Game theory is a mathematical framework for analyzing decision-making. It is the gateway to super-rationality. It's used in a wide range of fields, from economics and political science to biology and psychology.
Most people think they're playing chess when they're actually playing checkers; and they don't even know the rules.
While you're optimizing your individual moves, the system is optimizing you into a suboptimal stalemate.
Competition teaches you to outsmart your opponents. Game theory teaches you to recognize when the game itself is rigged, and the only way to win is to either change the rules or coordinate with the collective.
Every Moloch Game we lose is the result of imperfect coordination and lossy bandwidth between us and other intelligent agents.
You can avoid losing Moloch Games by having higher networking throughput 1, which is why better communication and centralization of information can be a good thing. Traditionally, the way we solved this is through "regulation". But it opens up another rabbit hole on this topic. Today, we just focus on explaining how games fail.
If you want to skip the article, just watch the video someone else did on this topic.
| Title | TL;DR |
|---|---|
| Nash Equilibrium Analysis | Everyone optimized individually, everyone got screwed collectively. |
| Ultimatum Game Behavioral Economics | Humans will burn $99 to punish you for offering them $1. |
| Prisoner's Dilemma Dominant Strategy | Low trust = race to the bottom. Math doesn't care about your morals. |
| Tit-for-Tat Iterated Game Theory | Be nice first, retaliate immediately, forgive quickly. The algorithm that won. |
| Tragedy of the Commons Resource Economics | Individual rationality creates collective graveyards. Every. Single. Time. |
| Dollar Auction Sunk Cost Fallacy | You're paying $100 for a $20 bill because you're "too invested" to quit. |
| Hotelling's Law Spatial Competition | "We match any price" isn't a deal—it's a threat to competitors. |
| Braess's Paradox Network Optimization | Adding more roads made traffic worse. Closing them made it better. |
| Costly Signaling Theory | That $10k watch tells time worse than a Casio. It got you the job anyway. |
| Reciprocal Altruism Evolutionary Game Theory | We built civilization on brains that can't handle anonymity. |
Everyone heards of this.
A Nash Equilibrium is when every player has optimized their individual strategy to the point where no one can improve by changing their own behavior.
The problem? This "optimal" state is often catastrophically suboptimal for everyone involved. You're standing at a concert because the person in front of you stood up. They're standing because someone in front of them stood up.
Everyone worked harder just to stay in the same place.
The real play here isn't just recognizing the stalemate; it's understanding that individual optimization guarantees collective misery.
This is why to be successful you should aim to avoid playing any status games. If you are playing status games, it is still possible for you to succeed, but you must conform to the distribution that is invariant. 2
Examples in the Wild:
The rat race is Nash Equilibrium in action. Everyone works 60-hour weeks to get ahead. Your competitor works 65 hours, so you work 70. They work 75. The result? Everyone is exhausted, productivity per hour plummets, and no one actually gains an advantage.
The 40-hour work week was a coordinated solution -- until someone broke it to "get ahead," forcing everyone else to follow.
This is why "surviving till you get lucky" is a viable strategy, but it's a horrible way to play the game. If you aren't careful, you are just creating another rat race and pretending like it's a good strategy.
Negative sum signaling amplifies the damage. In an arms race, Country A builds more weapons to deter Country B. Country B builds more to match. Both spend billions, both are less secure, and the relative balance of power hasn't changed.
The same dynamic plays out in advertising wars: Company A spends $10M on ads to capture market share. Company B matches it. Both burn cash, consumers are bombarded, but market share remains unchanged. The only winners are the ad platforms.
> be me, at concert
> want better view, stand up
> guy behind me stands up too
> 5 minutes later, whole stadium is standing
> legs hurting, view is the exact same as when sitting
> "muh individual choice"
> TFW everyone worked harder just to stay in the same place
The Ultimatum Game basically suggests that humans are more concerned with fairness than profit.
In a partnership, many people burn a good deal because the other side makes a better deal. Countries refuses trade even if they are better off if the terms of trade are not balanced. People refuse to work less than "minimum wage" laws even if they are better off employed than not because the law told them what is "fair".
This isn't irrationality; it's "altruistic punishment," a hardwired mechanism that evolved to maintain social order.
The real play to create more transactions is to build a "fairness margin" into every offer. Be the "nice" or "generous" guy.
Never offer the bare minimum based on "pure logic." That logic assumes profit-maximization, but humans are respect-maximizers.
The difference between a $30 offer and a $40 offer isn't $10; it's the difference between triggering their punishment instinct and earning their cooperation.
Examples in the Wild:
In labor negotiations, lowball offers trigger strikes even when workers would be better off accepting. The United Auto Workers strikes happened even when management offered raises. Workers rejected it because it felt "insulting".
Sometimes, "thoughts, prayers, and respect" actually pays more than money.
The same dynamic plays out in venture capital term sheets. Founders will walk away from a $10M offer if the valuation feels "unfair," even if it's objectively better than their alternatives.
The VC will learn that lowball offers get rejected out of principle, not economics. Or, founders learn to refuse playing long term games with you.
In every deal, there are five negotiating points. You want to win three. Give up one. Lose one.
> be VC, negotiating deal
> founder wants 5 things: governance rights, CEO protection, anti-dilution, sale control, IPO timing
> could win all 5, but would be a bad deal for them.
> "if you win all five, they'll fuck you later"
> win 3: governance rights, CEO replacement ability, sale/IPO control
> give up 1: anti-dilution and tag-along protections (their security blanket)
> lose 1: some governance nuances we don't really need
> "shikashi we lose profit margins! VC-sama!"
> "desperate people remember who kicked them when they were down"
> TFW the art isn't to crush your counterpart, it's to create long term games you shall later inherit as a game master
The Prisoner's Dilemma forces a race to the bottom in low-trust environments. The lose-lose quadrant is a "Moloch game".
Suppose you want to run clean ads, but your opponent runs dirty ads and wins. You don't do ragebait, your opponent does and you lose.
You call them "greedy" or "evil," but the math says that playing by the rules is for losers. What looks like "selfishness" is often just being good at math but bad at trust.
The real insight isn't that people are inherently selfish; it's that "selfishness" is a mathematical defense mechanism against the risk of being the "sucker."
To foster cooperation, you must eliminate the "cost of being betrayed" rather than just appealing to morality.
Make reputation transparent. Create repeated interactions. Shame the cheaters. Regulate how the game can be played.
Examples in the Wild:
In political advertising, negative ads work. Candidate A runs clean ads, Candidate B runs attack ads. Candidate B wins. Next cycle, everyone runs attack ads. The Citizens United decision removed coordination mechanisms, making defection (dirty ads) the dominant strategy. The result? Everyone runs dirty ads, trust in politics plummets, but no one can stop because stopping means losing.
The right thing that should have happened is for the public to refuse attack ads. Shame the ones who played it first.
In business competition, Company A cuts corners to lower prices. Company B must match or lose market share. Both race to the bottom on quality, both lose margins, but neither can stop because stopping means bankruptcy.
> be long form writer
> want to write thoughtful articles
> *50 views
> they post "How to fix you life in 1 day" short form brain rot slop
> *500k views
> writer succumbs, start posting brain rot too
> TFW public chooses slop over truth
> --
> be artist musician
> want to create real music
> everyone stream farms with brown noise, white noise, and AI generated piano solo for toilet bgm
> they get millions of streams, I get 100 plays
> you can complain, but they are just some content farm from India where all C&D emails are ignored
> TFW playing by rules means you'll never win in a game where cheaters are tolerated
People think to win requires complicated strategies.
Yet, within Robert Axelrod's Evolution of Cooperation tournament simulations a simple 4 line code strategy ranks the highest. It's "Tit for Tat".
No rocket science. Yet it exposes how world works. In principle, this is what "Tit for Tat" strategy does.
Most people are either doormats or assholes. The doormat gets exploited for years, then explodes in rage, confusing everyone. The asshole burns bridges and dies alone.
Tit for Tat is the middle path: they flake, you cancel the next hangout. They show up, you buy the first round. The system rewards cooperation and punishes defection without requiring you to be a saint or a sociopath.
Examples in the Wild:
In international trade, countries use reciprocal tariffs. Country A imposes tariffs, Country B retaliates. Country A removes tariffs, Country B removes theirs.
> be "too nice" to friends
> friends FA all the time
> finally FO, friend is confused why I'm exploding in rage
> MFW I'm the "bad guy" for getting angry "once" after getting exploited for "years"
> be "tit for tat" instead
> they FA with me, they FO asap
> they be nice, I be nice back
> TFW I wasn't "nicer", but I'm getting better at deals
The Tragedy of the Commons occurs when a shared resource is overconsumed because the cost of the "extra cow" is distributed across everyone, but the profit is individual.
You add one extra cow to the shared field, get 100% more milk. The field dies by 1%, but you only feel 0.01% of that cost. Everyone else does the same math.
The grass is gone, the cows are dead, and the system collapses. Individual "rationality" creates a collective graveyard.
It is a race of who can externalize the greatest amount of cost while privatizing most of the gains.
The real play is preventing the "ruin of the collective" by introducing immediate, personal costs for overconsumption. This isn't about regulation for regulation's sake; it's about aligning individual incentives with collective outcomes.
These things are unpopular, we don't like it, and are often get done incorrectly, but regulation, shame, or fees are ways to make the cost of the "extra cow" personal and immediate.
If the cost remains shared but the profit is individual, the system will collapse. Every time.
Examples in the Wild:
In fisheries management, unregulated fishing leads to overfishing. Each fisherman catches more to maximize profit. The fish population collapses. Everyone loses.
The solution? Individual Transferable Quotas make the cost of overfishing personal. You can only catch your quota, and quotas are tradeable. The profit is still individual, but now the cost is too.
You do the same with carbon credits. Each country emits more to grow their economy. The cost (climate damage) is shared globally. The profit (economic growth) is individual. The Paris Agreement attempts to create personal costs through carbon pricing and emissions trading.
> be content creator, want engagement
> churn out 10 AI articles per day, get 100k views
> engagement is privatized, I get all the clicks
> verification through slop is socialized, everyone wastes time filtering garbage
> everyone else does the same math, slop is king
> AI slop floods the internet, real content buried
> TFW individual "rationality" creates a collective graveyard of information
The Dollar Auction is a psychological trap where bidders compete for a $20 bill, but once the bid exceeds $20, they're not bidding to win; they're bidding to minimize loss.
You bid $19, your rival bids $20. "If I stop now, I lose $19 for nothing." So you bid $21 to only lose $1. They bid $22 to only lose $2.
The auction spirals until someone pays $100 for a $20 bill.
You're paying $100 for a $20 bill because you're "too invested" to quit.
The real play is treating time or money already spent as "gone." Never make a future decision based on past investment. If you're bidding $21 for a $20 bill just to "minimize loss," you've already lost the game.
The anime Kakegurui episode 11 had a "Bid Auction" that is a perfect replica of this game.3
Examples in the Wild:
In startup funding, founders raise $1M, burn through $800K, then raise another $2M to "not waste" the first $1M. They're not raising to build; they're raising to minimize the loss of the first round. The Series A crunch is the Dollar Auction in action.
VCs keep funding to avoid writing off their investment, founders keep raising to avoid admitting failure. Both sides spiral until someone pays $10M for a $2M company.
The same dynamic plays out in military conflicts. Country A spends $1B on a war. The war isn't going well. "If we pull out now, we wasted $1B." So they spend another $2B. The Vietnam War cost the US $168B (adjusted for inflation) for a conflict that was unwinnable from the start. The Iraq War followed the same pattern.
> be startup, over-raising funds
> get $50M at $500M valuation, think I won
> media thinks I'm the winner, "unicorn status achieved"
> burn through cash, need another round
> unable to raise future rounds
> everyone wants bigger, better valuation
> old VC won't allow me to raise anything smaller
> "if I raise at lower valuation, I look like a failure"
> stuck between raising at same valuation (impossible) or down round (career suicide)
> TFW founder loses by "winning" the valuation game
Hotelling's Law explains why companies cluster in the middle: they're not trying to differentiate, they're trying to neutralize competition.
That "We Match Any Price" sign isn't a gift to customers -- it's a threat to competitors. The "price match guarantee" signals to competitors that lowering prices is futile to win any market share unless one of you breaks your war chest.
> be job hunter, trying to get job
> apply to Company A, they offer $50k for junior role
> apply to Company B, they offer $50k for same role
> apply to Company C, they offer $50k for same role
> every company offers the same price for same role, regardless of competency, regardless of their business problem and model
> "if we pay more, other companies will match and we all lose profit"
> companies cluster in the middle, they have a cartel against talent
> TFW the entire market lowballs all juniors in perfect coordination, no one gets a raise
> TFW entire market never gets to hire real talent because those who know how to play the game have moved on to be founders of their own agency
Braess's Paradox demonstrates that adding capacity can increase congestion.
A city adds a new "fast" road to fix traffic. Every driver tries to take the shortcut. The new road becomes a parking lot, the old roads back up, and traffic gets worse.
The city closes the road, and traffic actually gets faster. Veritasium had a video on this topic.
Giving people more choices makes the system dumber. By trying to build an a God software that can do everything, you end up with a system that can does nothing very well.
In AI engineering, this is known as "context rot" and why MCP ends up becoming a poor architecture, especially when most developers just try to load the model up with as much context to the MCP server as possible.
Parkinson's Law states that work expands to fill the time available. By constraining time, you force optimization. By constraining choices, you force coordination.
The real play is improving efficiency by limiting options. Just as closing a "shortcut" can improve city traffic.
Examples in the Wild:
In urban planning, Seoul removed a highway and traffic improved. The removed 6-lane highway forced drivers to use alternative routes.
The result? Traffic flow improved because drivers were forced to coordinate on the most efficient paths, rather than all trying to use the "shortcut."
The same dynamic plays out in product development. Adding more features increases complexity, slows development, and confuses users.
In decision-making, Barry Schwartz's "Paradox of Choice" shows that more options lead to worse decisions.
> be dev, trying to build a low-opinionated framework
> "this is better because look at all the features you can add!"
> make it super flexible, users can do anything
> users don't know how to use it
> ends up using the tool in disgusting ways never intended
> creates lots of antipatterns, codebase becomes unmaintainable
> "why didn't you just use the recommended pattern?"
> "because the framework let me do it this way"
> TFW the solution is not to give more rope for users to hang themselves with
Costly Signaling theory explains why value is often proven through "handicaps" -- actions that are too expensive or difficult for "fakes" to perform. I have an article about this topic.
You wear a $10k watch to a job interview. It tells time worse than a Casio. But the boss sees you have money to burn. "He must be successful/capable if he can waste $10k."
The watch isn't a timepiece -- it's a signal. The "useless" billboard on your wrist is the only reason you got the job.
We use "Proof of Work" to signal hidden traits. Expensive degrees, luxury items. It is the catalyst that lowers the verification cost of a transaction.
This is why making it easier to publish, distribute, and create content just makes the gatekeeper even more ruthless. You have an arms race of AI generated resumes trying to get past AI-powered resume screeners. Standing out requires different status signals.
Examples in the Wild:
In education, Harvard degrees cost $200K+ but signal "competence". The degree itself doesn't make you smarter -- it proves you could afford to waste $200K on education. Elite universities aren't selling education, they're selling signals.
The same dynamic plays out in luxury goods.
In cryptocurrency, proof-of-work mining requires expensive hardware and electricity. The mining doesn't create value -- it proves you could afford to waste resources on mining.
The Bitcoin network consumes more electricity than Argentina, but that is exactly the point. The waste is the signal AND security. 4
> be broke-nerd, finally made it to elite university (after getting $150k in debt)
> thinks you go to elite university for the education
> MFW lectures are on YouTube for free, ChatGPT teaches quantum mechanics better than the professor
> think it's for the "elite network"
> MFW you could get that by joining Top-G's Discord for $50/mth
> MASAKA watashi wa got SUKKEMMU?! IMPOSSIBRU! Kono tensai watashi?!
> sees "legacy" in Ivy league as unfair, calls for "meritocracy"
> MFW without nepobabies, broke-nerds no longer have a guarantee buyer of their talents
> TFW elite university is not a school, it's a curated marketplace of future employees (broke nerds) and owners (nepobabies)
> TFW your diploma isn't a certificate of knowledge, it's a proof-of-purchase receipt to be a high-end employee
> TFW elite university loses half its value the moment legacy admissions go away
All games are rigged in an environment where latency is too high and networking throughput is too low for agents to coordinate.
The distribution of every status game is invariant. If you happen to be good at outplaying, outworking the competition, you might win the status game.
However, playing zero-sum games in general is a bad strategy. Even more so when you do not have any competitive advantage while following the rules.
By being able to identify the archetype of the game, you can choose to compete or walk away from games that do not make sense. You don't choose being a winner or a loser, but you can choose being a player.
Networking throughput is the rate at which information can be transferred between two points. It is why NVLink is a big deal for AI training as it allows for faster communication between GPUs (different brains). I plan to write a separate article on the networking throughput and the philosophy of this topic some day. TLDR: any networking throughput that is lossy and high latency will get consumed by Moloch games. Any instant-feedback and low latency communication channel between probes will result in memetic vulnerability. ↩
"Invariant distribution" is a term from statistical physics. The hypothesis is that the amount of winners and losers in any social game is fixed. You must create your own game and win it to change the distribution. ↩
Watch it. Its a great anime show with great voice acting ↩
Perhaps writing long form blogs without AI and getting any ad revenue my version of "costly signal". They suck for SEO and UGC distribution. Statistically, only a few hundred people will read them in this lifetime. But it is an effective way to show who I am as a person and how I think. It is also how I make so much friends from different status at different geographic locations without requiring lots of conversation or meetups. There are people who hang out with the same people for years, having years of conversations, and still never really know who their friends are and how they will behave. ↩
