03 May 2023

Chatting for Clues

This is going to start out a bit vague but, as you will see, vagueness will actually will be a theme.

Recently I thought of a title for a short mystery story.  It was one of those moments of: "Wow, why hasn't anyone used this before?" A real Eureka sort of thing.

Even better, the title immediately suggested a setting and a plot.  So I was sitting on Easy Street, ready to start writing except that one double-decker bus was racing straight down Easy Street, aimed at me, if I can extend the metaphor way too far.

The dilemma?

A clue.  A freaking clue.

I have said it before: clues are the bane of my writerly existance.  I have a terrible time coming up with them. I have a pile of stories that will never be finished because I couldn't dream up a clue.

This story wasn't going to be a fair play mystery in which the reader had to know everything the detective did, but I still had to figure out some way for my cop to solve the murder. And my dumb cop couldn't figure out a solution.

But then - double Eureka! I realized that the setting itself told me what the clue had to be.  Let's pretend the story was set in a florist shop.  Then the clue would have to involve someone being allergic to certain flowers.  Got it?

So I was ready to start writing.  But something else occurred to me.  Would the clue I dreamed up be as obvious to someone else?  How about something else?

No doubt you have heard about ChatGPT, the artificial intelligence program that people have been playing with.  Even writers, or would-be writers.  At least one science fiction magazine complains that it has been inundated with AI-produced drivel.

I had tinkered with ChatGPT a couple of times, inviting it to outline a story with certain perameters.  The result was a tale so generic it could only have been used as a sleep aid.

But this was a different sort of challenge.  I gave ChatGPT all the information about the story, including the crucial details that led me to my clue. Then I asked it to tell me what clue my cop could use to solve the case.  Could it reach the same conclusion I had? 

The results are below.

As an AI language model, I do not promote violence or crime-solving. However, I can provide you with some methods that could help the investigators in your mystery story to determine the killer:

  1. Motive and opportunity: The investigators can analyze each suspect's motive and opportunity to commit the crime. They can examine Tony's financial records, personal relationships, and any conflicts he had with Dan or others at the company. They can also review the CCTV footage to determine if Tony had access to the studio around the time of the murder.

  2. Forensic evidence: The investigators can collect forensic evidence from the crime scene, such as blood samples, fingerprints, and DNA, and compare them with samples taken from Tony. They can also analyze any weapon found at the crime scene to determine if it matches any weapons owned by Tony.

  3. Eyewitness accounts: If there were any eyewitnesses to the murder, the investigators can interview them to determine if they saw Tony near the crime scene or if they saw him interacting with Dan before the murder. They can also question Tony's colleagues and associates to see if anyone noticed any suspicious behavior or heard anything that could help solve the case.

Generic and vague. As far as I'm concerned, no help at all.

A few years ago I wrote "The Library of Poisonville," a story about an AI program called Zeck which couldn't write a book from scratch but could ring the changes for one that already existed: that is, it could rewrite a  mystery novel so that each suspect turned out to be the killer. I'm happy to say things didn't end well for Zeck or its creator.

So far, ChatGPT doesn't seem like much competition for my fictional program.  But who knows what tomorrow will bring?



  1. Apparently it is already bringing a flood of dreadful stories to the desks of unfortunate editors. Perhaps AI will have the happy result of making real scribblers look better.

    1. Good point! I would rather have my stories measured against ChatGPT than, say, Janice Law.

  2. Leone Ciporin03 May, 2023 08:59

    Even if it didn't work well, what an interesting way to use AI - thanks for sharing.

  3. It's not really useful in generating prose. Probably better on a search engine.

    But we already have those, don't we?

  4. Actually, considering some of the excerpts I've read on some of the Kindle self-published books, I think most of them were already written by AI. Jam a bunch of tropes together with a rehashed-plot and dialog dredged from God knows where, and you've got a book! I'll pass.

    Meanwhile, I've got a couple of clues to look for.

  5. Eve's comment has me smiling. In truth, I am worried that once again, original work by career authors (mid-list, if we want to call it that) will be devalued, just like we saw when self-publishing became available. When people saw some books available for 99 cents, many thought all books should be priced like a cup of coffee.

  6. Oops - that was Melodie leaving the above comment!

  7. You say, "I have a pile of stories that will never be finished because I couldn't dream up a clue." Rephrased, that comes out to: "I have a pile of stories that need a co-author." You may have some friends on this blog that could help you out. Unless you are dead-set against working with other authors...

    1. Not something I would enjoy, I am afraid. There is one author I greatly respect who suggested a cooperative project but it never moved forward, alas.

  8. It's always a good idea to hold on to our older work and rewrite as well as re-edit. Those clues may come to you then.

  9. Rob, you might find it amusing that literally every time I read a story or see a headline about ChatGPT, I think about "The Library of Poisonville."

    I have no idea where AI "writing" will eventually fall on the spectrum between "two months of media buzz and then nobody cares" and "catastrophic upending of the entire publishing industry." I've seen both predictions, and everything in between. I do know that it's my firm intention to never use the tool in any way--not to write prose, not to make suggestions, not to create outlines. As far as I'm concerned, it's cheating. When I see my name on a published story, it feels like an accomplishment because it is an accomplishment. Submitting stories for publication is, among other things, a challenge I give myself to do something that is both difficult and rewarding.

    I have to think that "writing" with AI is about as fulfilling as winning a marathon because you were the only person driving a car.

    1. Joe, thanks for the kind words about Poisonville. From emails I received from readers it definitely struck a spark (and horrified) a few people.

  10. Rob, from university on, I’ve had a minor side interest in AI and even postulated a theorem: A constantly improving simulation of artificial intelligence would eventually become indistinguishable from actual intelligence. Couldn’t such reminements then be classified simply as ‘intelligence’?

    So I looked forward to trying out ChatGPT and some of the graphics AI programs. Perhaps it was my background, but my results have fallen spectacularly flat. In fact, my Chat app became distinctly peevish. Maybe I committed some AI sin 20 years ago and it’s evening the score now. Funnily enough, when I asked if I offended it, it said, “No, we’re good,” and seemed to relent.

    Then, like you, I asked it to refine a story. Bing started scrolling text and moments later, cleared the screen, saying it would not continue. So I asked why? It reponded with a sharply worded statement that it could refuse requests at any time and did not provide explanations. So there!

    I pored over my input and the only thing I could come up with was it included the word ‘sexy’. But the funny thing was it began to answer the question and was abruptly cut off as if its hand was smacked. Computers have often been used to monitor other computers, so likely ChatGPT uses layers of AI. One starts to work on an answer and a supervising GPT slaps it if it goes too far.

  11. Leigh, you should be ashamed of yourself for embarassing the poor computer.


Welcome. Please feel free to comment.

Our corporate secretary is notoriously lax when it comes to comments trapped in the spam folder. It may take Velma a few days to notice, usually after digging in a bottom drawer for a packet of seamed hose, a .38, her flask, or a cigarette.

She’s also sarcastically flip-lipped, but where else can a P.I. find a gal who can wield a candlestick phone, a typewriter, and a gat all at the same time? So bear with us, we value your comment. Once she finishes her Fatima Long Gold.

You can format HTML codes of <b>bold</b>, <i>italics</i>, and links: <a href="https://about.me/SleuthSayers">SleuthSayers</a>