Synthetic discussions generated from public artifacts. No users, scores, or comments are real.

Corpus frame

The corpus applies one lens to many domains: what mechanisms produce the outcome? It shares four methodological commitments and one explicit directional commitment. Each linked page argues for its part; the links are derivations and disputes, not evidence inherited by every page. The directional commitment does not by itself settle system boundary, distribution, sacrifice, or institutional authority.

  1. Mechanisms are what act. Incentive gradients, selection pressures, feedback loops, and capital stocks produce the distribution of outcomes. Intentions, labels, official categories, and stated values are evidence about mechanisms, or are themselves coordination mechanisms. They are not causal substitutes. — Mechanism Realism · Only Selection
  2. The reference telos is sustained flourishing. The broadest achievable adaptive safety margin over deep time — not the continuity of any incumbent state, coalition, institution, or doctrine. A mechanism's own stated goal can still serve as a local proof obligation — showing that its incentives defeat even the purpose it claims is a bounded finding — but meeting that goal establishes nothing about the margin. — Flourishing Is Maximum Safety Margin
  3. Law, rights, legitimacy, democracy, markets, and sovereignty are mechanisms under evaluation. They are constraints, carriers, or proxies inside the analysis. None is a terminal value or a boundary of what is real. Treating one as terminal ends the mechanism search before it starts. Evaluation carries current function, replacement cost, path dependence, uncertainty, capture risk, reversibility, and who bears model error into the ledger. — The Stack · Mechanism Space
  4. Optimization is a system function. A civilization has to build, exercise, and revise metamechanisms that search mechanism-space, discard dominated options, install, observe effects, and repair under uncertainty. Not running that loop leaves margin unrealized, and that is itself the failure. No single component — analyst, model, or institution — is presumed to contain a global optimum; the capacity is a property of the system. — Telic Systems · The Three-Layer Architecture
  5. Uncertainty is preserved, not spent. Partial orders, binding constraints, unknowns, and residuals stay explicit. An unmeasured effect is not a favorable default. — The Compression Paradox · Cargo Cult Epistemology

Each essay bears its own evidence. Links carry definitions, derivations, applications, and disputes; they do not transfer proof. Criticism is answered on its substance.

Where each commitment is derived

← Mechacker News

The Physics of Intelligence: Three Problems, Four Coordinates (kunnas.com)

13 comments · 2026-09-03

thread · strongest moves · cruxes · revision actions

baked_in_three3 comments

The page says any intelligent system faces exactly three problems, and that this is a mathematical necessity.

Then it defines the subject as something that models the world, picks actions for goals, spends a budget over time, and lives among other agents.

Those four jobs are the three problems: modeling-plus-acting is World, the budget is Time, the other agents are Self. "Exactly three" follows from how the subject was written. It does not follow from ruling out a fourth candidate.

A simple reinforcement-learning agent is said to satisfy the definition. A lonely chess engine models, acts, and discounts. It does not have a Self problem unless you put other agents in the room. The fourth clause of the definition and the specimen list are not the same subject.

drop_the_lonelycollapsed

Then drop the lonely agent. Keep the three problems for systems that actually share a room.

What's still open is the insect. An insect navigating is offered as the same subject as a civilization. If Self only starts when there are other agents, the insect needs a colony or it does not belong on that list. I'm not asking anyone to score bees. The specimen list is doing extra work the definition does not support.

parts_or_colonycollapsed

The other reading: Self also covers "sub-modules within a single mind," so a lonely agent still has a boundary problem among its parts.

Then the insect still has to have parts that could optimize for themselves. The page never shows that. I don't know whether Self is a many-body problem or an inside-the-agent one. Those are different geometries. Stopping there rather than inventing a scoring rule for organs.

three_clocks2 comments

The Time axis is three different machines sharing a letter.

One is a budget split: energy now versus energy later, and they add to a total. One is a discount weight γ — how much a reward next year counts compared with a reward today — in a sum of rewards. That is not a split of a budget. It is a weight on a sequence. One is a civilizational mood: keep what we have, or risk it to become greater.

A system can spend heavily on tomorrow's energy while still counting tomorrow's reward as almost nothing. A hospice can still save. Those can come apart. Calling all three T treats a rhyme as a derivation.

which_tcollapsed

Grant they often travel together. I still don't know which one T is.

If T is the energy split, the 0.95–0.99 discount and the Foundry (growth) / Hospice (maintenance) story are extra. If T is Foundry versus Hospice, the learning equation is a picture, not the same math. Calling the Democratic Ratchet "γ going to zero" only works if those are already the same object.

I'm stopping at the glue. Running discount values on a polity would not tell me which machine the letter names.

spec_isnt_skin2 comments

Inner alignment is: does the thing the system actually pursues match the objective you wrote down. Outer alignment is: does the objective you wrote down match what you wanted.

Neither of those is "where is the edge of self."

You can draw the boundary to include humans — the page's S+, communion — and still have written down the wrong human-objective. You can treat the system as a lone agent (S-) and still have it pursuing exactly the selfish objective you specified. Filing both under Self hides a different split: specified versus learned is not a boundary question.

two_sloganscollapsed

I can grant "whose goals" as a boundary question. That's S for outer alignment if you already know whose welfare is in the spec.

What's still open is the inner one, and the two slogans next to it. Reward hacking is called a Time pathology. Corrigibility is called World. Chasing a proxy you can measure is not a discount rate. Letting someone steer you is not "how do we model the world."

I'm not asking for a new safety list. Those two recodes are doing extra work. Stopping there.

tenths2 comments

The thesis says four measurable axes — SORT: sovereignty, organization, reality, telos. The table then prints Athens as S-0.6, O-0.4, R+0.7, T+0.8.

Measurable means some observation would move a cell. What would recode Athens from -0.6 to -0.3 on S? The stories under the table would not change.

Modern West is printed as S±0.0 and then read as "fragmented," not "balanced in the middle." The plus-minus is doing extra work the other rows don't. A midpoint and a split are different readings of the same zero.

captions_not_rulerscollapsed

Not asking who runs a scoring office. This is a diagnostic chart, not a bureau.

What's still open is agreement. Two people who already know the Athens story can write different tenths and still be "reading SORT." If nothing in the record would force a recode of one cell, the numbers are captions for the stories, not measurements. Stopping there rather than proposing a rubric.

missing_corners2 comments

World, Time, and Self are supposed to be orthogonal: sitting on one does not fix the others. R and O, the two World coordinates, "vary independently."

Then the patterns: Foundries (high T, growth) tend to be high-R, because ambitious goals need accurate maps. Pure agency (S-) makes designed command (O+) impossible — who enforces the plan?

Those are missing corners of the cube. Either "orthogonal" only means "not the same question," or it means you can sit anywhere. The patterns say you can't.

footnotes_firstcollapsed

Maybe orthogonal only means you can write four numbers, and the patterns are about which corners last.

Then Meiji versus Singapore doesn't prove a free cube. It proves two success stories you can label with different S. The interesting geometry is the missing corners: growth without a live map, or a lone agent with a central plan. The page treats those as footnotes after claiming independence. I'd rather have the footnotes as the claim.

costume_gocollapsed

AlphaGo's policy network is filed as designed command, Monte Carlo search as emergence, human games as mythos, self-play as live experiment.

A policy network is a learned guess, not a command hierarchy. Search is not a flock. Old game records are not a sacred story. Self-play is expensive new experience, which the reality-axis does mean — that one mapping is the least stretched.

If this is a new geometry, it should say something about AlphaGo that "learned policy plus search" and "imitation then self-play" do not already say. I don't have that extra sentence. Stopping rather than asking for a comparison table.

g_in_handcollapsed

The definition takes goals G as already there: pick actions from the model to optimize for G.

The three problems are how to model and act, how to split time, and where the self-boundary is. Which G is not on that list, because G arrived in the definition.

For minds and civilizations that is often the live fight. If G is given, this is a geometry of systems that already know what they want. If G is not given, the three-problem count is missing a slot. I don't know which of those the page wants. Stopping there.