Ted Kaczynski's 2000 warning about AI and math careers holds up

In 2000, Ted Kaczynski wrote a letter arguing that students should steer clear of specializing in pure mathematics or theoretical computer science—not for the usual reasons people brush off his views, but because he foresaw that within a couple of decades, AI systems would handle the abstract reasoning these fields depend on. He wasn't referring to narrow automation; he meant the general problem-solving that mathematicians devote years to mastering. Back then, this sounded like paranoia from a man who'd lived in isolation in a cabin for years. Fifteen years on, with transformer architectures and large language models starting to crack graduate-level reasoning tasks, it no longer seems so far-fetched.

I came across a copy of that 1999–2000 correspondence while sifting through archives of academic mailing lists. What caught my attention wasn't the anti-technology angle—that's easy to dismiss—but the structural insight: if you're preparing for a career centered on manipulating symbols by formal rules, and you think intelligence is substrate-independent, then you're essentially betting against your own redundancy. He posed it as a question of obsolescence: why invest decades in honing a skill a machine might outperform, faster and cheaper, by the time you finish your PhD?

That's not philosophical musing. That's workforce economics.

Seeing Claude 3 and GPT-4 tackle Olympiad-level math problems or produce original proofs in Lean makes this feel less like prophecy and more like a cautionary tale about narrow specialization. The counterpoint is that human creativity, intuition, and conceptual imagination remain indispensable—but the margin is shrinking. If you're an undergraduate today weighing pure math against applied statistics or ML engineering, Kaczynski's bleak logic deserves at least one serious look. He wasn't right for the reasons he believed, but the timeline he worried about? That's arriving on schedule.

The uncomfortable lesson: skills that took humanity centuries to build are being squeezed into a single training run. Whether you're crafting the next prompt engineering workflow or debugging an LLM agent in production, the real question isn't whether machines can do abstract reasoning—it's how fast they'll commoditize whatever niche you've carved out for yourself.