Asking the Computer – Wednesday Blog by Seán Thomas Kane
This week, I address news that the latest version of ChatGPT will help with your math problems.
I’ve used ChatGPT on occasion, mostly to test the system and see what it will do if I prompt it about very particular things. What does it know about André Thevet (1516–1590), or about the championship run of my beloved Chicago Cubs from the 80s, the 1880s that is. I even asked it questions in Irish once and was startled to see it reply with perfect Irish grammar, better than Google Translate does. I’ve occasionally pulled up my ChatGPT app to ask about the proper cooking temperatures of beef, pork, or chicken rather than typing those questions into Google, and in one instance I used it to help me confirm a theory I had based on the secondary literature it had in its database for a project I was writing. The one thing that I would’ve expected ChatGPT to be best at from the start are logical questions, especially in mathematics.
There are clear rules for math, except that in America it’s singular in its informal name while in Britain it retains its inherent plurality. As much as I acted out a learned frustration and incomprehension when posed with mathematical questions in elementary, middle, and high school, I appreciate its regularity, the way in which it operates on a universal and expected level. Many of the greatest minds throughout human history have seen math as a universal language, one which they could use to explain the world in which we live and the heavens we see over our heads. The History of Science is as much a history of knowledge as it is the history of the development of the Scientific Method, a tool which has its own mathematical regularity. All our scales and theorems and representations of real and unreal numbers reflect our own interpretation of the Cosmos, and so it is logical that an advanced civilization like our own (if I may be so bold) would have developed their own language for these same concepts which are inherent in our universe. Carl Sagan took this idea to a fuller level in his novel and later film Contact, in which the alien signal coming from Vega is mathematical in nature.
Often, the lower numbers are some of the easiest words in a language for learners to pick up on. The numbers retain their similarities in the Indo-European languages to the extent that they were used as early evidence that the Irish trí, the English three, and the Latin trēs are related to the Sanskrit trī (त्रि) and the Farsi se (سه.) The higher the numbers go the more complicated they get, of course. An older pattern in Irish which I still use is to count higher numbers as four and fifty or ceathair is caoga, which is similar to the pattern used in modern German, and something that appears far more King James Bible in English. I love the complexity of the French base-twenty counting system, where the year of my birth, 1992, is mille neuf cent quatre-vignts douze, or one thousand nine-hundred four-twenties and twelve. Will the Belgian and Swiss word nonante to refer to the same number as quatre-vignts-dix ultimately win out in the Francophonie? Peut-être.
I was surprised to read in the New York Times last Friday that the latest version of ChatGPT called OpenAI o1 was built specifically to fix prior bugs that kept the program from solving mathematical problems. Surely this would be the first sort of language that one would teach a computer. As it turns out, no. Even now, OpenAI o1’s mathematical capabilities are limited to questions posed to it in English. So, as long as you have learned the English dialect of the language of mathematics then you can use this computer program to help you solve questions in the most universal of languages.
It reminds me of the bafflement I felt upon first seeing TurnItIn’s grammar correction feature, the purple boxes on TurnItIn’s web interface. For the uninitiated, TurnItIn is the essay grading and plagiarism detection system that most academic institutions that I’ve studied and taught at in the last 15 years use as a submission portal. I was proud to program into my Binghamton TurnItIn account several hotkeys that would allow me to save time retyping the same comment on 50 student essays every time they had a deadline. Thousands of essays later I can squarely say these hotkeys saved my bacon time and time again. Like legal documents, especially the medieval and early modern kind that I’ve read and written about in my studies, they are formulaic and expectable in their character.
The same goes for math: even with the basic understanding that I have (I only made it as far as Algebra II) the logic when explained well is inherent in the subject. Earlier in my doctoral studies, beginning in 2020, my two-sided approach to developing my own character and intellect beyond my studies came in the form of first signing up for Irish classes again, and second picking up where I left off with my mathematical studies in college and trying my hand at a beginner physics course. I’m sad to say I really haven’t had the time to devote to this mathematical pursuit as much as I would like. Perhaps I will be able to work it in someday, alas I also have to eat and sleep, and I’ve learned my attention will only last for so long. I too, dear reader, am only human.
Yet this is something where Open AI o1 differs from the average bear, for it is decidedly not human. How would we try to successfully communicate with a non-human entity or being when we have no basis for conversation to start with? The good thing about o1 and other AI programs is these are non-human minds which we are creating in our own image, ever the aspirant we are wrestling with the greater Essence from beyond this tangible Cosmos we inhabit. We can form o1 and its kind in the best image of our aspirations, a computerized mind that can recognize both empathy and logic and reflect those back to us in its answers to our questions. In the long run, I see o1’s descendants as the minds of far more powerful computers that will help our descendants explore this solar system and perhaps even beyond.
From the first time I saw it in work, I saw in ChatGPT a descendant of the fictional computers of Starfleet’s vessels whose purpose in being is to seek out new life and new civilizations and to boldly go where no one has gone before. Perhaps that future where humanity has built our utopia in this place, our planetary home, will be facilitated by AI. Perhaps, if we use it, build it, and train it right.
That said, the YouTuber Eddie Burback made a video several weeks ago about how he has seen AI put to use in his daily life in Los Angeles. In it, from the food delivery robots to his trips in several self-driving Waymo cars (manufactured by Jaguar), to his viewing of several AI films, Burback concluded that AI at this moment in 2024 is a net negative on human creativity and could remove more of the human element from the arts. I have seen far more AI generated images appear on my Instagram and Pinterest in the last year. I like Eddie’s videos, they may be long, but they are thorough and full of emotion, heart, and wit. They do a great service to their viewer at taking a long look at the world as he perceives it. I see much of the same thing, yet as the good Irish Catholic Cub fan that I am, I hold out hope that what today seems impossible to some: AI used morally and for the future improvement of our species and our advancement out of this adolescence in our story may still happen. I believe this is possible because I believe in us, that once this Wild West phase of the new Information Age settles down, we will see better uses of our new technologies develop, even as they continue to advance faster, higher, and stronger with each passing day.
Wednesday Blog by Seán Thomas Kane 