The Dark Miracle of Optics

by Suspended Reason7 min read24th Jun 20205 comments


Philosophy of LanguageGame TheoryEvolutionSlackWorld Modeling

Alternate titles:

  • The Public-Private Information Gap Rules Everything Around Me
  • Baudrillard’s Simulacra, Steelmanned
  • “Have your cake and eating it too”
  • The Precarity of Prestige Economies
  • “Goodhart’s is just a subset, mannn.”
  • “Costly signals are just a subset, mannn.”
  • The Tragedy of Appearances
  • On Truth & Lies in a Nonmoral Sense

Epistemic status: no idea how original any of this is; it just connects a lot of nodes in my brain. I’ve been told there’s a real debt to Robert Trivers, which I hope to educate myself on shortly. I may just be reinventing signal theory.

In the beginning there was defection.

We can treat a prisoner’s dilemma as an elegant stand-in for coordination more generally. A one-off dilemma has as its ideal solution defection. Bellum omnium contra omnes: the war of all against all, or, “hyper-individualism.”

At the same time, it is clear that many of the “benefits sought by living things”[1]—which is to say, that which assists survival—are more readily reached by group effort.

Crucially, an iterated prisoner’s dilemma has the opposite optimal equilibrium: tit-for-tat, or cooperation, in its many variations, its various guises. And so the trick becomes how to switch individuals over from one-offs onto iterated dilemmas. The technology which achieves this is reputation, allowing formation of a ledger anchored to names[2], faces, identities. Individuals sharing an ecosystem continually run into each other, and given a reputation ledger, cannot defect and “get away” with it, continuing to freeride into oblivion.[3]

Tit-for-tat is exceedingly simple. It enables mutualism (in symbiosis[4]) and is practiced internally to species as diverse as stickleback fish, tree swallows, all primates, bats. All it requires is a sense of continuous identity and tracking of that identity’s (recent-)historical actions. We can take bats as an example: Since mothers’ hunting returns are unequally distributed, but bat babies do better when consistently fed, mothers communally share food. But should a researcher pump up a mother’s gullet full of air, so it appears she had a strong return but is refusing to share, suddenly the other mothers will no longer feed her children, will allow them to starve.

We can read human progress as a history of instituting cooperation. The Stele of Hammurabi’s famous law is eye for an eye; many of its other decrees are variations thereof: if a building collapses and kills its occupant, its builder shall be put to death as well. The Old Testament introduces the Commandments, the laws of Exodus. Almost every major religion has its in-house variation on the Golden Rule. These are efforts at securing internal coordination of the group, which a la Studies on Slack” and multi-level selection theory, will outcompete other groups once instituted. I have heard from reliable sources that laws in the Quran, and many other major religious texts, have similar structures.

But vanilla tit-for-tat reputational ledgers, like a barter system, are difficult and costly to continuously verify. It requires small, local communities of recognition, and prevents civilizations from scaling up. And so there was a need for currency, for credit, for the accumulation, transportation, and commensurability of capital, all of which together say: this individual cooperates. (Or to generalize across contexts, since optics signal more broadly than mere cooperation: This individual carries value; an interaction with her will be positive-sum.) This currency needed to be legible and exchanged across markets, across subcommunities. For these and a thousand other reasons we invented proxies, heuristics, measurements; instituted CVs, letters of recommendation, titles of achievement and nobility, and of course, fashion. But currency is easily counterfeited.

Clothing arises to serve object-level purposes: warmth from cold, shelter from sun. But soon it gives rise in turn to fashion: equally tactical, but suited for the social, rather than literal, landscape. (For a social being, both are essential to survival.) Because the garments, the choices of paint pigment, the geometric patterns and metal hanging from the ears, reflected both the wealth of the individual and his affiliation to group, they became sources of information for recipients, on-lookers: ways of deducing whole from part, of grokking a person. As social reality devours the physical—in Baudrillard’s terms, simulacra—thus fashion devours its mother.

In the Upper-Middle Paleolithic Transition, human societies and economies grow increasingly complex. Trade deals and diplomacy are performed among credible spokesmen, and social hierarchies need preservation across interactions between strangers. Fashion enters as a technology for maintaining and navigating the social graph. “By the production of symbolic artefacts that signified different social groups and kinds of relationships, Aurignacian people were able to maintain wider networks that could exist even between people who had never set eyes on each other,” giving them a competitive advantage. The practice spreads through the law of cultural evolution: “The surface of the body… becomes the symbolic stage upon which the drama of socialisation is enacted, and body adornment… becomes the language through which it was expressed.”[5] We have entered the second stage of simulacra. The territory has a map, and there are parties interested in manipulating it.

Or consider the butterfly. A “protected” species (poisonous, inedible, etc) gains a survival advantage through honest signaling of this protection. An honest signal is a Pareto improvement—a win-win. The butterfly does not wish to be eaten; the predator does not wish to eat a toxic insect. How does it evolve this protection?

Brute association. The outward phenotypic expression of the butterfly—its public information—becomes associated with some interior, private information—its toxicity. Let’s say the distinctive pattern is black and red. A predator cannot tell whether an insect is toxic from sight, but it can tell by proxy. In other words, the butterfly develops a reputation.

Once this association between optics and essence, between appearance and reality, between signal and quality (the biological frame) or public and private information (the economic one), is formed, it can be freeridden. It becomes, in most cases, easier to pay “lip service”—to outwardly express the associated public characteristic—than it is to to develop the private characteristic. This is not entirely the fault of the freerider; it is a difficult situation he finds himself in. Imagine he “chooses” (I’m anthropomorphizing evolution) to remain with his blue and yellow colors: even if his “product” is “good” (I’m mixing metaphors, but I mean to say, his advertising is honest), it will take some time for a trusted association between signal and quality, public and private, to form. As consumers, we may initially disbelieve an advertiser’s claims, and for good reason, since there is incentive to deceive. And thus it is with the sun-basking lizard, deciding which butterfly to eat. Far easier for a precarious insect to ride coattails, to imitate and pretend toward what he is not—and so, quite simply, it does.

The connection with fashion should come into view now. The “barberpole” metaphor of fashion, where lower classes continually imitate higher classes, who are themselves engaged in a continual quest for “distinction” from the chasing masses, is a popular one in rationalist circles for good reason. Its cyclical nature is the result of limited options and a continual evasion of freeriders who exploit an associative proxy: clothing for caste.

A quick inventory of where we are: Individuals profit from cooperation, but are always personally incentivized to defect. Reputation ledgers switch us from the one-off tit-for-tat, and its incentivized defection, into an iterated tit-for-tat, and its incentivized cooperation. As civilizations scale, and we wish to do more with what we have, achieve new complexities, we move to an alternate system. A credit system of heuristic and proxy. Thus an individual who wishes to enter the art world will work internships in which she forges relationships of trust, in the hope that she will be recommended. And the employer who takes the recommendation will do so on account of having built up trust with the recommender; this trust is built by history, and its credits are transferable. (Each exchange, of course, comes with a diminishment.) Across many recommendations and positions, across many cities, the accumulating recommendations become virtualized: not only can one fabricate a CV, but one can embellish it, and the latter behavior is so ubiquitous it is hard to call it “cheating,” even though this is what a dishonest signal is. And, at the same time, this intern will find herself competing in a much larger implicit landscape of associations, in which the clothes she wears, the way she speaks, and a hundred other variables come together to—by proxy—provide further evidence of value.

Imagine that a bat mother, instead of having her gullets pumped full of air by researchers, developed a technology to achieve the opposite: to appear as if she had not caught insects, when in reality she had. In other words, to appear as if she was cooperating when in fact she was defecting. To present public information at odds with private information. This bat’s offspring would be most fit, would pass on its genes at higher rates. This bat would have discovered the miracle of optics. But it is a dark, and short-term miracle: the population as a whole would lose its fitness, as its ability to cooperate diminished.

It is better to cooperate than defect. But it is better still to defect while others around you cooperate: to reap the advantages of coordinated effort while contributing none of the work (freeriding). This behavior is blocked insofar as it is noticed. Social systems are not two-player, but N-player games, and resemble public goods games more than prisoners dilemmas, and thus even in the presence of parasites, it can be optimal for other players to invest in the pool.[6] But freeriders remain a burden on the system that rational players will wish to eliminate.

While an honest signal is beneficial to all parties involved—it adds true information to the informational ecosystem which actors can base actions on—a dishonest signal is definitionally exploitative. It causes another self-interested actor to behave against its interest, because its premises are malformed. It causes the sun-basking lizard to pass up on the butterfly, believing it to be protected, when in reality, it is only masquerading.

This is the tragedy of appearances. The cheater is punished if he is caught cheating; a society which punishes cheaters (or “parasites”) outperforms that which does not; and thus his optimal behavior will always be to cheat and pretend otherwise, to evade enforcers. He can do this by means of appearance, and the more that appearance is selected for, the more easily he can simply pretend, while embodying none of the internal, proxied-for qualities. freerider situations don’t work when the supporting actor can recognize freeriding, thus the trick, if one wishes to continue freeriding, is to prevent such a recognition.

This is the superset of Goodhart-Campbell. The solution is the superset of costly signaling. The greater the divergence in the incentive structure between proxy and proxied, the greater the incentives to optimize for appearance. Thus we can understand politics, where everything “real” is hidden behind a great veil, and public image carefully manipulated. Thus we can understand Baudrillard’s simulacra, at least in its steelmanned form: the first level is honest signaling, a one-to-one relationship between public and private. Levels 2-4 are self-aware manipulations, “complex patterns of strategic interactions,”[7] and if you believe Baudrillard, we are long past naivete, between simple one-to-oneness. An unsophisticated relationship to maps is a departure point, not a finish.

The tragedy of appearances, and our incessant optimization thereof, is a problem society does not yet seem to have stable solutions to. Taleb might admonish us, in Skin In The Game, to never trust a surgeon who looks the part, to never employee a straight-A student—but while wise as manipulations of the current fashion field, these are inherently unstable and contingent solutions. As soon as we would follow his advice we would see surgeons trending toward slovenliness, students strategically achieving B-grades in Bio for the sake of seeming interesting. Those familiar with Goodhard-Campbell know the pattern well, and the only answers are the same: diminish the gap between incentivized appearance and desired behavior. Easier said than done.

Or we might move away from proxy, heuristic, appearance; we might ditch the resume and credential. But would we move ahead or backwards? Would we become more or less encumbered, more or less handicapped? Currency can be more easily counterfeited, a silver finish over a nickel core, a nice embossing. “If it looks the part…” But look at currency’s advantages.

I wrote in a recent comment to Zvi’s post on simulacra:

But the actually toxic butterflies—the original honest signalers—they can't go anywhere. They're just stuck. One might happen to evolve a new phenotype, but that phenotype isn't protected by reputational association, and it's going to take a very long time for the new signal-association to take hold in predators. Once other insects have learned how to replicate the proxy-association or symbol that protected them, they can only wait it out until it's no longer protective.

Thus there is an arms race toward manufacturing and recognizing what can only be called “bullshit,” following Harry Frankfurt. It is speech designed to improve one’s image. And as our world becomes more mediated by representation, it in turn becomes more exploitable. We enter the Simulacra.

[1] Axelrod & Hamilton 1981.

[2] The Wire season 2: Barksdale’s crew develops a bad reputation for their underwhelming H, renames it to ditch the old baggage and keep slinging shitty product.

[3] See “recognize and retaliate.”

[4] Hence the parasite, which is a freerider (or worse).

[5] David Lewis-Williams, The Mind in the Cave: Consciousness and the Origins of Art

[6] Thanks to romeostevensit for pointing me toward related literature.

[7] Axelrod & Hamilton 1981.


5 comments, sorted by Highlighting new comments since Today at 1:38 AM
New Comment

You may be interested in my recent convo on super cooperation:

also crossposted from fb:

Social orders function on the back of unfakeably costly signals. Proof of Work social orders encourage people to compete to burn more resources, Proof of Stake social orders encourage people to invest more into the common pool. PoS requires reliable reputation tracking and capital formation. They aren't mutually exclusive, as both kinds of orders are operating all the time. People heavily invested in one will tend to view those heavily invested in the other as defectors.There is a market for narratives that help villainize the other strategy.

[5] David Lewis-Williams, The Mind in the Cave: Consciousness and the Origins of Art

[6] Thanks to romeostevensit for pointing me toward related literature. 

The end of your footnotes don't appear on (the version of this post on) your blog.

Ah thank you, looks like they all got jammed together in the formatting.

Hope you enjoyed reading!

I did. It's great to see this all in one place - it connects a lot of dots, and a long, good read.


After checking both I discovered

[7] doesn't appear on either.


Have you been doing this for a long time?

Thank for the tip on fn 7, it's a pull from an Axelrod & Hamilton paper! Updated.

How long have I been doing what?

Edit: I'll give some possible answers, been blogging regularly six or seven years, been lurking LessWrong four years, writing here a few months, and got into game theory six to eight months ago, though haven't had as much time as I'd like to dig around in it. Still really need to ready Schelling, and would love to do reading on multiagent simulation.