Proposed Census Changes/AI Chicanery

Drastic changes have been proposed for the 2030 Census. Should the changes go through, they would greatly reduce our knowledge of the population and would radically alter the provision of public services and resources. Census data is used by our library patrons across many disciplines, so the changes would greatly impoverish research and writing. Many higher education organizations have signed onto a letter opposing these changes. I don’t look forward to having to explain to a future undergraduate why, for example, race and ethnicity data collection simply stopped with the 2030 Census. There won’t be any good reason I can give them since the justifications given for the changes are specious and ahistorical.

While these regulations would be challenged in court the instant they’re enacted and may be blocked, the fact that they’re even being proposed will probably serve to cause confusion and depress response rates for the 2030 Census. Their proposal also comes at a critical juncture for the Census Bureau because they’re already knee deep in 2030 operational planning and testing. This uncertainty needlessly complicates that process.

Fortunately, you can comment on the proposed regulation until November 2nd and I encourage you to do so. (The comment deadline has already been extended once, but I don’t count on that happening again.) I found the Hands Off Our Census Comment Guide helpful in putting my comment together. Even if you don’t comment, it will be important to our profession to be aware of the implementation of these changes.

---

In the past month, I’ve seen a couple examples of AI chicanery in our library. First, on the morning September 29th, Christina sent an email to our reference ticket system seeking a digitized copy of a University of Pittsburgh PhD dissertation from 1969. The email address given by the user was a Gmail address. After a response from one of my colleagues seeking more information, Christina sent an odd reply in a couple of respects. The text was on separate lines rather than in a paragraph form. Also, they noted that the author was dead and 50 years had passed so “copyright is really a moot issue.” That’s not at all how copyright works. On the morning of September 30th, my colleague suggested requesting the text via ILL as an option and that satisfied Christina, ending the interaction.

About the same time as we sent the email response to her, Christina contacted our chat service using the same email address and seeking the same dissertation. A different colleague directed her to the same resources as she got in the email. She again made an odd copyright assertion-Pitt owns the copyright to all Pitt dissertations. (We don’t, of course.) My colleague tried to move the conversation along by telling them we couldn’t send them the dissertation because they weren’t a Pitt affiliate. Christina responded by saying “This is rather pathetic to erect FAKE arguments to erect barriers.” Christina got more belligerent from there, so my colleague ended the conversation. About an hour later, we got a chat from Ron, who stated he needed a librarian. Once the chat was initiated, he asked whether my colleague was “an experienced Ref lib on campus ? or a student sitter?” Eventually, Ron asked about the same dissertation that Christina did. It was explained to him that he could request the dissertation via ILL. At this point, Ron had a temper tantrum and started insisting the university had copyright and questioning our logic. My colleague didn’t respond, so Ron logged off. He rated the chat interaction a 0/10 and said we were “terrible.” Sorry, Ron!

After looking at all these interactions, we concluded they were the work of an agentic AI, which Alex wrote about in June. We reached this conclusion primarily because all the interactions were initiated from the same IP address in California. The odd syntax and belligerence were other tip offs. We also learned that other university libraries have had similar encounters. We banned the IP address, but that’s a short-term solution when an AI agent can presumably use (or spoof) any IP address. Like other libraries, we’re now working on policies and language to address non-human queries.

In the second example of AI hijinks, a company called Eldil contacted me in my role as the FDLP depository librarian on October 6th. The email stated that they were “looking for older technical and scientific material that still exists on microfiche or microfilm, especially government technical reports” with the idea that they’d acquire them. I had never heard of this company, and their website states they “track down and purchase hard to find print media.” That’s basically the extent of the information presented on the website – nothing about ownership, products, etc. Super sketchy to say the least. It turns out they’ve messaged many FDLP depositories and one of my fellow Depository Librarians said that Eldil are “a 3rd-party, venture-funded AI start-up company sourcing content for AI model-training by an undisclosed list of clientele described as ‘frontier AI labs.'” Needless to say, we won’t be doing any business with them!

These two stories demonstrate the importance of acquiring high-quality information about, and delivery of information to, our fellow citizens and patrons. The human-centered nature of our library services is one of our greatest strengths. We need to lean on that orientation to steer our libraries through uncertain times.