<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Daniel Niiaziiev on Medium]]></title>
        <description><![CDATA[Stories by Daniel Niiaziiev on Medium]]></description>
        <link>https://medium.com/@danielniiaziiev_24831?source=rss-1f0341ffa2aa------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*ywaIP7Muxw7skBsj0Je2bQ.jpeg</url>
            <title>Stories by Daniel Niiaziiev on Medium</title>
            <link>https://medium.com/@danielniiaziiev_24831?source=rss-1f0341ffa2aa------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 04:23:38 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@danielniiaziiev_24831/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Why Cooperation Wins: The Story of a Mathematical Problem]]></title>
            <link>https://medium.com/@danielniiaziiev_24831/why-cooperation-wins-the-story-of-a-mathematical-problem-2d685074fd87?source=rss-1f0341ffa2aa------2</link>
            <guid isPermaLink="false">https://medium.com/p/2d685074fd87</guid>
            <dc:creator><![CDATA[Daniel Niiaziiev]]></dc:creator>
            <pubDate>Fri, 04 Sep 2026 10:27:50 GMT</pubDate>
            <atom:updated>2026-09-04T19:48:12.138Z</atom:updated>
            <content:encoded><![CDATA[<p>An essay on game theory, repeated interactions, and practical conclusions for a team</p><h3>Why read this</h3><p>In 1950 two mathematicians working for the American defense establishment devised a problem meant to show why rational people are incapable of cooperating. Thirty years later a computer tournament proved the opposite: under certain conditions cooperation is not merely possible — it is the most profitable strategy, even when every participant is entirely selfish.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1mCdVui9v4OvAiOfaAzO0A.jpeg" /></figure><p>This is not a moral sermon. It is a result derived formally and verified experimentally. And it has quite concrete consequences for the way teams, departments and negotiations are built.</p><h3>Part I. The problem</h3><h4>The situation</h4><p>Two players independently and simultaneously make a choice: cooperate or defect. They cannot agree in advance and cannot see the partner’s choice.</p><p>Four outcomes are possible:</p><p>Both cooperate — 3 / 3: both come out ahead.<br>A cooperates, B defects — 0 / 5: A is the sucker.<br>A defects, B cooperates — 5 / 0: A profits at B’s expense.<br>Both defect — 1 / 1: both lose.</p><p>The numbers are conventional “utility points” (more = better). They come from Robert Axelrod’s classic work and have become the standard in the literature.</p><h4>Where the dilemma actually lies</h4><p>Look at the situation through player A’s eyes. He does not know what B will do, so he works through both options:</p><p>If B cooperates: by defecting, A gets 5 instead of 3. Defecting is better.<br>If B defects: by defecting, A gets 1 instead of 0. Again defecting is better.</p><p>Whatever the partner does, defection pays more. In the language of game theory this is a dominant strategy — a choice that is better than every alternative under any circumstances.</p><p>But B reasons in exactly the same way. Both defect. Both get 1 point — instead of the 3 they would have got by simply cooperating.</p><p>That is the essence of the paradox: individually flawless logic leads to a collectively miserable result.</p><p>The cell (1/1) is what is called a Nash equilibrium: a state from which no player gains by deviating alone. If A suddenly starts cooperating on his own, he drops from 1 to 0. The equilibrium is stable — and at the same time worse for both than the cell (3/3). Such a state is called Pareto-suboptimal: another outcome exists that is better for everyone at once, but there is no way to reach it single-handedly.</p><h4>The formal condition (in plain words)</h4><p>In the literature the four payoffs are labelled with letters:</p><p><strong>T — Temptation:</strong> I defected against someone who trusted me — 5<br><strong>R — Reward:</strong> both cooperated — 3<br><strong>P — Punishment:</strong> both defected — 1<br><strong>S — Sucker:</strong> I cooperated and was cheated — 0</p><p>For the situation to be a dilemma proper, two conditions are needed.</p><p><strong>1) T &gt; R &gt; P &gt; S</strong></p><p>Read it like this: best of all is to defect against the trusting; next, honest reciprocity; next, mutual war; worst of all is being cheated. It is precisely this ordering that creates the temptation to betray and at the same time makes mutual betrayal a bad outcome.</p><p><strong>2) 2R &gt; T + S</strong></p><p>That is, “playing fair twice” is better than “defecting once and being defected against once.” Check it: 2×3 = 6 against 5 + 0 = 5. Six is more.</p><p>This second condition matters, because without it the players could simply take turns cheating each other and come out no worse in total. When it holds, cooperation creates value that exists nowhere else in the matrix. The game becomes a non-zero-sum game.</p><p>A non-zero-sum game is one where the size of the “pie” is not fixed: depending on the players’ actions the total payoff can be larger or smaller, so your gain does not necessarily mean someone else’s loss.</p><p>In a zero-sum game (poker, chess) everything I win someone else has lost — the sum is always the same. In the example above, mutual cooperation gives both more than any combination of defection — that is, it creates value out of nothing, and that is exactly why the game is non-zero-sum.Part II. The history</p><h4>1950: RAND Corporation</h4><p>Merrill Flood and Melvin Dresher worked at the RAND Corporation, the think tank serving the U.S. Air Force at the beginning of the Cold War. The question on the table was entirely literal: does a preventive nuclear strike against the USSR make sense?</p><p>To study the structure of such decisions they constructed an abstract game and ran an experiment: two colleagues played 100 rounds in a row, recording not only the moves but also their comments. The result surprised the authors. The participants cooperated far more often than theory predicted — and appeared to have grasped something about repetition that the model did not see.</p><h4>1950: Albert Tucker gives the problem its name</h4><p>Tucker, John Nash’s doctoral advisor, was preparing a lecture for psychologists at Stanford and needed a vivid wrapper. He invented a story:</p><blockquote><em>Two accomplices are arrested and put in separate cells. There is no direct evidence. Each is offered a deal: turn the other in and you walk free, while he gets ten years. If you inform on each other, you each get five. If you both stay silent, you each get one year for a minor charge.</em></blockquote><p>Staying silent = cooperating (with the accomplice, not with the investigation). Informing = defecting. Hence the name — the prisoner’s dilemma.</p><h4>1950s–1970s: the shadow of the future</h4><p>It became clear fairly quickly that the whole pessimistic logic rests on one hidden assumption: the game is played once. If the players meet again, defection stops being free — the partner can answer.</p><p>This observation circulated among theorists for so long, and with such unclear authorship, that the result was named accordingly — the folk theorem. James Friedman formalized a version of it in 1971. Its content: in a game that lasts indefinitely, mutual cooperation can be a stable equilibrium — provided the players value the future enough.</p><h4>1980: Axelrod’s tournament</h4><p>Robert Axelrod, a political scientist at the University of Michigan, posed the question differently — not “what does theory prove” but “what actually works”.</p><p>He sent invitations to mathematicians, economists, psychologists and sociologists: submit a computer program — a strategy for the repeated prisoner’s dilemma. The programs would play each against each for 200 rounds, and the scores would be added up.</p><p>Fourteen programs arrived. Some were very complex — building probabilistic models of the opponent, analysing sequences, using adaptive thresholds.</p><p>The shortest one won. Four lines of code, submitted by Anatol Rapoport — a psychologist and mathematician born in Kharkiv, who emigrated as a child and became one of the founders of conflict research.</p><p>The strategy was called Tit-for-Tat:</p><blockquote><em>Move 1: cooperate. Every move after that: repeat what the opponent did last time.</em></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lWy_YJh_0zKAtJc3-0BgSQ.jpeg" /></figure><p>That is all. No model of the opponent, no forecasting.</p><h4>The second tournament: an attempted rematch</h4><p>Axelrod published a full analysis of the results and announced a second tournament. Now every participant knew that Tit-for-Tat had won, knew why, and had time to work out how to beat it.</p><p>Sixty-two programs arrived from six countries. There were purpose-built predators tuned to exploit the naive. There were more sophisticated versions of TFT itself. The number of rounds was indefinite, so that the ending could not be calculated.</p><p>Tit-for-Tat won again.</p><h4>1981: the result leaves economics behind</h4><p>Axelrod joined forces with the evolutionary biologist William Hamilton, and they published the paper “The Evolution of Cooperation” in Science. The central claim: this mechanism explains the emergence of cooperation without intelligence, without morality and without kinship — among bacteria, fish and birds. It is enough for organisms to meet repeatedly and be able to recognize one another.</p><p>The 1984 book of the same name became one of the most cited works in twentieth-century social science.Part III. Why it works</p><h4>The main formula, explained in human terms</h4><p>Axelrod’s key notion is “the shadow of the future”. Let w stand for the probability that after the current round there will be another one.</p><p>The question: at what w does defection stop paying?</p><p>Let us count. You are playing against Tit-for-Tat. You decide to betray.</p><p>The one-off gain: you got 5 instead of 3. You are up +2 (that is T − R).</p><p>The payback: from the next move on, TFT retaliates. You are stuck at 1 instead of 3, losing 2 every time (that is R − P) — and it goes on until someone apologizes.</p><p>Now it is obvious: the longer the game runs, the more the accumulated loss exceeds the one-off gain. Formally the threshold is:</p><p><strong>w ≥ (T − R) / (T − P)</strong></p><p>Substituting our numbers: w ≥ (5 − 3) / (5 − 1) = 2/4 = 0.5</p><p>Conclusion: if the chance of meeting again is greater than 50%, betrayal is mathematically unprofitable. Not “unkind” — unprofitable, on purely selfish arithmetic.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kS0ggaTCzoEV-as5fxqyZw.jpeg" /></figure><p>And from this comes the main lever: to build cooperation you have to increase w. Not to persuade, not to appeal to decency — but to make the future meeting more likely and its consequences more tangible.</p><h4>Four qualities of the winner</h4><p>Analysing what separated the winners from the losers, Axelrod identified four traits.</p><p><strong>1. Niceness</strong> — never defect first. All eight of the best programs in the first tournament were nice. All the worst ones were not. This is the single strongest predictor of success in the whole study.</p><p><strong>2. Retaliation</strong> — answer betrayal immediately. The strategy “always cooperate” loses catastrophically: it feeds predators. Kindness without the capacity to respond is not a virtue but a vulnerability.</p><p><strong>3. Forgiveness</strong> — return to cooperation the moment the opponent does. TFT punishes exactly once for one betrayal and holds no grudge. Strategies that “remembered everything” and took revenge forever turned out markedly worse: they could not get out of the spiral.</p><p><strong>4. Clarity</strong> — be comprehensible. An opponent works out TFT’s logic within a few moves. That is an advantage, not a flaw: as soon as the partner understands the rules, he sees that cooperation pays and starts cooperating. Complex, unpredictable strategies lost, because the opponent could not coordinate with them and out of despair switched to permanent betrayal.</p><h4>The most important detail of the whole tournament</h4><p>Tit-for-Tat did not beat a single individual opponent.</p><p>Not one. Head to head, TFT never scores more than its opponent — by construction it cannot, because it never defects first and retaliates exactly as much as it received.</p><p>It won the tournament because in every pairing it extracted the largest joint payoff. Aggressive strategies beat their opponents — and sank to the bottom of the table together with them, because their games produced 1/1.</p><p>This is a non-zero-sum game. To measure success by the gap between yourself and your partner is to lose systematically. Envy here is not merely an unattractive trait; it is a strategic error.</p><h3>Part IV. What was learned later</h3><p>The model did not stay in 1984. Three refinements worth knowing.</p><h4>1. Noise breaks Tit-for-Tat</h4><p>In reality mistakes happen: a signal does not arrive, an action is misread, a letter goes astray. Imagine one TFT move is accidentally distorted and read as a betrayal.</p><p>The consequence: the partner retaliates → TFT retaliates back → endless mutual revenge over a misunderstanding that never happened. Axelrod called this the echo effect.</p><p>Two well-known solutions:</p><p><strong>Generous Tit-for-Tat:</strong> forgive some betrayals at random — roughly every third one. A single forgiven move is enough to break the spiral.</p><p><strong>Pavlov (Win-Stay, Lose-Shift),</strong> Martin Nowak and Karl Sigmund, Nature, 1993: if the previous move produced a good result, repeat it; if a bad one, change. This strategy recovers from errors on its own and in many environments outperforms TFT.</p><p>Practical conclusion: wherever misunderstandings are possible, mechanical reciprocity is dangerous. You need a built-in reserve of generosity — or the habit of asking again before answering with a blow.</p><h4>2. A known ending destroys cooperation</h4><p>If both players know for certain that there are exactly 200 rounds, then on move 200 betrayal is free — there is no more future. But then move 199 effectively becomes the last meaningful one — and there too betrayal pays. And on 198. This chain of reasoning is called backward induction, and it unravels cooperation all the way back to the start.</p><p>The way out was found by Kreps, Milgrom, Roberts and Wilson (1982): a tiny amount of uncertainty about the opponent’s type is enough — not being 100% sure he is a cold maximizer — and reputation sustains cooperation almost to the end.</p><p>Practical conclusion: an announced end to a relationship is the beginning of its decay. The last month before someone leaves, the final quarter of a contract, “this is a one-off deal” — all of these switch the mechanism off. The riskiest interactions are the ones where somebody knows he will not meet you again.</p><h4>3. Extortion — and why it loses</h4><p>William Press and Freeman Dyson (PNAS, 2012) discovered the class of “zero-determinant strategies”. Some of them are extortionate: they can impose on the opponent a fixed, deliberately unequal share of the payoff. This looked like a blow to the whole optimistic picture.</p><p>But Alexander Stewart and Joshua Plotkin showed the sequel: in a population where strategies compete and spread, extortionate variants are unstable. They milk the compliant well, but they fail against each other. What proves evolutionarily stable are the generous ZD strategies — the ones that split the surplus in the partner’s favour.</p><p>Practical conclusion: squeezing the maximum out of a partner works in isolated cases and against the weak. As a systematic policy it does not scale.</p><h3><strong>Part V. Examples</strong></h3><h4>The trenches of the Western Front, 1914–1916</h4><p>The most convincing historical evidence. Working from diaries and reports, the historian Tony Ashworth documented a system the soldiers themselves called “live and let live”.</p><p>On static stretches of the front, units spontaneously stopped actually killing each other. Artillery struck the same spot at the same time every day — so that the other side had time to take cover. Snipers demonstratively fired high. During food deliveries both sides kept quiet.</p><p>Note the conditions: these men had direct orders to kill, hated each other courtesy of the newspapers, shared no language and had no way to negotiate. Cooperation arose anyway — because the same companies sat opposite the same companies for months. The shadow of the future was enormous.</p><p>The command understood the problem and suppressed it with two measures: rotating units, and compulsory night raids with a report on the result. Rotation made the encounters one-off. The raids made quiet non-compliance impossible. In other words, headquarters deliberately shortened the shadow of the future — and cooperation disappeared, exactly as the model predicts.</p><h4>Vampire bats</h4><p>Gerald Wilkinson (Nature, 1984): a vampire bat dies after roughly 60 hours without blood, and unsuccessful nights happen regularly. Well-fed individuals regurgitate part of their catch for the hungry — and do so mainly for those who shared with them earlier.</p><p>All three necessary ingredients are present: repetition (a shared roost for years), recognition (individual memory for particular animals), and asymmetry of value (for the well-fed bat the donated portion is cheap, for the hungry one it is a matter of life). This is cooperation without intelligence and without morality — pure mechanics.</p><h4>Cartels and OPEC</h4><p>The textbook case. It is in every country’s interest to hold to the quotas and keep the price high (R). It is in each one’s individual interest to quietly exceed its quota while the others hold back (T). When everyone starts doing it, the price collapses and everyone loses (P). The history of OPEC consists of cycles: agreement → quiet cheating → collapse into a price war → painful return to quotas.</p><h4>The arms race</h4><p>The USA and the USSR armed themselves into the state (1/1) instead of mutual disarmament (3/3) — because neither could risk being the only disarmed party (0/5). The SALT and START treaties are, structurally, not a display of goodwill but a verification technology: a way to make betrayal visible, and therefore punishable.</p><h3>Business</h3><p>Marketplace reviews, referrals, industry reputation, public supplier ratings, LinkedIn — all of it is essentially infrastructure for turning one-off deals into repeated ones. They artificially raise w between strangers who will never meet again.</p><h3>Part VI. Conclusions for a team</h3><p>Axelrod’s main result is worth restating bluntly, because it is counter-intuitive:</p><blockquote><em>Cooperation requires no altruism, no morality, no trust and no boss watching over everything. It arises between wholly selfish participants — if the structure of the interaction is right.</em></blockquote><p>The structure rests on three pillars:</p><p><strong>Repetition</strong> — “Will we meet again?” Without it, betrayal goes unpunished.<br><strong>Recognizability</strong> — “Do I remember who behaved how?” Without it, there is no one to punish or reward.<br><strong>Response</strong> — “Does betrayal have a price?” Without it, niceness turns into feed.</p><h4>What follows in practice</h4><p><strong>Lengthen the shadow of the future.</strong> This is the strongest lever. Break a big one-off deal into a series of smaller stages. Make interaction regular rather than project-based. Every meeting that guarantees a next one automatically makes honesty profitable — with no exhortation at all.</p><p><strong>Make contributions visible.</strong> If it is impossible to tell who helped and who rode free, the mechanism does not work at all. This is not about control — it is a necessary condition for reciprocity to exist.</p><p><strong>Be nice by default, but not defenceless.</strong> The first move is cooperation. It costs almost nothing and starts the whole cycle. But systematic exploitation of that niceness must have consequences, otherwise it simply subsidizes parasites.</p><p><strong>Forgive quickly.</strong> The costliest mistake in the model is not being cheated once, but getting stuck in a spiral of mutual revenge. One betrayal ≠ one conflict forever. Especially where misperception is possible — and between people it always is.</p><p><strong>Do not measure success by the gap with a colleague.</strong> That is literally the strategy that lost the tournament. The department that beat the neighbouring department usually ends up in cell (1/1) together with it.</p><p><strong>Beware of “final round” situations.</strong> Someone on their way out, the last month of a contract, a one-off contractor you will never work with again — that is precisely where the model predicts a collapse of cooperation. It is not about bad character; it is about structure.</p><p>And the mirror-image conclusion: if you need to break up someone else’s undesirable collusion — do the opposite. Shorten the horizon, destroy transparency, make it harder to identify the offender. That is exactly how antitrust authorities break cartels: a leniency programme for whoever informs first turns a stable (3/3) into an unstable race to (5/0).</p><h3>Sources</h3><p>Axelrod R., Hamilton W. D. The Evolution of Cooperation. Science, 211 (1981), 1390–1396.<br>Axelrod R. The Evolution of Cooperation. Basic Books, 1984.<br>Axelrod R. The Complexity of Cooperation. Princeton University Press, 1997.<br>Flood M. M. Some Experimental Games. RAND Research Memorandum RM-789, 1952.<br>Friedman J. A Non-cooperative Equilibrium for Supergames. Review of Economic Studies, 38 (1971).<br>Kreps D., Milgrom P., Roberts J., Wilson R. Rational Cooperation in the Finitely Repeated Prisoners’ Dilemma. Journal of Economic Theory, 27 (1982).<br>Nowak M., Sigmund K. A Strategy of Win-Stay, Lose-Shift that Outperforms Tit-for-Tat. Nature, 364 (1993), 56–58.<br>Nowak M. Five Rules for the Evolution of Cooperation. Science, 314 (2006), 1560–1563.<br>Press W. H., Dyson F. J. Iterated Prisoner’s Dilemma Contains Strategies that Dominate any Evolutionary Opponent. PNAS, 109 (2012).<br>Stewart A. J., Plotkin J. B. Extortion and Cooperation in the Prisoner’s Dilemma. PNAS, 109 (2012).<br>Ashworth T. Trench Warfare 1914–1918: The Live and Let Live System. Macmillan, 1980.<br>Wilkinson G. S. Reciprocal Food Sharing in the Vampire Bat. Nature, 308 (1984), 181–184.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2d685074fd87" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>