Since ChatGPT gained national prominence in November 2022, the ethical facets of AI use have become more and more complicated. Amaranth helps humanists to navigate the fast-changing dynamics of the debates over ethical uses of generative AI products.
The AI Steering Committee at UNM has drafted a Statement of Values, Ethics, and Principles for Artificial Intelligence at UNM. This document provides an excellent framework for an ethical approach to use of generative AI tools for teaching and research. At Amaranth, we offer advice on how to match the goals of documents like this one to specific situations in the classroom and in the archive.
An ethical use of AI will often hew closely to the purpose of the specific project. For a research paper submitted as a course requirement, the instructor will provide guidance on what parts are AI-friendly and what aspects of the research process must be done without AI. Ethical use, in this case, means following the guidelines from the course instructor.
Scholars considering the use of AI when preparing a piece for publication face a different set of ethical questions. The debates over AI disclosure continue and rules in place today might not apply a year from now. Princeton University Library maintains a helpful guide on disclosure policies from several large academic publishers.
Amaranth consults with researchers about finding the best ethical and productive way to leverage AI for specific humanities projects. Email us at amaranth@unm.edu to book a consultation.
At this time, an advanced use of generative AI will likely require a corporate frontier model such as Anthropic’s Claude, OpenAI’s ChatGPT, or Google’s Gemini. An ethical use of these applications requires understanding some background about how they were made.
All of the frontier AI models were created by a process of large-scale data scraping which could have violated intellectual property law and copyright law. Litigation, with The New York Times as lead plaintiff, is currently pending. In addition, the models in their initial stages require human beings to censor anti-social output deemed unfit for public release. The frontier AI companies employed low-wage labor for this work, known as Reinforcement Learning from Human Feedback. Similar to the content moderation labor used by social media companies, there are concerns about how moderation work for AI caused psychological trauma. The construction of data centers to run the frontier AI models, in the past and in regard to future plans, has been criticized on environmental justice grounds. Many of these concerns are not unique to AI and LLMs, but the rapid advacnement and infusion of those tools into digital products have rightfully brought them into focus.
The concerns also go beyond how the models were built. Anything you put into a commercial AI tool, including unpublished research, student writing, or oral histories shared in trust, may be stored by the vendor and, depending on the product, used to train future models. What comes out can be fluent, confident, and wrong. In the classroom, a tool that does the struggling for students can take away the learning the struggle was for. And a handful of companies now make the tools that all of higher education is being asked to adopt.
Keeping these ethical considerations in mind implies that care should be taken when deciding when to use the frontier models. For some this means using frontier AI models sparingly, and they will focus on running local LLMs to learn about AI. For others this means using frontier AI models while searching for ways to choose and promote companies that address ethical considerations with something more than lip service.
Thoughtful scholars come down in different places on all of this. Some decide the costs outweigh the benefits for their work, and that’s a defensible choice. Others use AI for specific tasks where they can check the results and keep it away from everything else. We’re not here to make that decision for anyone. We want people to make it with a clear picture of what these tools are, how they were built, and what they do well and where they are not yet reliable.
Ethics notwithstanding, the genie is not going back in the bottle.
Students are already using AI, research databases and publishers are building it into their platforms, and it is changing how scholarship gets produced and evaluated whether or not any of us use it. Everyone in higher education now needs to know how to navigate it: what it can do, where it fails, and how to tell the difference.
For faculty, the best place to build that understanding is usually your own research. Try it on a problem you know well, such as documents you’ve read closely, a translation you can check, or a literature whose gaps you already know. You’ll see its possibilities and its failures more clearly than any general overview can show you. The goal is awareness, not necessarily adoption, and you may decide it has no place in your work. Either way, that firsthand judgment is what students need from you in the classroom, where many of them are already using these tools with far less context. This should be based on informed and skilled use of AI, not stereotypes and cursory prompts that can predictably produce unhelpful output.
Amaranth is here to help with both: exploring what AI might open up in your research, and bringing it into your teaching at whatever level feels right to you, from a single class discussion about its limits to an assignment built around it. Book a consultation or drop by studio hours.