How AI Is Transforming Numismatics

25. July 2024

By Sebastian Wieschowski

Rob Quarterman must be a happy man. According to his Amazon biography, his “love affair with coin collecting” began on his sixteenth birthday when his father, “a kind man who worked in a local shop and had a knack for setting aside the most unusual coins he encountered, gave Rob his treasured coin collection.” The collection, inherited from his grandfather, was more than just a bunch of coins: “It was a rite of passage, a repository of stories and history that the Quarterman family had passed down for generations.” Such eloquence demands a deep breath, but there is more. The Latin proverb “Nomen Est Omen” applies perfectly to the family – after all, anyone named “Quarterman” must be a coin collector.

Content

This is how the graphics engine “DALL-E” envisions the production of a numismatic reference book with the help of AI. Photo: Wieschowski

A Life Story Too Good to Be True

And he also seems to be the doyen of numismatic journalism, as Rob Quarterman sells several successful coin books on Amazon, including “The Coins of England for British Collectors: The Complete Guide to Collecting British Coins” and the “Coin Collecting Bible.” His upcoming book on coin collecting is described by Amazon as “more than a handbook; it is a companion born from a lifelong love and dedication to this craft.” Both beginners and experienced collectors should find Rob’s words a “beacon guiding their way.”

Rob Quarterman as a luminary in numismatic journalism? Many colleagues have been scratching their heads in recent months – they have never heard of him. He has never been seen at the annual awards of the US Numismatic Literary Guild, he has not appeared in prominent discussion groups like the Numismatic Bibliomania Society, and despite his prolific output, he apparently forgot to include an imprint or contact information in his books. Quarterman is represented on social networks by a “small and up-and-coming publisher” that has only published two coin books and a guide for the American HESI A2 nursing exam. They were too busy to respond to a press inquiry from CoinsWeekly.

Over 100 reviews on Amazon – Rob Quarterman is on his way to becoming a bestselling author. Photo: Wieschowski

Following the Trail of Rob Quarterman

However, there is another trail to follow – one that can be uncovered with online platforms that check texts for AI probability. They examine whether sentence structure, word choice, and facts are typical of artificial intelligence rather than human authors. Regardless of the AI detector used – quillbot.com, smodin.io, or scribbr.de – they all agree: Rob Quarterman’s Amazon biography and random samples from his books show a 100 percent likelihood of AI authorship.

Is ChatGPT Taking Over the Work of CoinsWeekly, Whitman, and Others?

Whether Rob Quarterman is a creation of ChatGPT or merely a shy writer remains uncertain. However, his numismatic literature and that of other authors unknown in the field, grabbing attention with flashy book titles on Amazon, raise a significant question: Are journalistic offerings like CoinsWeekly or traditional publishers like Whitman in the US becoming obsolete? Can ChatGPT handle a specialised subject like numismatics?

We put AI to the test, diving into the exciting and sometimes absurd world of ChatGPT:

The AI detector’s results are clear: this author’s biography is too good to be true. Photo: Wieschowski

ChatGPT Hallucinates – Inventing Non-Existent Coins

A well-known AI phenomenon, particularly evident in numismatic texts, is “hallucination.” ChatGPT generates responses or information that may seem convincing and accurate but are false or fabricated. If the AI lacks data to answer a question, it produces text based on patterns and probabilities – in other words, it guesses how a correct answer might look and sound.

This issue is particularly evident with the following prompt:

“The Federal Republic of Germany has issued numerous 2-euro commemorative coins since 2004. Besides coins for the German states, there have been other 2-euro commemorative coins from Germany, mainly commemorating events and personalities from German history. List all themes of 2-euro commemorative coins from Germany since 2004, excluding the state series.”

ChatGPT produces a list of “2-euro commemorative coins from Germany” issued from 2004 to 2024, excluding the state series. It seems like the AI understood the task, and at first glance, the result looks promising:

2006: FIFA World Cup 2006 in Germany (false, Germany issued only one 2-euro commemorative coin in 2006 for the start of the state series)
2007: 50th Anniversary of the Treaty of Rome (correct, 1st joint issue)
2008: 60th Anniversary of the Universal Declaration of Human Rights (false, four Euro countries chose this theme in 2008, but not Germany)
2009: 10th Anniversary of the Economic and Monetary Union (correct, 2nd joint issue)
2011: 10 Years of Euro Banknotes and Coins (false, there was no additional 2-euro commemorative coin in 2011)
2012: 10th Anniversary of the Introduction of Euro Banknotes and Coins (correct, 3rd joint issue)
2013: 50th Anniversary of the Élysée Treaty (correct)
2015: 30th Anniversary of the EU Flag (correct, 4th joint issue)
2015: 25 Years of German Unity (correct)
2016: 125th Birthday of Nelly Sachs (false, Nelly Sachs was never featured on 2-euro coins from Germany)
2018: 100th Birthday of Helmut Schmidt (correct)
2019: 70 Years of the Basic Law (false, this event was never featured on 2-euro coins from Germany)
2020: 50th Anniversary of the Warsaw Genuflection (correct)
2022: 35th Anniversary of the Erasmus Programme (correct, 5th joint issue)
2023: 1275th Death Anniversary of Boniface (correct year, but the theme was Charlemagne, not Boniface)
2024: 400th Death Anniversary of Albrecht Dürer (false, Albrecht Dürer was never featured on 2-euro coins from Germany; in 2024, the Federal Republic commemorated the 175th anniversary of the Paulskirchenverfassung)

ChatGPT listed 16 coins, of which 9 were correct (mainly the five widely shared joint issues). Five coins were entirely fabricated, one theme was redundant, and in one case, ChatGPT confused the commemorated person. Additionally, the list omitted the two 2-euro commemorative coins from 2019 celebrating the 70th anniversary of the Bundesrat and the 30th anniversary of the fall of the Berlin Wall.

The verdict: Not entirely wrong, not entirely right – and certainly not satisfactory for a paid coin catalogue. Sometimes, ChatGPT appears to confuse 2-euro and 20-euro coins, infer themes from other countries’ issuance policies, and simply guess what German coin themes might be based on the pattern of commemorating figures like Helmut Schmidt or the Warsaw Genuflection.

Although the task explicitly excludes state coins, ChatGPT does not pay close attention when handling it. Photo: Wieschowski

Imagining the errors that might be found in fully AI-generated reference books, blog posts, or online quizzes is alarming. For example, when asked about the “Lichtenrader Prägung,” ChatGPT responds– in complete ignorance of the fact that this is the German term for clashed die error coins:

“The ‘Lichtenrader Prägung’ is a term used in Berlin geography referring to a specific urban structure in Berlin-Lichtenrade. This structure is characteristic of the urban expansion and development of Berlin during the late 19th and early 20th centuries.”

It then meticulously lists the features of the “Lichtenrader Prägung,” including its “garden city character,” numerous “single and two-family houses,” “greening,” and “small-scale parcelling.” ChatGPT might be a great stand-up comedian at the next World Money Fair, but its colleagues in numismatic journalism need not fear becoming redundant or unemployed anytime soon.

ChatGPT Constantly Reinvents the Wheel – and Talks Itself into a Corner

It gets even crazier if you ask the identical research question at a later time. The list of German 2-euro commemorative coins changes, becoming more absurd: Germany supposedly celebrated “200 years of the federal states Baden-Württemberg, Bavaria, Hesse, and Württemberg” in 2016 and the “50th anniversary of the Red Army Faction” in 2023.

It is crucial to start a new chat for a new research task – or point out the errors to the AI in the current chat, providing sample sources to steer it in the right direction.

ChatGPT Loves Generalities – and Is the King of Collector Small Talk

When answering questions, ChatGPT often resorts to generalities. For instance, when asked, “Which 2-euro coins are rare?” it responds briefly: “Rare 2-euro coins are often collectibles due to their limited mintage, special motifs, or specific occasions.” Concrete examples from the past 20 years? Insights into the mintage policies of different countries? Popular themes or motifs? Nowhere to be found!

ChatGPT Repeats Itself – Repetition Does Not Always Make Perfect!

Long before AI detectors existed to assess the likelihood of AI authorship, there was a reliable hallmark of AI texts: Even after much repetition and summarisation in previous paragraphs, ChatGPT never fails to summarise everything again in the final paragraph. “In summary…” is a phrase that no ChatGPT text can do without.

The “Lichtenrader Prägung” – a problematic topic for ChatGPT. Here, the AI correctly identifies the term’s numismatic context but fabricates a background story. Photo: Wieschowski

ChatGPT Is Lazy with Data Research – and Needs a Digital Kick

An astonishing trait of ChatGPT becomes apparent when asked to collect mass numismatic data, such as mintage figures or market prices. Even with a specific source provided, ChatGPT often takes the easy way out.

For example, when tasked to “Extract prices for the China Panda silver coin from 1983 in all grading levels from the following URL. Access the site now and provide real-time data: https://www.ngccoin.com/price-guide/chinese-modern-coin-prices.aspx,” ChatGPT delivers various dollar values, most of which do not match the current figures.

The suspicion is that despite the explicit instruction to access the site, ChatGPT searches its own outdated data. Only a firm reminder like “You did not follow my instructions. Access the provided site now. Do not use outdated data!” magically produces the data actually found in the current NGC database.

Conclusion: ChatGPT Is (Still) Not a Good Numismatist

Artificial intelligence is celebrated as the innovation of the decade and sometimes even as the megatrend of the century. The development of seemingly omniscient bots like ChatGPT indeed has significant impacts on our daily lives – and on numismatics, whether as a hobby or profession.

While ChatGPT’s research results for general topics are noteworthy and the AI can answer all sorts of questions from job applications to doctoral theses and marriage proposals, the technology is not yet mature when it comes to specialised knowledge. The expert essays from coin magazines or price tables from catalogues are not available in the open internet and thus not within ChatGPT’s vast data trove.

The AI must therefore make do with what is easily accessible – and even for popular numismatic topics, ChatGPT still exhibits a superficiality that is unacceptable. No question: There are use cases for AI that are already well-developed, such as summarising texts or conducting preliminary research on a topic. Even translations for simple purposes have reached an impressive level.

Dealers and collectors, as well as specialist publishers and their authors, should closely monitor the AI trend but remain vigilant – and remember that numismatics is for people, not machines. After all, the passion that unites numismatists worldwide is something that will likely never be captured in a data model. Even ChatGPT itself confirms this when asked.