Epoch ReadingField NotesOne-pager 030

The world is its own best model.Rodney A. Brooks · Intelligence without representation · Artificial Intelligence 47 · 1991

Read straight after Lucas as the easier one — and it turned into the anti-representation bookend of the whole arc. Brooks’s move: stop building intelligence top-down out of symbols and central reasoning, and build it bottom-up out of complete simple “Creatures” that act in the real world, with no central model and no central control. The margins ran it onto Dennett’s inner arena (“input, output — Dennett here”) and Lucas’s Gödel (“opposite of Gödel”), then landed the whole thing on purpose: real-world robotics starts and ends with the “why.”

read 08.21.26pp. 139–15908 · 21 · 2026
◆

Where the margins reached

Dennett · inner arena Lucas · Gödel the spar · mind-as-machine Simon’s Ant 2026 · the labor tier the self-archive BROOKS 1991
Book → Book Book → World Book → Self
▲

The lines he pulled

“It turns out to be better to use the world as its own model.”

§ abstract · boxed — the thesis in one sentence

“Representation is the wrong unit of abstraction in building the bulkiest parts of intelligent systems.”

§1 · marked H — the load-bearing claim, tagged

“Out of the local chaos of their interactions there emerges, in the eye of an observer, a coherent pattern of behavior. There is no central purposeful locus of control.”

§5 · margin: “the president giving an eloquent commencement speech the whole time he is thinking about trying not to poop his pants”

“Much of even human level activity is… a reflection of the world through very simple mechanisms without detailed representations.”

§6 · Simon’s Ant · margin: “where a person is from says a lot about them”
●

The read, section by section

§1–2
  1. ■Boxed the thesis outright: use the world as its own model, and “representation is the wrong unit of abstraction.” — he marked the load-bearing sentences directly, not the scaffolding around them.
  2. AAt the being-and-reacting-took-billions-of-years passage: the hard part is mobility and perception, not abstract thought. — evolution’s ledger as the current race.
§3
  1. AContextual ROI placement — what is worth representing is a value call, potential over the long term. — reading the represent-versus-use-the-world choice through a build-and-value lens.
  2. DThe essence of intelligence is being solved by the researchers, leaving little for the AI programs to do. — caught Brooks’s core charge: abstraction hides the genuinely hard part.
§4.1–4.2
  1. ASymbols — or the world itself triggers the register. Build one very simple complete autonomous system and test it in the real world. — the fork stated in his own words, then the build-a-Creature maxim.
  2. CKinda like us growing from crawl to walk — minus learning ABCs or potty training, concurrently. — layered Creatures as human development, with the sharp catch that the layers run at once and without a taught curriculum.
  3. DSuch as your lungs breathing autonomously while your brain wants to go swimming. — decentralized control: the body runs itself while “you” want something else.
§5
  1. CThe layers are all unaware of each other’s tasks under the decomposition. — at the separate-hardware, parallel-goals passage.
  2. DThe president giving an eloquent commencement speech the whole time he is thinking about trying not to poop his pants. — coherent surface, chaotic competing behaviors underneath: no central purposeful locus.
  3. EInput, output — Dennett here. — written exactly where Brooks says coherent behavior emerges with no central author. The arc connection, made in ink.
§6
  1. AOpposite of Gödel. — on Brooks’s claim that the “representation” exists only through an external mapping: meaning is imputed by an observer, not by formal self-reference.
  2. AWhere a person is from says a lot about them. — Simon’s Ant: the environment carries the complexity, not the creature.
  3. BStuck on “tutorial mode” versus a high-pressure lobby. — the maxim: never build or test in a simplified world and then ship into the real one.
§7
  1. ANo rules except: react, and do not step on the other layer’s toes. — why subsumption is not a production-rule system.
  2. CWhere are the connectionist people at today? — at the “what this is not” pages: the question that points straight at modern deep learning.
§8
  1. CWhat would 2026’s analogy be — a non-high-school-grad, an Amazon warehouse worker? — updating Brooks’s insect-versus-bacteria intelligence tier to a modern labor tier.
  2. CThey built the layers and yet cannot keep up to fully trust them. — caught the author’s own concession: an isolated module, inputs and outputs left dangling.

Structure is scaffold; every quoted line is his, verbatim from the annotated pages. Letters are per-page comment order; section markers stand in for the journal pagination. A handful of margins — a Gödel spelling, one symbol, one starred word — are flagged low-confidence in the decode and held off this page rather than smoothed into false certainty.

The one-sentence summary · his §8.4 stamp
“The lab is every movie’s cookie-cutter out-take; real-world robotics starts and ends with the ‘why’ — purpose, i.e. philosophy.”
★ 1 sentence summary ★ · pp. 139–159 · Artificial Intelligence 47 · read 08.21.26