There’s a disgusting trend reoccurring lately. The most recent misuse of “AI” set me off a ledge regarding the Voynich Manuscript. For anyone unfamiliar, the Voynich Manuscript is a 270-page book that has been carbon-dated to roughly 15th century written in a language using an alphabet no one recognizes. There have been dozens of theories, and dozens of people have been dedicating their lives to deciphering this cryptic text. A computer algorithm recently tried, and failed, to solve it too.
Every single written article regarding this is wrong. This stems from another issue plaguing American media, where journalists will report what another journalist reports without performing the due diligence of thorough research. If the first journalist is inaccurate then everyone else follows suit. It’s Chinese whispers (or the telephone game) in real life. I apologizing for digressing. The point is that the words “Artificial Intelligence” or “AI” are not even in the published source (available here: https://transacl.org/ojs/index.php/tacl/article/view/821/174). Written in the conclusion: “We have presented a multi-stage system for solving ciphers that combine monoalphabetic letter substitution and unconstrained intra-word letter transposition to encode messages in an unknown language.” Throughout the entire published work, the authors discuss how they train a computer to use anagrams and other methods to compare the letters in the manuscript against known letters in various languages. This is called analysis, not AI.
There are several things that the computer is analyzing such as the frequency of letters to find vowels and grammatical structure using Google’s translate. However, we need to keep in mind that Google’s translate product focuses on languages that exists now and not 500 years ago. The texts from 500 years ago would have a drastically different structure. For example:
“man com & se how schal alle ded li: wen yolk comes bad & bare
moth have ben ve awaẏ fare: All ẏs wermēs yt ve for care:—
bot yt ve do for god ẏs luf ve haue nothyng yare:
yis graue lẏs John ye smẏth god yif his soule hewn grit”
The above statement is written in English in 1371 from a monumental brass in an Oxfordshire parish church. Sure, it may be a 50 years older than the Voynich Manuscript, but again, writing wasn’t common until a few centuries later so it’s difficult to compare it. It reads in modern English as follows:
“Man, come and see how all dead men shall lie: when that comes bad and bare,
we have nothing when we away fare: all that we care for is worms:—
except for that which we do for God’s sake, we have nothing ready:
under this grave lies John the smith, God give his soul heavenly peace”
Hebrew, another language used in an attempt to decipher the Voynich Manuscript, is much more complicated. Unfortunately, they are no where closer to solving the mystery and using AI is the wrong terminology to use to describe how they attempted to solve it. At what point would we consider the computer algorithm as an artificial intelligence trying to solve this? To answer this, I need to put analysis in another perspective.
Let’s say I want to sell 500 cars by the end of the year. Next year, I want to sell 750 cars. After each car I sell this year, I will enter in data into a computer: the buyer’s gender, age, ethnicity, gender, job title, employer, whether or not it was a sale, and the car purchased.
At the end of roughly 50 sales I would be able to get a clear picture of what kind of people bought cars. By the end of 500 car sales I would have robust data that would give me a lot of information to be able to use the following year. This information would tell me, for example, that people between 35-45 who are white and female are more likely to buy a car than anyone else. Again, this is just an example thrown out there for the sake of argument. Thus, I would focus most of my efforts on people fitting that description in the next year to reach 750 car sales.
In predictive analysis I’m using that same data to forecast the future. Rather than focusing on white women between the age of 35 and 45, I would begin understanding the cars they would more likely purchase. If a Hispanic 20-year-old male comes to my dealership, I would enter his demographic information into my system. The system would then analyze the data and inform me what car they’re more likely to buy.
Using artificial intelligence would do something entirely different. The concept of artificial intelligence is that it learns from mistakes. In this example, the system would run thousands of simulations to determine the best sales method with the goal selling a car. It would attempt the sale itself, instead of me. In addition to capturing the demographic data of the buyer, it would also review the questions the buyer asked, the tonality used between both the buyer and the system, and how the buyer responds to different answers. Initially, we would expect a lot of failure by the system. When the system makes a sale, it will capture that as a “success indicator.” Then, in the next sale it will try again and most likely fail because each person is different. However, it will keep attempting other methods until it gets another successful sale. Over thousands of attempts it will start to gain a clear picture of who wants what kind of car. It will also learn how to respond to questions that will make the buyer more interested in buying a car.
Returning to the Voynich Manuscript, artificial intelligence could solve the problem by running thousands of simulations of different languages, letters, grammar, and possibly deciphering it. The programmers would need to feed AI the Voynich Manuscript, then say, “we need to translate this to coherent English. Good luck.” A true AI would run hundreds of thousands of simulations, a majority of which will ultimately fail, to solve the puzzle. Even though many people who study it believe they have possibly solved it, they always get countering evidence from other people who also study it. Perhaps AI may be our only chance at solving one of the greatest writing mysteries of all time.