TB: Academics Need to Wake Up on AI — Parts I & II (Kustov)

Core Thesis

Alexander Kustov argues that academia's reigning posture toward AI — debate the philosophy of "understanding," dismiss with anecdote, refuse to update workflow — is intellectually unserious, professionally suicidal, and uncalibrated to what the tools actually do. Resistance is futile; the only live question is how to redesign research to take advantage of the inevitable.

Key Takeaways (Part I)

Academia applies double standards

  • AI is held to a perfection bar that human-authored papers are not. P-hacking, reproducibility failures, citation cartels — all tolerated. A model that hallucinates once is dismissed as "fundamentally broken."
  • Confusing philosophical disputes about machine "understanding" with practical disputes about capability is the single biggest intellectual error in the discourse.

The Bluesky echo chamber

  • Pile-on culture on Bluesky and similar platforms has created a denialist consensus that does not survive private conversation. Junior scholars learn quickly to keep heterodox views off-line.
  • "AI is useful and fun … Instead of doomscrolling, I now slack off by trying side projects in Claude Code. May be the most productive form of non-work there is."

The redesign prescription

  • Spend concentrated time with agentic AI (not chatbot AI). Capability has moved.
  • Stop debating "understanding"; pick problems where AI's verifiable contribution is now — data verification, security, p-hacking detection, institutional design.
  • Pause to redesign workflows intentionally. Otherwise AI fills the gaps with the same incentive structures that produced the current dysfunctions.

Key Takeaways (Part II — ten further theses)

Jaggedness explains polarization

  • "AI can write a serviceable literature review but struggle with a basic visual puzzle." Each scholar's experience depends on which sub-task they happened to test. Conclusions vary because the capability is jagged, not because some users are more honest.

Publication lag is fatal

  • A peer-reviewed critique of GPT-N is obsolete by the time it appears, because GPT-(N+1) is out. Academic timelines were designed for slow-moving objects of study; AI is not one.

Disclosure backfires

  • Mandatory AI-use disclosure penalizes the honest. "The rational incentive is to lie." Disclosure rules without enforcement raise the dishonesty rate.

Qualitative research gains relative value

  • As AI compresses literature reviews and large-N analysis, the comparative advantage of fieldwork, ethnography, and qualitative interpretation rises.

Skill atrophy is real

  • Future scholars who never learn the underlying craft because AI did it for them are a real, not rhetorical, worry. Counter-arguments about "higher-level abstraction" do not dissolve the worry; they just relocate it.

Public funding justifies adoption

  • If AI lets the same dollar buy more research, the social compact requires using it.

Mental Models

  • Incentives Matter — disclosure mandates without enforcement select for dishonesty; tenure incentive structure selects against workflow change
  • Second-Order Thinking — first-order: AI does literature reviews. Second-order: that lifts qualitative methods up the value stack
  • Inversion — instead of asking "what is wrong with AI for academia?" ask "what is wrong with academia that AI exposes?"
  • It Pays to Get the Design Right — Kustov's prescription is workflow redesign, not chatbot adoption

See also

Source

Alexander Kustov, Popular by Design Substack, 2026

Backlinks