Concept Map

The concept files are raw material. The article should not present them as ten independent ideas. They form a smaller argument with four layers.

Layer 1: Capability Mechanism

Main concept:

  • ../concepts/cheap_feedback.md

Supporting concepts:

  • ../concepts/feedback_cost_map.md
  • ../concepts/plateau_agi_partition.md

Claim:

AI progress is fastest when the system can generate many candidates and receive cheap, reliable feedback. This includes proof checkers, unit tests, formal systems, benchmarks, and fast simulations.

Layer 2: Mathematical Bottlenecks

Main concepts:

  • ../concepts/importance_oracle.md
  • ../concepts/synthesis_debt.md

Supporting concept:

  • ../concepts/conjecture_economy.md

Claim:

Correctness is only one bottleneck. Mathematics also needs question selection, proof compression, explanation, reuse, and teaching. AI can close more questions while increasing the amount of material that still needs to be digested.

Layer 3: Human Formation And Community

Main concepts:

  • ../concepts/acquisition_asymmetry.md
  • ../concepts/community_layer.md

Claim:

Tools that help experts can weaken the training path for novices. If fewer people learn a domain deeply, the documents may remain while the working knowledge around them weakens.

Layer 4: Institutions And Labor

Main concepts:

  • ../concepts/measurability_distortion.md
  • ../concepts/conjecture_economy.md
  • ../concepts/amplification_to_replacement.md

Supporting concept:

  • ../concepts/plateau_agi_partition.md

Claim:

Institutions reward what they can observe. AI makes some tasks easier to score and display, which can redirect prestige, hiring, funding, and training toward the AI-tractable part of the frontier. At the labor level, “assistant” workflows are unstable when the human remains only a review bottleneck.

Recommended Article Order

The article should follow the layers above, not the chronological order of ../ai_views.md.

  1. Start with cheap feedback.
  2. Use math proof as the clean example.
  3. Separate proof generation, verification, and digestion.
  4. Introduce the importance oracle.
  5. Move to synthesis debt and community.
  6. Discuss education and acquisition asymmetry.
  7. Discuss metrics, conjectures, and institutions.
  8. End with plateau/AGI and labor as consequences of the feedback-cost split.

Concepts To Merge In The Final Draft

  • cheap_feedback.md and feedback_cost_map.md should probably become one early section plus one later generalization.
  • importance_oracle.md and conjecture_economy.md overlap around question selection.
  • synthesis_debt.md and community_layer.md should be adjacent.
  • plateau_agi_partition.md should not become a long standalone essay. It is a consequence of the feedback-cost map.