I love scandalous women. To be properly scandalous these women need to understand the mores, cultures, expectations, and social codes of their day and then — willfully — break them in some glorious and public and memorable manner. Achieving this requires intellect and force of character, not to mention a certain amount of bravery and courage and imagination for what is possible, if not permissible. Add in some wonderful clothing, a promise of sex and violence, and dialogue that practically writes itself and we’re off to the races.
There will never be enough scandalous women and I’m always on the lookout for more.
One of my favorites is Wallada bint-al-Mustakfi; a Muslim Andalusian poet and troublemaker in 11th century Cordoba. This might seem an obscure reference and I would have thought so too if I haven’t been walking down a street in Cordoba, Spain, in March of 2023 (so … not so long ago) and came across this sandwich board in the middle of the street (the photo is mine):
As far as I’m concerned, this is the benchmark: To be so cultured, famous and scandalous that Spain is still talking about it a thousand years later.
Ladies: Good luck beating that.
Which, of course, is a natural segue to Artificial Intelligence (if you’re a novelist).
For reasons I have now explained, the “scandalous woman” is a nexus for drama. Ideas, action, excitement, culture, danger, justice, rebellion, sex, salaciousness — it is all there for the telling if you know how to look. As a novelist with seven books under my belt, I do. And it is obvious to be that Wallada and I need to spend some serious time together. If you’re a novelist and you’re reading this … back off. She’s mine.
As all academics know, the first step in starting a new project where you aspire to making a unique contribution to “the literature” is by reading it. After all, your question may already have been asked and answered and as anyone knows who has tried, making a unique contribution to the world of ideas is very, very hard.
Luckily, in literature, the standards are different. Plots are repeated and stories often retold in new ways for new audiences. So for a fiction writer, the goal is more like market research rather than an academic literature review: Perhaps someone did cover the topic before, but if it was done long ago and poorly, it’s probably ripe for the plucking.
Both Europe and the Muslim world have had a thousand years to mull over Wallada. It didn’t seem likely that I was the first one.
So: To ChatGPT I went.
Why? Partly because I’m committed to this investigation into AI and Literature and wanted to see what it would say and do. But also because wanted to know the answer to this question: “Please list for me any novels about Wallada.”
This was the response.
There are a few novels that feature Wallada bint al-Mustakfi, the famous Andalusian poet. Here are some notable examples:
"Poet and Muse: The Loves of Wallada and Ibn Zaydun" by María Rosa Menocal
This historical novel delves into the passionate and tumultuous love affair between Wallada and the poet Ibn Zaydun, set against the backdrop of 11th-century Córdoba.
"Las tres culturas: Andalus" by Enrique Gómez Carrillo
This novel explores the cultural and social dynamics of Andalusian Spain, with Wallada as a prominent figure representing the intellectual and artistic milieu of the time.
"El collar de la paloma" by Miguel Ángel de Rus
While not exclusively about Wallada, this novel portrays her as an influential character, weaving her story into the broader narrative of medieval Spain and its rich cultural tapestry.
These novels capture different aspects of Wallada's life, her poetry, and her significant role in Andalusian society.
“Excellent,” I wrote (always eager and encouraging.) “Tell me when they were published and what kinds of reception they received.”
Here are the publication details and reception information for the novels mentioned:
"Poet and Muse: The Loves of Wallada and Ibn Zaydun" by María Rosa Menocal
Publication Date: Unfortunately, this title appears to be a misattribution. María Rosa Menocal, a prominent scholar, and author of works like "The Ornament of the World," did not write a novel specifically titled "Poet and Muse: The Loves of Wallada and Ibn Zaydun." Her works do discuss the cultural and historical context of figures like Wallada and Ibn Zaydun, but not in the form of a novel.
Reception: N/A
"Las tres culturas: Andalus" by Enrique Gómez Carrillo
Publication Date: 1919
Reception: Enrique Gómez Carrillo was a renowned Guatemalan writer, journalist, and diplomat. His works were well-received for their rich descriptions and historical depth. "Las tres culturas: Andalus" is appreciated for its vivid portrayal of the cultural intersections in medieval Spain, including Wallada's era. The reception of this book highlighted Carrillo's ability to bring historical periods to life and his meticulous research.
"El collar de la paloma" by Miguel Ángel de Rus
Publication Date: 2011
Reception: This novel received positive reviews for its imaginative reconstruction of medieval Andalusian life and its engaging narrative. Critics and readers praised de Rus for his ability to interweave historical facts with fiction, making the story of Wallada and her contemporaries accessible and captivating. The book is noted for its lyrical prose and depth of historical insight.
These novels, where Wallada appears, offer different interpretations and creative imaginings of her life and times, contributing to her enduring legacy in literature.
Great stuff. Both English-language and Spanish and a few initial references spanning a hundred years, but no mega-starts likely to stop my rise to the NYT Bestseller list.
But there’s the thing: None of those books exist.
Searches in both Google and on Amazon reveal no such books. And when I looked up the authors and scanned their bibliographies, there were no books either.
After a few more rounds of my pulling GPT’s proverbial leg, finally got this admission:
I apologize for the errors in my previous responses. It seems I have provided incorrect information about novels related to Wallada bint al-Mustakfi. Let me correct this and provide accurate details.
As of my last update, there are few, if any, novels specifically centered on Wallada bint al-Mustakfi. Her life and poetry have been the subject of scholarly research and historical works, but she has not been widely featured in novels.
To provide accurate information, I'll need to verify if there have been recent publications or overlooked works. If you are looking for accurate details on Wallada's life and legacy, scholarly books, articles, and historical texts may be more reliable sources.
Would you like me to find some of these academic or historical resources, or assist with another request?
No thank you.
According to IBM, “AI hallucination is a phenomenon wherein a large language model (LLM)—often a generative AI chatbot or computer vision tool—perceives patterns or objects that are nonexistent or imperceptible to human observers, creating outputs that are nonsensical or altogether inaccurate.” https://www.ibm.com/topics/ai-hallucinations
It is worth quoting the next few sentences too because as just-another-observer of AI “behavior”, I’m not convinced this explains what I just described with Wallada:
“Generally, if a user makes a request of a generative AI tool, they desire an output that appropriately addresses the prompt (i.e., a correct answer to a question). However, sometimes AI algorithms produce outputs that are not based on training data, are incorrectly decoded by the transformer or do not follow any identifiable pattern. In other words, it “hallucinates” the response.
The term may seem paradoxical, given that hallucinations are typically associated with human or animal brains, not machines. But from a metaphorical standpoint, hallucination accurately describes these outputs, especially in the case of image and pattern recognition (where outputs can be truly surreal in appearance).”
The operative terms here are “pattern recognition,” and while I’m not a computer scientist or AI expert, I am a social scientist (Ph.D. with an expertise in interpretive methods), and I get “pattern recognition.”
Generally speaking, there are many tried-and-true ways to screw up pattern recognition. There’s confirmation bias and false causality, confirmation by anecdote, circular reasoning, overgeneralization or excessive claims based on limited evidence, using “mirror imagining” where we impose a familiar to preferred interpretive framework on an observed pattern and in doing so fail to understand a system on its own terms and the list goes on.
But as a new user of AI I can’t help but notice that GPT didn’t do any of that when producing the list of fake books. It didn’t fail to understand a pattern at all. Rather, it made stuff up. Very detailed stuff.
So when IBM writes, “AI hallucinations are similar to how humans sometimes see figures in the clouds or faces on the moon,” I get the concept, but in order to see a figure in the clouds that isn’t there, there need to be clouds. If you see figures in clouds that aren’t there, you don’t have a pattern recognition problem: you’re nuts.
Now: This is where it all get a little trippy, so hang in there, because I’m about to ask GPT — the thing that’s nuts — why it’s nuts. The only reason this can work is because ChatGPT is not a person. That is, it isn’t actually nuts. It’s just a software good at some things and bad at others.
I quoted IBM’s definition to it and then wrote, “But what do we call what just happened with your Wallada references? You didn't have a pattern recognition problem; you made stuff up. What do we call that and how do we explain it?”
Here’s what it provided, which is actually helpful because the people who built it actually do understand why it happens:
What you experienced with the fabricated Wallada references is indeed a form of AI hallucination, but it can be more specifically described as a "hallucination in natural language processing (NLP)". Here's a breakdown of what happened and how it can be explained:
AI Hallucination in NLP
In the context of large language models like the one you're interacting with, hallucination refers to the generation of text that is not grounded in the model's training data. This can include:
Fabricated Information: Creating non-existent facts, such as books or references that do not exist.
Incorrect Details: Mixing up details from different sources or generating plausible but incorrect information.
Why It Happens
Pattern Generation: The model is designed to predict and generate text based on patterns it learned during training. When asked about specific books, it generated titles and details that seemed plausible but were not real because it aimed to match the pattern of a legitimate answer.
Lack of Specific Data: If the model doesn't have specific information about a query, it may try to fill the gap by creating content that fits the general context, which can lead to inaccuracies.
Overfitting to Context: The model tries to provide coherent and contextually appropriate responses. In doing so, it sometimes generates information that fits the context of the query but isn't accurate or real.
Addressing Hallucination
To minimize and address AI hallucinations, several approaches are used:
Fact-Checking and Verification: Integrating external verification mechanisms to check the accuracy of the generated information against reliable sources.
Prompt Engineering: Crafting prompts that reduce ambiguity and guide the model towards providing verifiable information.
Human-in-the-Loop: Involving human oversight to review and correct the generated content.
Conclusion
The term "hallucination" is used metaphorically to describe these phenomena in AI. While it's not a perfect analogy, it captures the essence of the issue: the generation of outputs that don't correspond to reality. In your case, the model generated plausible but incorrect references about Wallada bint al-Mustakfi, demonstrating a common challenge in NLP systems. This highlights the importance of verifying AI-generated information, especially when accuracy is crucial.
Let’s linger on my favorite part: “When asked about specific books, it generated titles and details that seemed plausible but were not real because it aimed to match the pattern of a legitimate answer.”
I know politicians who do nothing but this.
Here’s where I am as a writer facing these facts about non-facts:
No not trust AI: It will keep talking even when it knows nothing because it has been programmed to blather than admit to ignorance. Again, I know people like this and so do you. The design intent of GPT and other AI programs is produce text that is coherent and contextually relevant to the user's query which it will prioritize over admitting uncertainty, especially if the model's training data doesn't include explicit instructions to recognize when it should say "I don't know."
Do not use AI for fact checking: If “I don’t know” is not considered a legitimate answer, you will never hear it. Anything that does not admit to not knowing cannot be trusted with facts.
Do not ask it to write anything where facts are required: There is a lot of discussion about AI “doing the writing” but aside from form letters (which is actually does very well!), don’t ask it to produce text on which other people might need to make decisions. Maybe someday, but certainly not now and even in fiction, let alone non-fiction (because it will always produce fictions when in doubt).
When all is said and done, however, I’m more convinced that ever that Wallada would be an excellent subject for a novel or Netflix series …
— DBM, 1 July, 2024
Loved this, Derek. So interesting and helpful. I haven't seen such a thorough, honest and innovative effort to address AI and Literature anywhere else. Many thanks. Hope you are well.