ON AI IN NUMISMATIC PUBLISHING

PREV ARTICLE NEXT ARTICLE FULL ISSUE PREV FULL ISSUE

V28 2025 INDEX E-SYLUM ARCHIVE

The E-Sylum: Volume 28, Number 35, 2025, Article 7

Last week, Len Augsburger wrote that the "September issue of The Numismatist included this notice: 'A large percentage of the July 2025 cover story ('Coins of the Indian Subcontinent,' p. 26) was generated using artificial intelligence (AI) software, a fact the author did not disclose at the time of submission...' " -Editor

"I thought the Coins of the Indian Subcontinent article in the July issue of The Numismatist was far below the standards of what I expect from that publication. It was superficial, skipping over many important aspects and periods of Indian coinage. I was surprised it was the cover article. I assumed it had been written by a high school student.

My experience in using Microsoft Copilot AI for numismatic research is that about 20% of the time it comes up with incorrect information (e.g. incorrect metal content, reporting designs or denominations that do not exist, mixing up United States and Canadian coinage, etc.). It is a tool, but everything must be checked against reliable sources."

The notice also said that "Additionally, since late 2024, our review process includes the use of detection software to help identify AI-generated content." Justin Hinh submitted these thoughts. -Editor

As someone who has been closely following AI development over the past few years, I have to warn that AI detectors are inaccurate at best, fraudulent at worst. There are numerous stories about poor use of AI detectors by college campuses against students. Many of the so-called 'AI Checkers' like ZeroGPT, QuillBot, and others are just fancy frontends overlaid on ChatGPT with simple prompts such as, "Analyze the following text and give a percentage confidence on how much of the text was written by an AI." Of course, these AI detectors proceed to sell a service where they can 'Humanize' texts so that their systems can't detect that an AI wrote it.

ZeroGPT famously got in trouble in early 2023 when its AI claimed that an AI wrote the US Constitution.

This has led to countless false accusations against college students by uninformed professors.

I plugged in Len's response regarding The Numismatist kerfuffle, and it came back saying that only 48% of the text was written by a human.

I'm sorry to say, but AI detectors at this stage aren't reliable. These companies are offering inferior products and capitalizing on people's fears.

As a lifelong techno-optimist, I think today's generative artificial intelligence is a great tool that will only get better in time, but it is and will likely always remain just a tool that requires careful use. We all have to sharpen our critical thinking skills, not allow them to atrophy. -Editor