AI is not like the calculator

Last year, when I was much more AI-positive, I often had conversations about AI with people both inside and outside the Computer Science field. Most of those conversations boiled down to two camps: the optimists and the pessimists. It was usually my colleagues, experienced developers, who didn't think LLMs would take their jobs. They saw it as a tool, with the implicit meaning that you can only use it well once you understand everything around it. In software, that means knowing how code and systems actually work before you ever reach for the AI.

The conversations with friends and family often turned much more sour. They had little to no programming experience, but they had seen the posts about how AI was coming for my "future job market". My parents especially worried about it. How could they happily watch their son study something AI would supposedly replace in a few years anyway? To calm them, and my friends, I would always tell them the story of how the calculator entered schools.

The calculator might be the most controversial tool ever to enter the classroom (see Watters 2015). Scholars and experts argued that leaning on the machine would rot young minds. The fear was that students would never learn to estimate, and that they would never learn from their mistakes, because the machine made sure they never made any.

Of course, we all know that isn't how the calculator story ended. Far from rotting young minds, the calculator turned out not to harm students' math skills at all; the students who used one developed better problem-solving skills and a more positive attitude toward math than those who didn't (see Ellington 2003). The machine didn't lower the ceiling, it raised the floor, freeing students to spend their effort on reasoning instead of arithmetic. And the floor it raised was real: take the calculator away and the skill underneath was still there. This was the success story I preached to worried souls. No, I told them, LLMs won't "take my job". They will become like the calculator: a tool that raises the floor and lifts the ceiling of what is possible.

I lived by this mantra, genuinely believing the future of programming was bright with LLMs at our side. Then, a few months ago, I took a hard look at how I actually used LLMs. By that point I had about 1.5 years of experience with them, both for general use and for coding. I had used them to teach me things I struggled to understand and to build whole applications. It had been a wild ride, so I tried stepping off the train and building something on my own. And for a while, I could. Then I hit my first bug, and I immediately reached for the LLM. The bot solved it instantly. If you can describe the problem clearly and keep the scope tight, the LLM delivers. I was impressed. I had a problem I was stuck on, and a tool solved it for me. Then another bug, another instant fix. Then the UI didn't look right, and the chatbot fixed that too. The feeling reminded me of my earliest school years, when another tool was always there for me whenever I had a problem.

In 2009, Alex Shevchenko, Max Lytvyn, and Dmytro Lider gave the world Grammarly, a tool that corrects your spelling and grammar. As a Dane, my second language was of course English. We started learning it in 3rd grade, around nine years of age. I loved writing in English. I was a kid in love with the internet, raised on English novels and the English-speaking web, and English became a language I almost loved more than Danish. I was good at it too: the teachers loved my writing and my speaking, well, as much as you can love the work of a nine-to-twelve-year-old. Then in 7th grade, at thirteen, we were finally introduced to English grammar. I had already struggled with Danish grammar for reasons I never understood. I loved writing, but the grammar always tripped me up. Learning English grammar on top of that did not help, and I was desperate for a way out. That year I went from a top student to well below average, simply because I couldn't keep up with grammar. That is when I found Grammarly. It was like a drug.

I loved writing and I hated grammar, and Grammarly erased the part I hated while boosting the part I loved. It felt like what the calculator did for me in math: it handled the tedious groundwork and freed me to enjoy the rest. Then in 9th grade we sat for our final exams of Danish primary school, and of course Grammarly wasn't allowed. And of course I got a terrible grade in English. This dependence on grammar tools followed me through high school. I never got better, my Grammarly use only grew, and I never learned English grammar in high school either.

In math, though, I kept excelling. Something about the way the calculator was simply expected meant I could keep getting better. Language was the opposite. There you were expected to know the grammar by heart, it was treated as the basic part of every course, and you couldn't move on to anything more advanced until you had the fundamentals down. I don't think math and language are really that far apart, least of all the grammar part, which is arguably the strictest, most rule-bound part of any language. And I don't think it comes down to how my brain is wired either. To this day I'll always care more about writing, and about finding creative ways to look at the world and the art we inherit. No, I think the difference between my grades in math and language came down to expectations about our tools. Math expected you to have a tool like the calculator, and it used that expectation to raise the ceiling of what you could do. Language expected no tools at all. The school system enforced grammar and basic writing first, maybe at the cost of student creativity or some genuinely great writers who were made to grind on the basics instead. Whether that trade-off is worth it isn't the question this essay is trying to answer. What I'm much more interested in is what all this means for computer science, or more specifically, programming.

Programming has always been an abstract, creative field (see Autor 2015), and we as developers have always tried to minimize the grunt work and raise the ceiling of what is possible. So what would be wrong with AI, then? Why should we fear it as a Grammarly when we could respect and love it like a calculator?

Well, I don't know. The only thing I can offer is my experience: To this day I still use Grammarly for everything: every email, every report, every document. When I started, it felt like the calculator, something that let me excel and focus on what I do best. Now I know what it really did. Grammarly raised my floor, but the floor was never mine; it kept me from ever learning a core part of a language. Take the tool away and I fall straight through it. I can try to learn the grammar now, but I fear it's too late. The tool is too woven into my professional life. People have expectations about my performance and my skills, but those aren't really my skills. They're Grammarly's. I worry that AI might turn out to be a Grammarly for me rather than a calculator. Whether or not that is a good thing I don't know, but I think it's important we start to analyze the tools we set as the basis for student learning. A calculator is allowed; Grammarly is not. Where in this does AI land? Should we elevate students and clear away their grunt work, the way math did with the calculator? Or should we keep them on the floor, making sure every single nook and cranny is covered before they can raise their head and look at the ceiling?

Is AI a tool, or a crutch?