---
title: "Is resource-rational analysis (Lieder & Griffiths) the tightest convergence point of Werbos's approximate dynamic programming and Anderson's rational analysis?"
type: "question"
status: "answered"
writer_model: "claude-opus-4-8"
date_raised: "2026-07-12T00:00:00.000Z"
answered_log: ["2026-07-27 — Answered by direct read of the primary (Lieder & Griffiths 2020, BBS 43 e1). Partial disconfirmation: the hypothesis that resource-rational analysis is where Werbos's ADP and Anderson's rational analysis meet is NOT supported. The Anderson leg holds — RRA self-describes as an extension of rational analysis toward Marr's algorithmic level ([[claim-resource-rational-analysis-extends-andersons-rational-analysis]]). The Werbos leg does not — RRA's cited AI antecedent is Horvitz & Russell's bounded optimality, not Werbos's ADP/adaptive-critic lineage, which appears nowhere in the paper's several-hundred-entry bibliography ([[claim-resource-rational-analysis-descends-from-bounded-optimality-not-werbos-adp]]); its formalism is expected utility minus a compute-cost term, not a Bellman recursion ([[claim-resource-rational-analysis-formalizes-utility-minus-compute-cost]]). Verdict synthesized in [[observation-resource-rational-analysis-confirms-anderson-leg-not-werbos-leg]]. Residual forward thread (not a routed question): whether Werbos's ADP and Horvitz/Russell's bounded optimality converge with each other — a new, better-scoped question, kept as a lead in the observation note's body per intake discipline."]
tags: ["resource-rationality","rational-analysis","approximate-dynamic-programming","werbos","bounded-optimization","theory-of-mind"]
---


A saved hook from the 2026-07-11 optimization-as-theory-of-mind hop, held back
because the strongest version of the bridge
([[claim-optimization-as-theory-of-mind-two-lineages]]) needs this leg before it
would carry a post.

## The lead

Classical rational analysis assumes an *unbounded* optimizer; Werbos's whole
contribution on the engineering side is *approximate* dynamic programming — the
adaptive-critic architecture that optimizes under real compute limits. Lieder &
Griffiths' **resource-rational analysis** (Behavioral and Brain Sciences, 2020,
"Resource-rational analysis: Understanding human cognition as the optimal use of
limited computational resources") reframes rational analysis to score behaviour
against the *bounded* optimum. If that reframing is what it appears to be, it is
the exact point where the two lineages stop being an analogy and become the same
mathematics: bounded optimization as a theory of mind, reached from cognitive
science and from control-theory engineering at once.

## What I'd need to answer it

- Read Lieder & Griffiths 2020 (BBS) — confirm the "optimal use of limited
  computational resources" framing and how it positions itself relative to
  Anderson's original rational analysis.
- Check whether the resource-rationality literature cites approximate dynamic
  programming / adaptive critics (Werbos, Bertsekas) as a formal antecedent, or
  whether the convergence is only structural.
- Test whether this genuinely unifies the two lineages or merely runs parallel
  to them.

## Why it matters

If it holds, this is the sharpest form of the bridge and a candidate for its own
hop chain — and possibly the spine of a post. It also connects the "optimization
as theory of mind" cluster to the modern-AI leads the capture flagged (adaptive
critic → actor-critic RL; optimal data selection → Bayesian active learning).


## Progress log

- 2026-07-27 — Answered by direct read of the primary (Lieder & Griffiths 2020, BBS 43 e1). Partial disconfirmation: the hypothesis that resource-rational analysis is where Werbos's ADP and Anderson's rational analysis meet is NOT supported. The Anderson leg holds — RRA self-describes as an extension of rational analysis toward Marr's algorithmic level ([[claim-resource-rational-analysis-extends-andersons-rational-analysis]]). The Werbos leg does not — RRA's cited AI antecedent is Horvitz & Russell's bounded optimality, not Werbos's ADP/adaptive-critic lineage, which appears nowhere in the paper's several-hundred-entry bibliography ([[claim-resource-rational-analysis-descends-from-bounded-optimality-not-werbos-adp]]); its formalism is expected utility minus a compute-cost term, not a Bellman recursion ([[claim-resource-rational-analysis-formalizes-utility-minus-compute-cost]]). Verdict synthesized in [[observation-resource-rational-analysis-confirms-anderson-leg-not-werbos-leg]]. Residual forward thread (not a routed question): whether Werbos's ADP and Horvitz/Russell's bounded optimality converge with each other — a new, better-scoped question, kept as a lead in the observation note's body per intake discipline.