Classic self-reference in planning & control

Self-reference in "doing work" = a system reasoning about, modeling, or modifying itself. The classic traditions:

① Meta-reasoning / bounded rationality — reasoning about your own reasoning

Russell & Wefald, Do the Right Thing (1991) — deliberation as an action; rational metareasoning ("how much to think before acting"). I.J. Good's Type II rationality; anytime algorithms + meta-level control (Hansen–Zilberstein).

② Cybernetic self-reference — the controller must model itself/the world

• Good Regulator Theorem (Conant & Ashby, 1970): "every good regulator of a system must be a model of that system" (good-regulator-theorem). Internal Model Principle (Francis–Wonham, 1976).
• Second-order cybernetics (von Foerster), autopoiesis (Maturana–Varela), Viable System Model (Stafford Beer) — observer inside the system; system produces itself; recursive self-similar management.

③ Reflection / self-modification — programs that read & rewrite themselves

• Brian Cantwell Smith's 3-Lisp / procedural reflection (1982), Maes' computational reflection, the metaobject protocol.
• Schmidhuber's Gödel machine — self-rewrites its code only when it can prove the rewrite is beneficial; plus recursive self-improvement.

④ The mathematical core — fixed points

• Bellman equation / DP: the value function is self-referentially defined (V(s)=max_a[r+γV(s′)]); value iteration converges because the Bellman operator is a γ-contraction. Optimal control / RL is built on a self-referential fixed point.
• Kleene recursion theorem / diagonalization — the formal core (a program referencing its own code).

⑤ Meta-learning — learning your own learning

Meta-learning / learning-to-learn / meta-RL (RL²): improving your own learning and exploration.