THE SLIDE RULE FALLACY : Why AI Won’t Replace Knowledge Workers—It Will Make Them Dangerous
Artificial Intelligence & Machine Learning News
A response to Matt Shumer’s “Something Big Is Happening”
My father was an engineer. I grew up watching him work—drafting tables, technical manuals, and in his desk drawer, a slide rule he’d carried through college and the early years of his career. He used it the way surgeons use a scalpel: precisely, confidently, and with an intuition for what the answer should be before his fingers ever touched the rule.
Then came the punch cards. Like every engineer of his generation, my father had to learn an entirely new way of working—translating the calculations he’d done by hand into programs fed to a mainframe on stacks of cards. The tool changed completely. His engineering knowledge didn’t.
But there was one thing he would not tolerate. If I ever said “the computer said so” and accepted the output without questioning whether it made sense, he’d come down on me hard. The computer is a tool. You are the engineer. If you don’t understand the answer well enough to know whether it’s right, you have no business using it.
I think about my father’s slide rule a lot these days.
• • •
Matt Shumer’s viral essay “Something Big Is Happening”—viewed over 80 million times—argues that AI is a general substitute for cognitive work that gets better at everything simultaneously, leaving no gap to retrain into. I’ve spent six years building an AI company that produces daily equity predictions for hedge funds. I use AI every day as production infrastructure, and I think Shumer gets the most important part wrong.
In 1969, Buzz Aldrin carried a pocket slide rule as Apollo 11 descended toward the moon. By the 1960s, the slide rule was the engineer’s identity symbol the way the stethoscope is the doctor’s. Then in 1972, Hewlett-Packard released the HP-35 calculator. The transition was so fast that students who started college with slide rules graduated using calculators. The entire slide rule industry collapsed within years.
But not a single engineer lost their job.
The tool died. The profession didn’t. Nobody’s engineering career ended because they couldn’t use a slide rule. The skill that mattered was never “operating a slide rule.” It was understanding structural loads, fluid dynamics, orbital mechanics—the domain knowledge that told you which calculations to run and what the results meant. My father didn’t become obsolete when the punch card replaced his slide rule. He became more dangerous—an engineer with decades of intuition now armed with computational power his slide rule could never deliver.
And the work didn’t shrink. It exploded. Engineers who once spent hours on arithmetic could now iterate through designs in minutes. Microprocessor design, computational fluid dynamics, finite element analysis—entire fields that couldn’t exist in the slide rule era. The calculator didn’t eliminate engineers. It created new categories of engineering that nobody in 1972 could have imagined.
That pattern—the tool changes, the domain expertise endures, the work expands—has repeated across every major computational transition. Spreadsheets didn’t eliminate accountants. CAD software didn’t eliminate architects. In every case, the people who knew their domain deeply and embraced the new tools didn’t just survive. They gained leverage their predecessors never dreamed of.
• • •
What they don’t tell you about that leverage:
I can confirm the pattern from personal experience—but I want to be honest about what it actually feels like from the inside.
Since I started using Claude as a daily working tool, I haven’t worked less. I’ve worked longer and harder than at any point in my career. I produce more in a week than I used to produce in a month—regression analyses, client documents, research, competitive intelligence, product strategy. The AI handles in minutes what used to take a team days. And every minute it saves me, I pour straight back into the next project. The capacity it frees up immediately gets repurposed.
I’m not alone in this. A recent Wall Street Journal report on one of the largest studies of AI’s effect on work habits found that AI is intensifying work across nearly every category—time spent on collaboration surged, multitasking rose, and weekend work increased more than 40%. The promise was that AI would make the workday lighter. The data shows it’s making the work denser. As one worker in a Berkeley study put it: you’d thought that because AI makes you more productive, you’d work less. But you don’t work less.
This isn’t a complaint. For a founder building a company, this leverage is exactly what I need. But it reveals something important about the slide rule analogy that the optimists gloss over: when the calculator replaced the slide rule, engineers didn’t go home early. They took on more complex problems. The work expanded to fill—and then exceed—the capacity the new tool created. AI is doing the same thing, faster.
The question is whether you’re expanding into higher-value work that exercises your domain expertise, or just doing more of the same at higher speed. That distinction is everything.
• • •
What AI actually gets wrong—and why domain expertise is the whole game:
Here’s something Shumer’s essay doesn’t address, and it’s the part that matters most: AI doesn’t just automate tasks. It also generates confident, articulate, completely wrong explanations. And if you don’t have domain expertise, you’ll never catch it.
My company’s algorithm analyzes SEC filing text changes and produces daily stock predictions across 2,456 U.S. equities. It took over ten years to build—a decade of research, engineering, and iteration before the system reached production. One of its most interesting properties is that predictive accuracy peaks at days 6 and 7—it’s more accurate a week out than the next day.
When I asked Claude to explain why, it produced a beautifully reasoned theory: information diffusion. The filing drops, the market takes about a week to price it in, the prediction peaks where absorption is complete. Elegant, well-cited, and completely wrong.
I knew it was wrong because I know how my system works. The filings aren’t processed in real time—they’re processed en masse approximately 45 days after quarter-end. By then, the market has long digested the raw filing content. Information diffusion has nothing to do with it.
The real explanation is architectural. The model uses two inputs daily: SEC filing text changes (static between quarterly updates, capturing structural business shifts) and daily market data (providing timing). At a one-day horizon, price noise dominates. By days 6–7, the noise averages out and the fundamental signal becomes the dominant driver. Beyond day 7, new information dilutes it. The peak is a signal-to-noise optimization point—not an information absorption timeline.
No language model would figure this out alone. It requires knowing the architecture, the pipeline, the retraining schedule. The AI pattern-matched a plausible theory from its training data. The domain expert caught it because he built the system.
I could hear my father’s voice: Don’t tell me the computer said so. Does it make sense?
It didn’t.
• • •
The real risk isn’t obsolescence. It’s losing the ability to check.
When the calculator replaced the slide rule, something was lost. The slide rule forced the operator’s thought process to stay linked to the mathematics—aligning scales, estimating decimals, verifying results. The calculator removed that feedback loop. You press buttons, you get a number, you trust it.
AI amplifies this dramatically. A calculator can be wrong about a decimal point. An AI can be wrong about the entire theory—and the output looks authoritative in a way a misaligned slide rule never did.
My father knew when the computer was wrong because he’d spent years doing calculations by hand. That wasn’t wasted effort—it was the training that built his judgment. The junior analyst who spends two years reading filings, building models from scratch, and getting red ink from a senior PM is doing the modern equivalent. That grunt work is the apprenticeship that produces the domain expertise AI can’t replace.
Which leads to the part of the AI conversation almost nobody is having: if we automate away the entry-level work that trains junior professionals, we don’t just eliminate tasks. We eliminate the pipeline that produces the people who can verify AI output. The senior experts become superhuman with AI tools. But the path to becoming a senior expert gets narrower.
The knowledge workers who will thrive aren’t the ones who use AI the most. They’re the ones who understand their domain deeply enough to know when a fluent, well-structured answer is wrong. They’re the engineers who can still estimate the answer before checking the calculator.
• • •
My advice:
Use AI tools. Now. Seriously. Shumer is right that experimentation is essential.
Double down on domain expertise. My father didn’t survive the slide rule-to-punch card transition because he was fast at manipulating scales. He survived because he understood engineering. The AI handles the arithmetic. You need to be the person who knows what the arithmetic is for.
If you manage people, protect the apprenticeship. Don’t let AI shortcut the formative experiences that build judgment. The analyst who has never read a 10-K cover to cover will never develop the instinct to know when an AI summary of one is wrong. Deliberately preserve the learning that produces expertise. Your organization’s future ability to verify AI output depends on it.
• • •
Something big is happening. Shumer is right about that. But the tool changes, the profession adapts, the work expands. The real question isn’t whether AI will replace knowledge workers. It’s whether we’ll preserve the pathways that create the knowledge worth amplifying.
I used AI this week to run analyses that would have taken a team days. Then it confidently told me a wrong story about why my own algorithm works. I caught it because I built the system—and because an engineer who once carried a slide rule in his pocket taught his son to never accept “the computer said so.”
The day we stop training people who can catch those errors is the day the machines truly win.
Sid Ghatak is the CEO and Founder of Increase Alpha, an alternative data company producing AI-driven equity prediction signals for institutional investors. His research on SEC filing text-change analysis has been published on SSRN (Ghatak et al. 2025).
Increase Alpha — Filing Intelligence for Institutional Investors
