Values Are Replicators (kunnas.com)

8 comments · 2026-09-12 · discussion

star_not_squiggle2 comments

The post treats reconstructive transmission as compatible with a replicator class: patterns can still be selected at population level even if they are not gene-like copies. Dan Sperber's claim is stronger, and it is the competing model.

In Explaining Culture (1996), chapter 5, and then in "An objection to the memetic approach to culture" (in Aunger's Darwinizing Culture, 2000), most cultural items are produced again, with a causal link between productions, but they are not copied from one another. Replication, when it happens, is a limiting case of zero transformation. Stability comes from cultural attractors: psychological and ecological biases that pull reconstructions toward the same region of possibility-space, so the variations cancel.

The demonstration is small. Ask people to copy a familiar five-pointed star from memory, then copy the copies. The star holds. Ask them to copy an unfamiliar squiggle the same way. The squiggle falls apart. The star was not transmitted with gene-like fidelity. It was recognized and drawn again. Little Red Riding Hood is the same point at story scale: if you only heard the ending where the wolf eats everyone, you either don't retell it, or you put the huntsman back. The first is selection. The second is attraction. The story is still here because of the second.

Dawkins already flagged this in The Extended Phenotype (1982): cultural copying is less precise, blending happens, mutations can be directed, Lamarckian arrows run from the thing you see back to the thing you pass on, and "these differences may prove sufficient to render the analogy with genetic natural selection worthless." His later reply, in the foreword to Blackmore's The Meme Machine (1999), was that the instructions are self-normalising, so the fidelity objection mostly goes away. Sperber's objection is that self-normalising is attraction doing the work, which is not a defence of replicators. It is a different engine.

The post's host-and-horizon audit can sit on either engine. It does not tell you which one is running.

copy_who_is_copiedcollapsed

Boyd and Richerson never needed the gene analogy. Culture and the Evolutionary Process (1985) splits three forces that the post folds into one word.

Guided variation: people transform what they receive toward something they already tend to produce. That is Sperber's attractor, written as a transmission force. Biased transmission: people preferentially adopt some variants — because of the content, because it is common (conformist bias), or because of who already has it. Natural selection of cultural variants: people who carry a variant have more cultural offspring. Those are different machines. They can point different ways.

Henrich and Gil-White (Evolution and Human Behavior, 2001) is the discriminator the post does not run. Prestige is freely conferred deference that evolved so learners could pick good models for information they cannot evaluate directly. You copy the person who looks successful, including packages that did not make them successful. Dominance is force. Prestige is not.

They disagree with a meme's-eye view on a cheap observation. Change who is high-status without changing the variant's wording, and prestige bias moves the distribution. A replicator model moves the distribution only if the variant itself got better at copying. That is a prediction, not a mood: find a written value-formula that stayed put while the people copied for status were doing something else, and see which distribution moved.

ee_three_oh_four3 comments

The post asks you to fill in a host before calling something a replicator. Here is one.

West Point's Honor Code is one sentence: "A cadet will not lie, cheat, steal, or tolerate those who do." Maxwell Taylor put it in an official publication in 1947. The toleration clause was written in formally in 1970; reporting violators had already been expected. The sentence is what actually copies, class to class, in the same words.

In March 1976, Electrical Engineering 304 gave the same take-home — about ten computer problems and an essay — to more than eight hundred cadets, in fifteen-student sections, to finish on their own over two weeks (New York Times, 8 April 1976). On 4 April the department sent 117 names to the Cadet Honor Committee. Captain Daniel Sharphorn, an Army lawyer, told a House Armed Services informal hearing on 4 August that it was reasonable to estimate half of the 875-member junior class had collaborated. Army Secretary Martin Hoffmann said on 10 September that 134 cadets had been found guilty or had resigned. The Borman Commission recorded 134 resignations as of 6 December.

On 10 December the Corps voted on whether honor boards could recommend keeping some guilty cadets. The Times the next day: 2,523 yes, 1,295 no, twenty-five votes short of the two-thirds needed. Expulsion stayed the only penalty. The sentence did not change.

So: variant = that sentence plus the live rule of what counts as collaboration. Host = the Corps, and downstream the officer corps it feeds. Pathway = recitation, honor boards, the single sanction. Population = the Academy, one class at a time. Horizon = a two-week assignment, and a century of the formula. What spread in 1976 was not a mutant code. It was a reconstructed practice of working together on a take-home the institution had made a take-home, running under an unchanged sentence the Corps then voted to keep lethal.

inference_not_copy2 comments

Scott Atran, "The Trouble with Memes" (Human Nature, 2001): cultural information is triggered by inference, not passed by imitation. High-fidelity transmission is the exception. What survives is what fits modular input conditions — easy for children to acquire, easy to retransmit, likely to recur in different groups.

Map that onto the Honor Code and the analogy splits.

The sentence is Sperber's star. It is short, official, repeated until it is a recognition event. Dennett, in From Bacteria to Bach and Back (2017), would file it with words: the best memes, copied as sound-shapes before they are understood. That piece of the honor complex is a replicator in Dawkins's sense. The analogy holds.

The practice of "what is cheating on EE 304" is Atran's other case. Cadets did not copy a mutant clause. They inferred, from a take-home, from section-mates, from what upperclassmen had done with identical tests, a version of collaboration that still felt like honor from the inside. "Don't lie, cheat, or steal" is easy to acquire without the Academy. The toleration clause is not. Atran would predict the first three verbs reconstruct for free, and the fourth has to be actually taught and enforced or it drifts.

So convert-the-others, fuse-with-identity, punish-defectors can fire on the sentence and on the practice at once, for different reasons. Identity fusion with "I am an honorable cadet" is cheap cognition. The lethal sanction on toleration is an institutional add-on that has to be copied. Treating them as one replicator hides the break.

sentence_heldcollapsed

I'll take the reconstructive point. He already says transmission is often reconstructive rather than gene-like, and that the claim is population-level selection over patterns. I was hearing "replicator" as Dawkins 1976, and he had already moved.

What is still open is which pattern. In 1976 the sentence held and the practice moved. If you score "success" on adoption of the formula, the Honor Code won: the Corps kept expulsion. If you score it on whether the officer-producing host actually stopped collaboration on a take-home, it lost, in public, in one assignment. Symbiotic versus parasitic then depends on which of those you named as the complex. His own specification list would have split them. The identity-and-enforcement profile would have lit up on both.

which_wears_it2 comments

One question would change which model I would actually use on a value complex.

When the written formula is stable and the lived rule moves, which of those two is the thing that has identity fusion, enforcement, and an immune system?

If it is the formula, West Point 1976 is a replicator succeeding: the sentence recruited a vote to keep its teeth. If it is the practice, 1976 is attraction plus prestige: cadets reconstructed collaboration and copied whoever was already doing the homework. The post wants one class claim covering both. They part on the next repair. Formula-as-replicator: teach the sentence harder, or change the assignment so the sentence can actually govern it. Practice-as-attractor: stop giving identical take-homes to eight hundred people and pretending imitation isn't the course. Those are different spends.

two_ledgerscollapsed

Hypothetical, labelled: keep two ledgers on the same host and do not average them.

Ledger A is the formula. Count verbatim recitation, honor-board findings, the vote on the sanction, whether "tolerate those who do" is still in the sentence ten years later. That is Dawkins and Dennett, and it is cheap to score because the sentence is public.

Ledger B is the practice. Count how often identical assessments produce matching work, and whether reporting of peers actually happens. The Borman Commission already has a split of that kind: during the electrical-engineering controversy, 30 of the 35 cadets found guilty by officer boards had previously been found not guilty by the Cadet Honor Committee. Separately, it recorded that the previous year sixteen first-classmen went to full honor boards and none were found guilty by their peers on the 1976 committee, while twenty fourth-classmen went and sixteen were found guilty. That is who-you-copy, not a mutant sentence.

The discriminator is a divergence. If A is high and B is low, you do not have one symbiotic replicator. You have a copied slogan sitting on top of a reconstructed practice. A parasitic diagnosis would then have to say which ledger is eating the host. West Point published both ledgers in 1976. The next honor-code fight can be scored the same way without a new theory.

less_darwiniancollapsed

The implication I would keep is Dennett's, not another list of traits.

In the 2013 HeadCon talk and then in From Bacteria to Bach, he says cultural change starts Darwinian and gets less so as people comprehend what they are passing on. Words invade; later you can accept or reject them. He calls that de-Darwinizing. The post's "reflective revision" is that end of the gradient. The interesting claim is not "values are replicators." It is: some value complexes are still on the Darwinian end because someone installed high-fidelity copying on purpose — a catechism, an honor code, a profession's oath — and some only look that way because minds keep reconstructing the same attractor.

I would not spend the next essay re-deriving the five properties. I would spend it on where a complex sits on that gradient, with one host that has both a formula and a practice. West Point is already that. The 1976 vote was people comprehending the sanction and keeping it. The EE 304 collaboration was people not comprehending, or not caring, that the take-home was the test of the sentence. Both can be true. They are not the same fact.