Thinking Fast, Slow, and Artificial: How AI Is Reshaping Human Reasoning

Original Article ↗ Hacker News Discussion ↗

Role of “System 3” / AI in Reasoning

  • AI is framed as a de facto “System 3”: an external reasoning layer that people increasingly defer to.
  • Commenters note AI introduces its own “cognitive biases,” shaped by training data, marketing, and culture.
  • Some argue this isn’t fundamentally different from asking another human; others say AI is uniquely opaque and confidently wrong, making errors easier to miss.
  • One view: AI doesn’t add a new system, it just moves existing cognition into “autopilot” and hides the struggle.

Impact on Individual Cognition

  • Several users report feeling cognitively stronger: more high‑level thinking, better problem solving, and “rubber‑ducking” into their own insights through dialogue with models.
  • Others see AI as an “amplifier”: it boosts those who already think deeply, while many users get lazier or struggle to use it effectively.
  • A strong opposing view: delegating core thinking to LLMs inevitably weakens critical thinking, likened to offshoring manufacturing and later realizing the skills are gone.

Tools, Atrophy, and Historical Analogies

  • Comparisons to calculators, GPS, cars, and databases:
    • Pro‑AI side: tools free us from low‑level work and historically leave us better off.
    • Cautious side: overuse leads to atrophy (mental arithmetic, navigation, physical fitness); AI should be used with similar discipline as diet or exercise.
  • Some intentionally avoid calculators/GPS to preserve “mental muscle.”

Reliability, Use Cases, and Verification

  • Multiple comments stress that LLMs must be treated unlike calculators or CPUs: they are probabilistic and wrong far more often.
  • Suggested safe uses: tasks that are easy to verify, tasks you already understand, one-step-beyond-your-skill learning, or low‑stakes outputs.
  • Subtle but plausible errors (e.g., color spaces) show how easily non‑experts can be misled.
  • Users describe strategies like multi‑model cross‑checks and carefully phrased prompts, but admit this is frustrating.

Social and Epistemic Effects

  • Concern that AI-written text makes “everyone sound like an expert,” eroding cues for real depth and expertise.
  • Worry that attention-optimizing LLMs resemble addictive feeds, blurring usefulness and manipulation.
  • Speculative fears: AI could accelerate a long‑term “dumbing down,” contribute to an “Idiocracy” scenario, or even be part of Fermi‑paradox style collapse.
  • Others counter that AI is already matching or exceeding humans in some domains (e.g., coding, math), and denial is identity-protective.

Foundations and the Paper Itself

  • Some note that classic System 1/System 2 work has replication and theoretical critiques, which may weaken the paper’s conceptual basis.
  • The study’s specific finding highlighted: AI improves performance when it’s right but reliably degrades it when it’s wrong, even under time pressure and incentives.
  • Several suspect parts of the paper were written with AI; opinions vary on whether that undermines its trustworthiness or is simply the new normal.