To be fair to those who use it:
the calculator precedent offers one genuinely useful lesson. When calculators
arrived in classrooms in the 1970s, 72 percent of math teachers opposed giving
them to seventh graders. They feared that students would never learn to think
mathematically if a machine could do the arithmetic for them.2 Those fears did not materialize. Math education adapted, and students
ultimately engaged with more sophisticated problems than the pre-calculator
curriculum allowed.
The lesson worth keeping from
that history is that schools can adapt to new tools without abandoning the
underlying goals of education. That much applies. But the lesson stops there,
and treating it as a full blueprint for AI is where the trouble starts.
First, calculators produce
correct answers. Punch 6 times 9 into a calculator and you get 54. Every time.
Generative AI produces plausible-sounding answers that are sometimes wrong,
without labeling which is which. As historian Stephen Jackson has put it, the
more accurate image would be a calculator that occasionally returns 2+2=5 with
the same confidence it returns 2+2=4.3 A student who does not already know enough to evaluate the output cannot tell
the difference. That is a fundamentally different risk profile than the one
calculators introduced.
Second, a calculator outsources
arithmetic. Generative AI can outsource reasoning. This is the distinction that
Stanford political scientist Rob Reich raised at Stanford's AI Education
Summit: writing is how students learn to think.4 A calculator frees a student from the burden of long division so they can
engage with more complex mathematical ideas. An AI that writes a student's
essay does not free them to engage with more complex ideas. It removes the
process by which the thinking was supposed to happen. Those are not the same trade-off
and treating them as equivalent causes schools to underestimate what is at
stake in the decisions they are making about academic integrity.
Third, a calculator is a
bounded tool. It does arithmetic. A generative AI system can write code, draft
legal arguments, produce images, hold extended conversations, simulate
emotional relationships, and generate content that has nothing to do with
learning. Researchers at the University of Western Australia have described
this as the difference between a domain-specific tool and what they call an “everything
machine.”5 The policies and guardrails appropriate for a calculator do not translate to a
technology that students are already using for mental health support, creative
writing, and social interaction, as well as homework.
We are not suggesting schools treat AI as uniquely dangerous or unprecedented in every respect. The calculator precedent is a reasonable reminder that institutional panic about new tools tends to be misplaced. What we are pushing back on is using the analogy as a reason to skip the harder thinking.
The questions the calculator analogy tends to foreclose are the ones that most need answering. What specific tasks are we comfortable with students offloading to AI, and at what stage of their learning? How do we assess understanding when the tool can produce a finished product without the student having engaged with the underlying concepts? What does it mean to use AI responsibly when the tool can behave differently than expected?
Calculators did not require schools to answer questions like those, because calculators do not write essays, hold conversations, or simulate relationships. AI does. The school districts handling this well are not the ones that found the most reassuring analogy. They are the ones that treated AI as something worth understanding on its own terms, which it is.
