Hokku

A log of open questions. Notes to follow.

  1. 01

    How do we encode human judgment, evidence, context, standards, uncertainty and quality criteria so an agent can make better decisions?

    docs — soon
  2. 02

    When should an AI system make a fast judgment, when should it retrieve more evidence, when should it reason longer, and when should it refuse to decide?

    docs — soon
  3. 03

    What important human cognitive work is repeatedly performed but has not yet become a software primitive?

    docs — soon
  4. 04

    What important human cognitive work does software still fail to represent properly?

    docs — soon
  5. 05

    How do we represent a design system as a machine-readable set of tokens, semantics, invariants, component contracts, patterns, and flows that both humans and AI can reliably build against?

    docs — soon
  6. 06

    How can an AI agent query, compose, and modify a design system without violating the system's semantic and behavioral invariants?

    docs — soon
  7. 07

    How do we make AI outputs trustworthy without making people blindly trust them?

    docs — soon
  8. 08

    How do we represent evidence, uncertainty, assumptions, and competing explanations in software?

    docs — soon
  9. 09

    How do we help people distinguish plausible answers from things actually supported by evidence?

    docs — soon
  10. 10

    How should humans and AI divide cognitive work when the AI is good at producing answers but bad at knowing when it is wrong?

    docs — soon
  11. 11

    How do we design interfaces for uncertainty, disagreement, ambiguity, and incomplete information?

    docs — soon
  12. 12

    How can AI support expert reasoning without quietly replacing it?

    docs — soon
  13. 13

    How do we preserve context across complex workflows without pretending that a knowledge base is a substitute for understanding?

    docs — soon
  14. 14

    How do organizations make better decisions when incentives reward speed, local optimization, and confidence rather than learning?

    docs — soon
  15. 15

    How can software make reasoning processes observable, revisable, and auditable?

    docs — soon
  16. 16

    What happens when an abstraction becomes more influential than the reality it was meant to represent?

    docs — soon
  17. 17

    How do we know whether a research method actually produced knowledge, rather than merely activity or a persuasive narrative?

    docs — soon
  18. 18

    How can design systems encode consequences, constraints, and relationships rather than just visual consistency?

    docs — soon
  19. 19

    What cognitive work remains poorly supported after documentation, search, and routine analysis become automated?

    docs — soon
  20. 20

    How do we build tools that help experts think better rather than simply helping them produce more output?

    docs — soon