Artificial intelligence programs that analyze and generate text are changing the way we read and learn. To parse the sentences, the AI model looks at text cues such as word choices to see how they are related.But what if those clues are deliberately vague and confusing? Cain’s Jawbone, 1934 Murder Mystery Puzzle Book.
This book came into my life as mysteriously as any literary detective would like. One afternoon in October 2022, a random package from Amazon was dropped on my doorstep.I had never heard of the book inside, but a Google search turned up that Cain’s Jawbone It’s part murder mystery and part brain teasing puzzle. This book was intentionally published with all pages out of order. To solve the case, the reader must first rearrange the pages, then he must name the six murderers and their victims.
advertisement
The author of this heinous conspiracy was (surprise surprise) a puzzle expert. Edward Mathers worked as his crossword compiler. observer Publishes a newspaper under the pseudonym Torquemada.he published Cain’s Jawbone It was the height of the so-called Golden Age of Detective Stories, but only two were able to solve it before the book went out of print. In 2019, John Mitchinson, co-founder of the publishing platform Unbound Publishing, came across a copy of the story and its solution in a British literary museum and decided to reprint the 100-page puzzle. “I was like, ‘Well, this is great. It’s a detective story, so it’s hard to organize, isn’t it?'” he recalls.
The answer is “very difficult”. Only four other people have solved this puzzle in the last few years. The book then went viral, thanks to a few of his TikTokers trying to rearrange the pages with a colorful “murder wall.” Its newfound popularity prompted Mitchinson to print additional copies on top of his initial run of 5,000 copies.
when my copy Cain’s Jawbone Instead of designating page wall space, my husband and I spread them out on the guest bed. One dark night, as I perused a flowery, intentionally nebulous language, I suggested using an AI algorithm to solve the novel.
advertisement
I’m not a software expert, so I started looking for AI companies to tackle this puzzle. But most AIs aren’t specially trained to sort the pages of a book or analyze the linguistic quirks of 1930s English. Finally, we reached out to Zindi, an Africa-based company that hosts an AI competition where 50,000 data scientists use algorithms to solve puzzles and win prizes. Zindi was interested in running a contest, and with Unbound’s approval, I created his 2022 Cain’s Jawbone Murder Mystery Competition. We digitized his 90-year-old book and challenged the world to rearrange the pages using natural language processing (NLP) algorithms.
NLP algorithms, such as the famous ChatGPT, try to understand the information in the text by comparing the context and language of the text with the training data they receive. Such algorithms can analyze text like never before by converting each word into a “token” and analyzing how each token fits into the complete task. This allows AI algorithms to quickly and effectively analyze text, whether it is literature or scientific reports. Nobly resisting using AI to unravel the case of the person who sent you this interesting book, don’t text him to a friend or post on Instagram to reveal the culprit. bottom.
In our competition, participants started with an existing NLP model called BERT. BERT was developed by Google and is available as an open-source library that can be modified for specific uses. “These models are trained on only the gobs of data available to the creator of the model and are refined to follow a specific set of instructions,” says his research associate in computer science at Southern University. Professor Jonathan May said: California. In order to refine the model for this particular application, we asked participants to read Agatha Christie’s first mystery novel, The Mysterious Case of Stiles, The story was written around the same time, so we use it as training data Cain’s Jawbone It contains similar words, giving clues to the context of the classic mystery.
advertisement
AI has a long history of writing novels containing murder mysteries. In 1973, computer scientist Sheldon Klein proposed an automatic novel writer that could create his 2,100-word murder mystery story within 20 seconds of him. Since then, programmers and engineers have improved the output of these models with more data. “In some ways, murder mysteries are easy,” says Mike Sharples, Emeritus Professor of Educational Technology at the Institute for Educational Technology, The Open University, England. “It has a standard plot structure: find a corpse, detectives come, got a herring, etc.” It also helps the language program (in theory) to try to put these confused pages of stories back into the correct order.
Unfortunately, Cain’s Jawbone It creates the ultimate challenge for linguistic analysis algorithms. The narrative is not only completely chaotic, it is designed to confuse the reader. ), which is intentionally vague to make the ordering of the pages as difficult as possible. Additionally, the story is full of false cues, including pseudonyms and misleading names for some characters, which can confuse AI models and human solvers. As a result, no AI developer was able to solve the puzzle, but some made a little headway.
MG Ferreira, an econometrician from South Africa, was one of the AI contest winners, with a top score of 42%. This means that his program correctly sorted his 42 of his 100 pages. “NLP has an understanding like knowing that thunder and rain go together,” says Ferreira. “But the problem here is that the book is trying to mislead you with false clues. It breaks your understanding of NLP.” You have to look and identify which ideas work, he explains. “If we go in that direction, eventually we will be able to solve everything. is called machine-assisted,” he adds.
advertisement
The Murder Mystery Contest has revealed that current AI language programs may be capable of impressive feats, but they won’t be on par with Poirot anytime soon. These models are not very good at analyzing things without context, which can pose problems for researchers trying to analyze ancient languages using NLP. The lack of context makes it difficult for AI to learn how to translate lost languages, as there are few historical records about civilizations long ago.
At least this experience helped me solve one puzzle. I tracked down the person who sent me the book and embarked on this quest to solve it. The culprit was one of my elementary school friends, who doesn’t have social media, but likes murder mysteries like me.
This is an opinion and analysis article and the views expressed by the author or authors are not necessarily Scientific American.
