AI Is Not A Calculator

One of the popular pro-AI arguments these days is that adopting AI in classrooms is just like back in the day when calculators wormed their way into the classroom.

"Even with calculators, students still have to learn fundamentals like times tables," the argument goes. "But calculators simplified things, got rid of penmanship-related errors, and ultimately just helped students do their thing more quickly and efficiently." So AI will just get folded into the educational process, a sort of digital helper.

Well, no. How is AI not at all like a calculator? Let me count the ways.

Calculators are consistent and reliable. Punch in 3 x 12 and it will spit out 36. If you ask it to multiple 3 and 12 a thousand times, the calculator will spit out 36 a thousand times. But if I ask ChatGPT to write a response to the same prompt a thousand times, it will give me a thousand different answers. Here are just a couple of what I got by asking it to write a single sentence comparing Hamlet and Huckelberry Hound:

While Hamlet broods over existential dilemmas and the weight of revenge, Huckleberry Hound ambles through life with laid-back charm and a carefree tune.

Hamlet is a tormented prince consumed by introspection and tragedy, while Huckleberry Hound is a cheerful, easygoing cartoon dog who breezes through life’s mishaps with a song.

Broadly similar, but with significant differences. Structurally, each sentence uses a subordinate clause to center a different character as the main focus of the sentence. ChatGPT also gives us two Hamlets. One more passive (he's "tormented" and "consumed") and the other is active (even he's brooding). One worries about revenge and existential angst, while the other is introspective and --well, somehow consumed by tragedy, which leaves it unclear whether he's somehow part of the tragedy or just pre-occupied with it.

You may think I'm being picky in ways that only an English teacher could be, but word choice matters and these sentences are not just two ways of saying the same thing, but are two different statements. They represent two different thoughts--well, would represent two different thoughts if a thinking being had generated them.

AI is not reliable. You get a different answer every time.

Calculators are also accurate. 3 x 12 is, in fact, 36. AI presents incorrect information, often. Let's not call these errors "hallucination," because the word anthropomorphizes the algorithm has human-ish perceptions that have somehow been tricked. It produces incorrect information through exactly the same process that it produces accurate information; if you want to say its errors are hallucinations, you should acknowledge that it hallucinates 100% of the time.

Calculators work out matters of fact, and the manufacturer's biases are not a factor. Even if the calculator was manufactured by folks who believe that off numbers are way cooler than even ones, the calculator will still compute that 3 x 12 equals 36.

But AI deals with many matters that are not factual at all. Chatbots have repeatedly demonstrated a tendency to veer off into wildly racist or misogynist output--and that's just the obvious stuff. AI can be programmed to present any bias its operators care to feed into it. And yet we are encouraged to think of chatbots as objective arbiters of Truth even when there is every reason to believe they are stuffed full of human biases.

A calculator saves you the trouble of performing operations--operations that could be performed by any other calculator or any human being with the necessary operational knowledge. A chatbot saves you the trouble of thinking, of figuring things out.

A chatbot is not a calculator. There may be valid arguments for AI in the classroom, but this is not one of them.