Legal and Ethical Issues in AI

How should we think about AI? Is it fundamentally a copying technology that copies and creates new things? Or is it more like mining - an extractive process wherein a machine is able to remove resources or data, often at great environmental expense? Are there ways for AI to be useful, and are there ways to use it in the public interest? And regardless of what metaphor you use to understand machine learning and large language models, how do they intersect with copyright and with libraries?

In a wide-ranging conversation data journalist and NYU professor Meredith Broussard, Re:Create Coalition Director Brandon Butler, and Nick Garcia of Public Knowledge addressed these and other questions in Legal and Ethical Issues in AI, part of Library Futures’s Machine Learning and Artificial Intelligence for Information Professionals series.

AI: Copying, Mining, or Something Else?

The discussion began with Brandon (speaking only for himself and not as a representative of Re:Create or its members) asserting that AI is fundamentally a copying technology, noting that the way AI learns this is “a pretty straight line” to the way researchers learn things. Libraries have used copying–from photocopiers to book scanners–to preserve and share the knowledge they have acquired. From a copyright perspective, “really what’s at stake…is not access to expression, it’s access to knowledge”–and who controls that access. Copyright, in his view, isn’t one of the things that’s a problem for libraries and AI.

In Meredith’s view, AI companies are not copiers but miners: an AI model is a machine, and the AI companies are sucking up information on the internet, putting it into the machine, and spewing “new stuff” for which the original creators were not compensated. We have a model from mining: mineral rights. If you’re going to dig up minerals on someone’s land, you pay the landowner. If you’re going to use someone’s work to train a machine, you need to pay them. She did feel that AI (with human oversight) was useful for “boring projects,” including facial recognition in historical photographs and generating metadata.

AI and the Public Good

Nick Garcia spoke to an issue all the panelists agreed on: that AI needs to serve the public good, and that it cannot be controlled by companies whose motives are profit. Nick argued in favor of systems for “public AI,” from public AI datasets to legal language for public interest copyright exemptions (such as the text and data mining exceptions in the EU) to universal service and digital equity programs. Libraries, he noted, are poised to develop and house many such endeavors, as preserving and sharing knowledge for the public good are central to the library mission.

Brandon pushed back on Meredith’s mining metaphor, noting that mining leaves a landscape depleted, whereas copying preserves the originals. In his view AI training is not “taking” expression but deriving facts, and what comes out of a model is a new expression, not a copy. He maintained that using works for training doesn’t deprive authors of their works and that copyright shouldn’t hinder access to useful data for training.

The AI Narrative–And How to Step Back from the Brink

Nick noted that it is important to try AI in order to learn about it and understand it. We should approach new technology with humility, he said, and not immediately dismiss differing viewpoints as we develop narratives about what the technology is and what it can do.

Using metaphors and human-scale images to understand AI is helpful, Meredith agreed. It allows us to focus on what AI can do today and the real harms it causes in the present rather than speculating on the future. AI is “complicated, beautiful math” that produces more accurate outputs with more data, even if we don’t understand the inner workings.

Finally, the panelists addressed their views on how to step back from the brink of negative societal forces related to AI.

According to Nick, we need to challenge powerful entities with money and influence, but locking down AI with restrictions isn’t the answer, as large companies would be better equipped to bypass those restrictions. He suggested drawing inspiration from libraries’ radical principles of providing information and access.

Brandon, meanwhile, reiterated that copyright is a distraction from the core issues of AI use and control, advocating for AI training to be considered fair use. He warned against “walled gardens” created by licensing fees.

And finally Meredith introduced “technochauvinism” to the conversation, questioning whether AI is always the right tool. Librarians have a valuable perspective on these issues, she noted. “Everybody should be listening to librarians more.”

Links from the Panelists

Links from the Chat

Our webinar attendees also had a lot to share!