the end user of the technology?
What?
AI technology in the processing pipelines includes data, databases, algorithms, software, models, user interfaces, and output. For definitions of each element see the AI Key Terms resource at AI Law Maze Map. Numerous terms for AI technology are ambiguous. Understanding exactly what the technology is and its embodied format is foundational for any copyright analysis. Some types of AI technology are not copyrightable subject matter. Some types have thin copyright protection.
When?
When was the original AI technology created? When was it modified?
Where?
Because copyright law varies by country, knowing where technology is created, modified, deployed, and used is essential to any copyright analysis. It’s also important to know where elements of AI technology originated. The internet? A public repository? A professional publication? Underlying third party IP rights and licensing terms will alter the analysis.
Why?
Why was the AI technology created? For public good? For commercial gain? What were the reasons for the selection or arrangement of the technology? Was it for technical considerations or efficiency? What is the purpose in using the technology? Is the purpose for use related to the expressive content?
How?
How is the AI technology used? How is it distributed or deployed? As a development tool? As part of a larger product or service? The relationship to the AI technology and any allegedly infringing actions matters in the analysis. Raw or foundational AI technology like datasets, architecture description and code, models, software libraries, and framework tools may be provided directly to end users to enable AI development. Other AI technology, like models, may be incorporated into AI products or services, including computer vision modules for cars, computers, cameras, and phones or recommendation engines for content or ads. Some AI technology is provided as a service or part of a more complex platform including new generative AI chat bots like ChatGPT, software code generators like GitHub Copilot, and image generators like Firefly, DALL-E, and Midjourney. Also consider how related AI technology has been previously licensed or indemnified for copyright infringement claims.
Once you understand the genealogy of the AI technology and the answers to the basic questions, you can begin to assess the implications of the answers under U.S. copyright law.
