Answered By: Victoria Peters
Last Updated: Jul 08, 2024     Views: 9

Preface

This set of guidelines is intended for instructional faculty and staff when creating new open educational resources (OER) or revising existing OER using generative artificial intelligence (GenAI) tools. These guidelines assume that faculty and staff do not have access to the training datasets or validation datasets a generative AI tool uses in order to train its neural networks, and these guidelines are therefore focused on the use of the output of many generative AI tools. 

These guidelines are not intended to give advice on how to use GenAI in the classroom. Please consult the following for guides on GenAI and pedagogy: 

Dr. Jeanne Beatrix Law and Dr. Tamara Powell of Kennesaw State University, through USG's Coursera instance, have created an open course on Education and AI.

Administrative and pedagogical guides using GenAI are included in Chapter 4 of Generative Artificial Intelligence: Practical Uses in Education by Troy Heaps, CC BY 4.0 International License, Open Ed Manitoba.

Copyright / Trademark Fair Use in GenAI

Recommended Resources

Because GenAI platforms often use copyrighted resources on the web to train and validate generative neural networks, a GenAI tool may make an approximate copy of one web resource. It may also transform one copyrighted resource that may or may not be deemed in future court cases as Fair Use via transformation. When publishing OER which uses GenAI to create any part of the new resources, be sure to check on some general information on Fair Use. ALG recommends the following resources: 

Code of Best Practices in Fair Use for Open Educational Resources for the intersection of copyright Fair Use and OER.

INTA Fair Use of Trademarks (Intended for a Non-Legal Audience) fact sheet for how Fair Use works in trademark law.

Assessing Copyright Infringement Based on Output

Generative AI laws and regulations are still in a nascent stage in the United States, and most OER creators will not have access to the data sets used to train or validate a GenAI tool. Until more specific laws and regulations are passed, Affordable Learning Georgia recommends the following: 

Assess using your due diligence whether or not the output of the GenAI tool explicitly infringes on copyright.

  • Example: Prompting a GenAI tool to create a “painting of Sonic the Hedgehog” and then using the output as a decorative image without educational value may explicitly infringe on trademark law with the publisher of Sonic the Hedgehog games, Sega.
  • Example: Prompting a GenAI tool to create an “abstract book cover featuring a beach and a whale” may or may not result in GenAI output that infringes on an artist's copyright. After doing a due-diligence search for identical works and finding none, this work is likely to be able to be included in an open resource and open-licensed as such.
  • Example: Prompting a GenAI tool to write “15 short-answer questions about hydrocarbons” may or may not result in GenAI output that infringes on an author's copyright. After doing a due-diligence search for identical unique phrases and finding none, this work is likely to be able to be included in an open resource and open-licensed as such.

Searching for AI Copyright Infringement

GenAI tools are trained on one or multiple datasets and then validated using both human checking and more data. At the moment, most Language Learning Models (LLMs) such as GPT-4 and Claude use an extensive amount of copyrighted web content in their training sets. To check whether or not an AI output has duplicated copyrighted content: 

For text, search the web. Ensure that this exact language does not exist within copyrighted web content.

For images, do a reverse image search. Ensure that the image generated by the GenAI tool does not duplicate copyrighted web images. (Tools such as TinEye and Google Images can conduct a reverse image search.)

Note: US Copyright Office and AI-Generated Materials

The United States Copyright Office (USCO) has launched a copyright initiative on AI-generated materials. At the moment, USCO will not register any works under copyright that are entirely generated by machines due to US copyright law's requirement of human authorship (see Copyright Registration Guidance for Works Containing AI-Generated Materials). Works including GenAI-generated images, such as a human-authored book containing GenAI-generated images, can still be registered, but the images are not protected under copyright. 

This will likely undergo multiple clarifications, changes, and new legislation in the coming years - ALG will try to keep these guidelines up-to-date with US copyright law surrounding AI-generated content.

Ethical Use of Open Licenses when Using OER to Prompt New Works

It is possible to prompt a large language model (LLM) such as GPT to create new works using existing OER. If you are doing so, keeping the open license of the existing OER and attributing the original works would keep the openness of the original work alive and provide a history of where each resource came from. While open license practices on machine-generated works regarding GenAI are not currently required via the above human authorship requirement in US copyright law, it is an ethical practice to use the original resource's open license on the output of your GenAI prompt: 

Example: “Using https://openstax.org/books/introduction-python-programming/pages/2-2-type-conversion (OpenStax Introduction to Python Programming's section on Type Conversion), create ten short-answer questions on type conversion.” This open textbook is under a CC BY 4.0 International License, so any use of the output should include an attribution of the original work.

If you are using an OER with an SA (ShareAlike) designation in the license to create a new work using Gen AI, the output should both attribute the original work and be shared under the same open license as the original work.

When using more than one OER to create a new work, use the same guidelines as remixing multiple OER by complying with and applying the most restrictive open license: 

Example: “Using https://openstax.org/books/algebra-and-trigonometry-2e/pages/1-4-polynomials (OpenStax Algebra and Trigonometry's section on Polynomials, CC BY) and https://sl2x.aimath.org/book/aafmt/chapter9.html (Elementary Abstract Algebra's chapter on Polynomials, CC BY-NC-SA), create twenty questions on polynomials.” The output should be shared under CC BY-NC-SA (Attribution-NonCommercial-ShareAlike).

Adapted from Affordable Learning GeorgiaCreative Commons Attribution 4.0 International license 

Chat