We Are the Last People Who Know How It Works

Original Article ↗ Hacker News Discussion ↗

As computing becomes easier and more abstracted—from DOS-era config files to today’s touch interfaces and AI assistants—many engineers worry that we’re losing the hard-won, low-level understanding that once came from “fighting” the machine. Commenters debate whether this is just another historical shift like automatic transmissions and household electricity, or something new and riskier because AI systems are opaque, non-deterministic, and increasingly delivered as centralized subscription services. Some see AI as an extraordinary learning tool; others fear a future where few humans can verify or repair the complex digital infrastructure society depends on.

Abstraction, Specialization, and “Knowing How It Works”

  • Many argue this isn’t new: every technology moves from expert-tinkerable to opaque, with users living at higher layers of abstraction (cars, electricity, telephony, PCs).
  • Others stress that some understanding of a few layers above/below your work remains essential, especially for debugging and infrastructure resilience.
  • There’s debate over how much understanding counts: rough conceptual grasp vs being able to rebuild a CPU or fabrication process from scratch.

What’s Different About AI and Modern Computing

  • Some see LLMs/agents as just another abstraction layer; others say they’re qualitatively different due to non-determinism and inability to reliably check their own outputs.
  • Concerns include: hallucinations, erosion of human expertise, “model collapse,” and being locked into subscription-based cognition.
  • Counterpoint: tools can also deepen understanding if used as interactive tutors; the risk comes from defaulting to “do it for me” instead of “help me learn.”

Loss of Acquaintance vs Loss of Knowledge

  • Distinction drawn between:
    • Hard technical knowledge (which is well-preserved in documents, code, and some experts).
    • “Acquaintance” or hands-on struggle (e.g., IRQs, autoexec.bat, jumpers, modem tones) that built intuition and confidence.
  • Some see this loss as mostly nostalgic and acceptable; others fear a dangerous gap if too few people can maintain foundational systems.

Younger Generations and Computer Literacy

  • Mixed observations:
    • Some report students who can’t troubleshoot basic OS installs or think beyond smartphone-style UX.
    • Others note vibrant modding scenes, DIY 8‑bit projects, and very capable young systems programmers.
  • General agreement that curiosity persists, but consumer devices and frictionless UX disincentivize tinkering.

Tinkering, Financialization, and Enshitification

  • Several link the decline of hobbyist exploration to financialization, growth-at-all-costs, and attention economies.
  • “Smart” products are criticized as vehicles for lock-in and ads, not genuine user empowerment.
  • Some expect a growing indie/handmade computing scene as a reaction, where doing things “the hard way” gains renewed cultural value.

Meta: AI-Generated Text and Detectors

  • Part of the thread debates whether the original essay “sounds like AI.”
  • Many distrust AI detectors, report high false positives on older human-written text, and worry that LLM style is bleeding into human prose.