Introduction to Strategic Science bX) pert) FT 53 Tatiana BabichevaWhat is this book about? ‘This is an introductory course on game theory and operations re- search, This course is intended for anyone interested in game theory. It does not require any prior mathematical or economic knowledge. What is the subject? What is operations research? No! It's not about an operation where a person is incised and their anatomy is examined, Operations research is the application of mathematical and quan- titative methods to support decision-making in all areas of human activity. It begins when one or the other of the mathematical appa- ratuses is used to justify decisions. What is a game, in principle, everyone understands and has experi ‘enced more than once, But how are operations research and games related? And what do mathematics have to do with it2! Let's start with the main question: what is game theory exactly? ‘The game begins whenever people try in one way or another to in- teract with each other. Romeo and Juliet played the «let's get married» game, but due to incomplete information and other factors, both lost. When leaving home to go to work, each road user plays a game of interaction with many other participants, trying to minimize their travel time. This is also what we will mention when introducing thevi Waar 15 THIS BOOK ABOUT? concept of equilibrium in road networks. Auctioning is a very explicit type of game, like any transaction and pricing in general. For example, if I decide to open a new canteen at MIT, I will have to play a big game with the managers of other canteens, a crowd of students, and other protagonists. Even when looking for a job, you will have to play the game «lf I ‘want the highest possible salary, how much should I ask to be hired and get that salary?» Asyou may have already understood, game theory is one of the most, important subjects. Naturally, game theorists do not claim to be able to provide answers toll the world’s problems, ‘The best they can understand is what happens when people interact rationally. It’s fortunate that many people don’t always behave irrationally, al- though not everyone can be easily understood. Most of us generally try to spend our money wisely, while avoiding doing strange or un- predictable things; otherwise, economic theory could not exist Even if someone acts without thinking, it does not necessarily mean that they are acting irrationally. Game theory has had some suc- cess in explaining the behavior of insects and plants, even though. their ability to think is probably questionable. The explanation may simply come from the fact that these generations of insects and plants have evolved and adapted when those of them with inade- ‘quate genes disappeared. So perhaps we are going to study simple psychology? Maybe the subject should be called «methods of rational interaction» ratherthan «game theory»? In fact, we will consider a rather specific case: that of rational play- crs in «simplified» game conditions with known rules. The mathe- matics we will use will be quite simple, although some of the eco- nomic concepts we will address are only studied in university pro- grams. Most of the games proposed in this book have been experimented with: the players were both MIPT students and researchers from RATP Smart Systems and other types of audiences. To conclude the introduction of this book, we would like to thank the wonderful illustrator Natalia Kilianova for the fantastic work she has done. GOLContents Dedication What is this book about? Introduction. Opening the door to game theory. 1 Games with Nature, or What is expected value? 1.1. What is chance? 1.2. What is Probability? 13. Games with Nature 14 Mathematical Expectation 15 Birthday Paradox 1.6 Can you win at the casino? 1.7 Decision theory 1.8 Interaction within a Group 1.9 Fair Cake Cutting 1.10 Secretary Problem 2. Mathematics of Decision 21 Ultimatum 22. Saint Petersburg paradox 23. Cognitive bias 24 The two envelopes paradox 25 Monty Hall Problem 2.6 And what about the pigeons? 2.7. The (three) prisoners’ paradox . . 28. Surprise test paradox 29. Paradox of Parrondo 3. Prisoner’s dilemma 3.1 Prisoner's dilemma cone 3.2 Prisoner's dilemma in a group. Game of competing, companies 3.3 Prisoner's dilemma in politics 25 25 27 29 32 34 38 40 42 45 49 49 54 56Contents: ix 3.4 Public and private goods 58 35. Free rider problem 60 3.6 Game with contribution to the common fund 63 3.7 Some classification and terms 65 3.8 Nash Equilibrium 67 3.9 Pareto Optimality . . . - : 69 3.10 ‘The water fountain and Pareto optimum ...... 71 3.11. Penalties for deviating from strategies B 3.12 Sen's Paradox... « : sere BB 4 Repeated games 7 4.1. Stag hunt 7 4.2. Minimax principle . . . sere 43 The mad hatters : seen eee BB 44 Repeated games 85 45° Mixed strategies . . . : serene 86 4.6 A repeated prisoner's dilemma . eee 8B 4.7 Where is the coin? 90 5. Psychology of the game 93 5.1 Solution of a matrix game in mixed strategies 93 5.2 Rock-Paper-Scissors 94 5.3 One step further? . . : eee 54 Lizard? Spock! 9 5.5 Solution of the game in mixed strategies 101 5.6 Deer and Bison. . . . : cee es 108 5.7 Colonel Blotto and the Chinese sticks 105 5.8 Sequential games 108 5.9 Restricting choice? . : seve M2 5.10 Dollar auction 116 6 Sequential games. Games with imperfect or incom- plete information. 19 6.1 Winning and losing positions. eee M9 6.2 Subgame perfect equilibrium 122 6.3 The battle between duty and desire 12464 The baitle of the professor and negligent student 6.5 Games with imperfect information 66 More gold needed! 6.7 Games with incomplete information 68 Yield the way! 6.9 ‘The market for lemons 6.10 Weekend! Road network games 71 Directed graphs 7.2. Weighted graphs 73. Variable weighting 74 Playing traffic jams! . . wees 7.5 Anexample of seeking equilibrium in a transporta- tion network . . fetes 7.6 _Isit always profitable to build new roads? 7.7 Braess paradoxs 7.8 Other paradoxes and ways to resolve them There are kinds of lies: lies, damned lies, and statis- tics. 8.1 Decision Theory 8.2 What is utility function? 83. The utility of the movements 84 Time is money 8.5 How time can be unlinear? 8.6 And nothing else matters? 8.7. The struggle with statistics 8.8 Understanding type I and type Il errors 8.9 To treat or not to treat? 8.10. Cheshire cat's smile 8.11 ROC analysis 8.12, What is correlation? . . 8.13 Correlation vs casuality 8.14 Hunting for the casuality 129 132 133 138 139 342 44 47 447 uy. 150 153 195 156 158 160 163 163 164 167 168 170 172 174 176 178 179 181 185 188 191Introduction. Opening the door to game theory. Let's start this book by describing the role of game theory in modern, mathematical and economic sciences. If we turn to a formal defini- tion, then game theory is a branch of applied mathematics, using which the behavior of multiple subjects is modeled, where each sub- Jject’s decision criterion depends on the decisions made by others. In other textbooks and monographs, game theory is understood as the mathematical theory of decision-making in conflict situations. In fact, these two definitions are equivalent because any situation of conflicting interests is a conflict in the broad sense. Historically, game theory dates back to the 18th century, with the beginning of the Enlightenment and the development of economic theory. ‘The earliest scientific economic models that later became game theory were envisioned by Antoine Augustin Cournot (1801— 1877) and Joseph Louis Francois Bertrand (1822-1900), French re~ searchers, in the 19th century. Subsequently, they became classical models of production and pricing in an oligopoly named after their creators (Cournot and Bertrand oligopoly models). In the early 20th century, Emanuel Lasker (who was the world chess ‘champion for the longest time, for 27 years), Ernst Zermelo (German mathematician), and Emile Borel (researcher, also recognized as a French politician, deputy, and minister) advanced the idea of math- ‘ematical theory of conflicts of interest. “The formal mathematical aspects and applications of game theory were first presented in the 1944 book by John von Neumann and Oskar Morgenstern, «Game ‘Theory and Economic Behavior» [morgenstern1953theory]. At that time, the mainstream of scientific research was around antago- nistic games: games where one player's gain was equal to the other's loss, At the same time, until the mid-20th century, game theory wasconsidered only as a mathematical theory, despite the obvious pos- sibilities ofits application to economics. However, after World War Il, in the United States (notably due to increased science funding), attempts were made to apply game the- ory to practice in economics, biology, cybernetics, technology, and anthropology. ‘The military themselves, during and immediately after the war, se- riously looked into game theory, seeing it as a powerful tool for strategic decision-making. In the early 1950s, John Nash formulated the concept of Nash equilibrium in non-antagonistic games, which later became the key to the entire game theory. According to this concept, the parties involved in the conflict must use the optimal strategy, which leads to the creation of a stable equilibrium. It is ad~ vantageous for players to maintain this equilibrium, as any change will worsen their situation, Nash's work thus made a significant, contribution to the development of game theory, and the mathemat- ical tools of economic modeling were revised. In general, modern game theory heavily relies on approaches and results that emerged in Nash's works. In recent years, this scientific direction has greatly expanded. Some areas of modern economic theory cannot be for- mulated without the use of game theory. Furthermore, a significant portion of the Nobel Prizes in economies in recent years has been awarded to works where special attention has been paid to game models, Among them, the following prizes can be mentioned. In 2005: Robert Aumann and Thomas Schelling, «for having enhanced our understanding of conflict and cooperation through game-theory analysis.» In 2007: Leonid Hurwiez, Eric Maskin, and Roger Myer- son, «for having laid the foundations of mechanism design theory.» In 2012: Lloyd Shapley and Alvin Roth, «for the theory of stable allocations and the practice of market design.» In 2020: Robert Wil- son and Paul Milgrom, «for improvements to auction theory andinventions of new auction formats.» ‘The construction and study of decision-making models are tradition- ally referred to as «operations research problems». According to the ‘elegant definition proposed by one of the founders of the Russian school of operations research, academician Pavel Krasnoshchekov, «the theoretical aspect of operations research concerns the construc- tion and study of mathematical models for making optimal deci- sions». Obviously, any operation, which is a set of actions aimed at achieving a goal, is unthinkable without defining that very goal. ‘The goal can be formulated, in general, as desired, depending on the domain in which the decision is made. For example, in military af- fairs, the objective is to complete the assigned combat mission (in fact, of a binary value: either the detachment has completed the task or not), and in economies, the objective, generally, is to max- imize a certain quantitative indicator of the decision-maker’s suc- cess. Most often, in mathematical models of economics, profit (if the decision-maker is a company) or utility (if the decision is made by the consumer) is maximized. The outcome of the operation is in- fluenced both by controllable factors, whose values are determined by the player's actions, and by uncontrollable factors. The rules that the player takes into account to determine the values of the control- lable factors form their strategy. As for the uncontrollable factors, they include everything the player cannot influence. Depending on the quality of information and the decision-maker's ability to pre- dict the value of uncontrollable factors, among the latter, we can distinguish random factors (for which at least the distribution law and stochastic behavior are known) and uncertain factors, whose definition domain is known only. In game theory, the main subject of analysis and decision-making models are more complex: in these models, there are several decision-makers (or players) ranging from two to infinity. It is as- sumed that their interests do not coincide: that is, the objectives of these individuals are different. This is the main essence of theconflict situation: the decision is not made by an individual, but by several, and the payoff function of each of them depends not only on their strategy but also on the decisions of the other participants. ‘The mathematical model of this type of conflict is called a game, and, the participants in the conflict are called players. ‘Traditionally, mathematical game theory is divided into two main areas, + Cooperative game theory studies decision-making under the assumption that there is a mechanism to ensure the implemen- tation of a common decision. At the same time, the main task in this branch of game theory is to indicate a variety of mutu- ally beneficial solutions, taking into account the interests and independent capabilities of individual actors and their coali- tions, Le., groups of actors acting jointly. If this set includes multiple solutions, then there is also the problem of develop- ing an optimality criterion that would enable finding the only, and, in some sense, the best solution. + In turn, non-cooperative games reflect situations in which players act autonomously, independently of each other, and if some agreements are concluded, they are not binding: each player can deviate from the agreement. In addition to the division into cooperative and non-cooperative game models, there are also many other classifications of game the- ory models. For example, the difference between static and dynamic ‘games comes from the players’ ability to observe and react to each other's actions. Thus, in static games, players make decisions simul- taneously, and the decisions made are not subject to revision, In ‘dynamic games, there is a more complex order of moves. For exam- ple, in chess, moves are alternating, and in some board games, the order of moves depends on the players’ reactions.In the chapters of this book, we will discuss the basics of the most classic games and adopt a general view of game theory.Games with Nature, or What is expected value? 1.1 What is chance? We are all familiar (or we think we are) with the concept of «chance», oreven «random», What idea do you have about the meaning of this term? ‘The most common answer to this question is as follows: «chance oc- ‘curs when unexpected things happen. What are these unexpected things? I think you understand yourself that such a definition does not really make sense. Here is Aristotle's definition of the term chance: «when this accidental character occurs in the facts that are produced for a purpose, then we speak of effects of fortune and chance», but he asserts «... there isa determined cause of everything which we say happens by chance or fortune». Let's introduce a more formal definition. «Chance is a factor that determines the outcome of an experiment among many possible re- sults known in advance.» But can we speak of randomness if we do not know in advance the many possible results? For example, you take a test and are given problems to solve. Are they random for you? Are they random for your teacher? Think about it ‘The concept of randomness can be categorized into two different types:2__Cuaprer 1 Games wer Nature, on WHAT IS EXPECTED VALUE? Definition 1. Ontological randomness — it is part of being. For ‘example, tossing a coin can be classified into this category. Definition 2. Epistemic randomness — it is randomness that re- sults from ignorance or the inability to understand certain processes, but in fact everything is completely predetermined here. For example, you come to school and find out that you have an un- ‘expected quiz today. It may seem like a random event, but it was ae- tually planned long ago by the school board. You just didn’t know. ‘The expression «accidents are not accidental» refers to this type of epistemic randomness. To this day, not to mention an earlier period, many people still believe that ontological randomness does not exist and that all randomness is only epistemic in nature. For most Enlightenment scholars and until the early 20th century, everything was prede- termined. This trend is called determinism. Today, we simply don’t know exactly how things work. Classical physics is built on this deterministic principle, but not quantum physics, for which ev- erything is much more complicated. Philosophers, like physicists, are still not fully decided. For example, Paul Thiry d’Holbach, a German-born French materialist scientist and philosopher (1723- 1789), wrote [holbach1793systeme]: «In a whirlwind of dust raised by an impetuous wind, however confused it may appear to our eyes, in the most dreadful storm stirred up by opposing winds that lift the ‘waves, there is not a single molecule of dust or water placed at random, ‘which does not have its sufficient cause to occupy the place where it is {found and which does not act rigorously in the way it must act. A _geometer who would know exactly the different forces acting in these ceases and the properties of the molecules that are moved, would demon- strate that, according to given causes, each molecule acts precisely as it must act and cannot act otherwise than it does.» D'Holbach as- sociates determinations of different natures, notably a mechanical1.2. Waar Is PRonaniiry? 3 model based on the universal communication of movement and a ‘chemical model based on affinities, which encompasses human pas- sions and desires. He asserts the fundamental principle of general necessary causality: every effect has a natural cause. He thus in- sists on «simple and general laws» because they offer a constant, reassuring, and sufficient reference against the supernatural. ‘This opinion is particularly present among dualists, who affirm that in addition to matter, there exist elements of metaphysics, including the divine principle, Obviously, ontological randomness is abso- lutely incompatible with the acceptance of something omnipotent and omniscient, ‘The importance of philosophy in our subject, we hope, will not es- cape you. Yet adding philosophy to our course, in addition to math- ‘ematics, economics, and psychology, would be cruel on our part. 1.2 What is Probability? ‘No one can predict how a coin will land when tossed: heads, tails, or ‘even edge. That's why flipping a coin is so often used to determine what to do in a controversial situation: for example, whether to go to the first class or stay in bed. ‘The first use of this game in this form dates back to the creation of metallic currency. However, other forms existed before that used objects with two distinet sides, like a seashell, for instance. The choice was then left to chance. Even today, playing : from two tosses, on average, one side will be heads and the other will be tails. So, on average, out of 1000/2 = 500 tosses, 500 heads will be flipped. So, what is probability? Usually, within the framework of school programs, the following definition is given: Definition 3. When we can determine all possible and equally likely outcomes, the probability of an event is given by the quo- tient of the number of favorable outcomes by the number of possible outcomes. What is a favorable case? The case for which we are seeking the probability. For example, getting heads in a coin toss. The number of possible cases corresponds to all possible outcomes, here, «heads» and «tails», so it is equal to 2. Why did the term «equally likely» appear in the definition of prob- ability? Shouldn’t we just divide the favorable cases by the possible1.3 Games worn NATURE 5 ‘This definition above can be illustrated by the following anecdote: {\ ike Someone asks a passerby: What isthe proba- bility that when you step out onto the street, ‘you'll encounter a dinosaur? ‘The passerby answers: 50 percent. = But... How is that possible? ‘The passerby: Well, either Ill encounter one or Lwon't. A Here, this average passerby (not really intelligent) has just divided 4 favorable outcome of «encountering a dinosaur» by two possible outcomes. Well, you've understood that it doesn't work that way — it’s quite obvious that this probability can’t be that high. Moreover, if in the heads or tails» game the coin is rigged, it will no longer be per- fectly symmetrical, and the probabilities of landing on heads or tails, ‘can then become unequal. This is the method used by cheats, for example. 1.3 Games with Nature ‘One ofthe main reasons for the emergence and development of prob- ability theory, and even its popularity, comes from the possibility of making a lot of money in one go and without effort. For example, in the lottery or at roulette. Finding patterns to «beat the system» then becomes a powerful motivator to develop the corresponding mathematical apparatus.6 Cuaprer 1 Games wera Nature, on WHAT IS EXPECTED VALUE? It seems that if we know the laws of probability and the rules govern- ing chance, then we can win any game. These attempts, therefore the techniques aimed at securing winnings in games of chance, are called «martingales». In reality, this is a pipe dream, and we will show it, ‘The main thing we need to understand is that a game is random if the player cannot influence its outcome, Thus, chess is not random, bridge is not en- tirely random, while tossing a coin, roulette, and even Russian roulette are. We will call these games in which chance plays a significant role, although subject to certain mathemat- ical dependencies, «games with na- ture», that is, with randomness. For example, you can play with only na- ture by tossing a coin. You can also play with someone and with nature — this is the case, for example, with «Bridge». In this scenario, on one hand, nature has dealt you the cards (or someone may cheat, but we believe in the kindness and honesty of people, and generally, we have our own deck). As for yt! the actions of the second player, they are not random. ‘There are certain games where the player's action consists solely of buying a ticket, and after that, they have no further involvement. ‘This is the case, for example, with a simple lottery. Roulette is an ‘example of another class of games where the player has the oppor- tunity to choose the bet and the type of game. From a mathematical point of view, the game of roulette is not fair because in all cases, the14 MATHEMATICAL EXPECTATION 7 casino wins. How do we determine if a game is fair? This requires in- troducing the concept of mathematical expectation, first formulated in 1670 by the Dutch mathematician Jean de Witt. He published the first modern treatise on evaluating life annuities by mathematical ‘expectation (the present value of future payments). 1.4 Mathematical Expectation What is it? Let's imagine we're playing a game. It doesn’t matter to us at this point whether it’s a game with nature or with another opponent. Consider a dice game with two dice. We pay €10 for the opportunity to roll these dice. If the sum of the two rolled dice is 7, we win €50, If any other sum comes up, we win nothing, Is this game profitable? Should you participate? To find out, you need to calculate the probability of getting exactly 7 points on a roll of two dice. There are exactly 36 equally proba- ble outcomes in total (we believe that in this game, the organizers aren’t such obvious cheaters as to offer rigged dice). What are these outcomes? 1+ 1,1 +2,1+3, ete. Among them, exactly 6 (1 +6, 2+5,8+4,4+3,5+2,6 + 1) are favorable. In other words, the probability of winning is pun ‘The probability of losing is pu = 1-4-5 ‘The result of the roll is a random variable. Definition 4. ‘The possible values of a random variable could rep- resent the possible outcomes of an experiment, the existing value of which is uncertain, In our case, we know the set of possible roll values for each die: they8__Cuaprer 1 Games were Nature, on WHAT IS EXPECTED VALUE? are numbers from 1 to 6. We don't know which number will come up on the next roll; its ontological randomness (if the game organiz~ cers are cheaters, then for them it would be epistemic randomness). We're interested in the average winnings per game. Let's introduce the definition of the average gain in a more formal language. Definition 5. Mathematical Expectation is the arithmetic mean of the possible values a random variable can take, weighted by the probability of those outcomes. In other words, if the variable X takes values :r1,.02,... tn with probabilities p1,p2,..-, Pn, the ex- tion of X is defined as: E|X] = yp, + e2p2 +--+ tnPas Corollary 1. Since the sum of the probabilities is equal to 1, the expectation can be considered as the average of the zr; weighted by the BX] = Sti In the considered game, if we succeed, we win €50, so the amount of the gain is €40 (don't forget we already spent €10!), and if we fail, ‘we lose €0, so the amount of the gain is -€10. Therefore, E = Pwin’ Swin + Ploss * Sloss ‘The mathematical expectation of a gain in this game is negative, meaning the game is not profitable for us. The more we play, the larger the mathematical expectation of gain (in absolute value) will be, and consequently, the more we will lose.1.5 BIRTHDAY PARADOX 9 Christian Huygens, in his work «On the Calculation in Games of Chance» from 1657, was interested in the amount to bet for the «game to be fair. He established that if in a game we have p chances, of winning the amount a and q chances of winning the amount b, ‘we should bet the amount $ = “25%! for the game to be fair. In other words, Definition 6. If the mathematical expectation of gain for a game is ‘equal to zero, the game is considered to be fair. ‘The concept of mathematical expectation is quite basic and is used not only in game theory but in many other fields as well, which we will mention later. 1.5 Birthday Paradox How can we tell if a game is fair? Sometimes, our intuition fails. For example, I propose a game to all of you who want to play in your study {group orat the office: Ifamong you there are at least two people who have the same birthday, then they will buy you a pizza; otherwise, you buy them pizza. Is, this bet fair or not for you? Let’s check. Let’s calculate some probabilities. We'll derive a formula showing the probability that there is atleast one pair with the same birthdays in a group of n people.10__Cuapren 1 Games wri NaTuRe, on WHAT Is EXPECTED VALUE? Suppose there are 366 different birthdays, and they are all equally likely (infact, those born on February 29 should be fewer on average than those born on February 28, but this is not the most important ‘case here). The simplest way to get the required result isto calculate the probability that each person has a birthday different from the others: the opposite of what we are looking for. We can proceed by recurrence: the first person has 366 choices, the second has 365, the third has 364, the fourth has 363, and so on. Here we will proceed by counting, meaning we will count the num- ber of cases where n people have different birthdays and divide it by the total number of possibilities. In both cases, we assume equiprob- ability of birth dates. So, for two people, the probability of no match fi sig: When the third person arrives, the desired probability of no match becomes 38° x 34, since there were 364 «available» days. By reasoning in this way, we can derive a formula that determines the probability of non-coincidence of birthdays for one of the pairs in a {group of n people: 365 364 366~n 366 "366 366 Pr Avery simple action remains to be done. You just need to calculate this expression until the product of the fractions becomes less than 3. It’seasier to do on a computer. It turns out that already at n = 23, the fraction is less than 4, so if there are 23 people in your group, then the game is already profitable for you 1.6 Can you win at the casino? Let's look at another example, is time related to another popular ‘game with nature: French roulette.16 CAN YOU WIN AT THE CASINO? u In the modern world, almost everyone uses the Internet, and many of us have probably received emails or intrusive ads offering to dis- cover the secret to winning at roulette. ‘The game of French roulette involves spinning a small ball on a wheel with 37 slots. These are numbered from 0 to 36 (the numbers are part of the wheel) al- ternately red and black, except for zero, which is green. A player bets an amount M on one of the slots. If the ball lands on their chosen slot, they are paid 36 times their bet (their gain is then 35.M = 36M —M); otherwise, they lose their bet (their loss is then —M = 0 ~ M), ‘There are also other betting options in the game, for example, betting on red. So, when the ball lands on a red-numbered slot, the bet is doubled. ‘The probability of landing on a red slot is then #8, on a black slot is also $3, and on zero is 4. ‘The expected value ofa gain with a bet of one euro is then: is 19 g=84 at! +37 As we can see, on average, we will win @ negative amount, which means this game is unfair. But maybe there are other options? You can bet on a specific slot, and you will be paid 36 times your bet. Let's calculate the expected gain:12__Cuaprer 1 Games wr NaruRe, on WHAT Is EXPECTED VALUE? ‘The casino cannot be tricked; the rules are designed in such a way that the expected value of a gain for each possible variant of roulette games is always exactly — 31 of the bet. American roulette differs from French or European roulette in its wheel layout, which has an additional number: double zero. Conse- ‘quently, the rules there are even stricter. Let's try to trick the casino using the winning strategy offered on the Internet. To simplify, let’s assume the roulette game is fair, meaning the mathematical expectation of a win is zero. My first bet is 1 euro on red. If Lwin, I put the winnings in a bank account and stop playing. If [lose, I double my bet and bet again on red. I believe it doesn’t matter how many times the series lands on red. With this strategy, by constantly doubling bets after each loss and taking money when I win, 'l become invincible, Am I right? Let's see what the probability is of getting black 20 times in a row. Ie will be equal to gly = 10%, or about one millionth, How much money do I need to bet again in this case? It’s 2%", which is about one million euros. Do I have that much money on me? And if so. ‘what about when black lands 21 times in a row? That's roughly two million? And 30 times in a row? It's unlikely, but the loss in that case is astronomical! Also, for this strategy, there's another problem: in real casinos, the maximum bet size is limited. Let’s say you can’t bet more than one million euros, We just watched in horror as the ball landed on black 19 times in a row. To recover the loss (well, it can't possibly land on black for the 20th time, we believe), we must bet 1,048,576 euros, except the rules limit our bet to 1,000,000 euros.1.7 DECISION THEORY 13 Even if I get lucky (and players believe they surely will), then I will be paid 1,000,000 euros, and I've already bet 1,048,576 euros. My net loss was 48,576 euros, and to recover it, [must successfully play 48,576 times. Now remember, the probability of winning is not } but Calculations show that even with a fair game (which is impossible), during 20 consecutive and consistent years of playing, the probabil- ity of winning is 99%, Then you need to have at least 2" = 262,144 ‘euros pocket money to win one euro every day. Isnt it easier to put money in the bank at a minimal interest rate? Or even to find a job? Attempting to succeed in such a game is, af- \ ter all, quite anecdotal: Visitor to the bartenders How much does a drop of cognac cost? Bartender: Nothing. Visitor: Then, pour me a glass of drops. ‘The law of large numbers is at play here. And we conclude that in a fair game (and even more so in an unfair game), no strat- egy can lead to a guaranteed win, 1.7 Decision theory We started our book with mathematics because playing with nature is essentially an estimation of the expected gain based on the prob- abilities of each outcome. Each action by the player in game theory is called their strategy. In the games we've already examined, the player's strategies for each14_Cuapren 1 Games wri Nature, on WHAT IS EXPECTED VALUE? round were to «play» or «not play». Attempting to choose the right strategy is the simplest case of deci- sion theory. There's a well-known paradox called the Buridan’s Ass Paradox (Latin: Asinus Buridani inter duo prata) ‘The Buridan’s Ass paradox originates from the legend that an ass died of hunger between two piles of hay (or between a bucket of oats and a bucket of water) because it couldn’t make a choice. Strictly speaking, this isn’t a logical paradox but rather an extreme ‘example of a dilemma taken to the ab- surd ‘The Buridan’s Ass paradox doesn’t appear in any known works of Jean Buridan, although it aligns well with Buridan’s theory of free~ dom and animals. Instead, this theme appears in Aristotle's «On the Heavens», where he ponders how a man would proceed when faced with a choice between water and food, So, how can an ass receiving two equally tempting treats make a rational choice? Buridan addressed this by defending the position of moral determinism, arguing that when faced with a choice, one should lean towards the greater good. Buridan admitted that the choice might be complicated by evaluating the outcomes of each option. Later thinkers exaggerated this viewpoint, citing the example of an. ass and two equally accessible and appetizing piles of hay, arguing that the ass would surely starve to death making a decision, ‘This1.7 DECISION THEORY 15 version became widely known through Gottfried Wilhelm Leibniz, (1646-1716), a German philosopher and scientist. He was also a member of the French Academy of Sciences. Within the logic of the problem itself, one can show that a rational ass would never starve to death, although we can’t say which pile of hay it will choose. If we consider refusing to eat as a choice, and the ‘worst outcome among three options ~ a pile to the lef, a pile to the right, and starvation ~ the third option will be dominated by other strategies (we'll discuss this later), so the ass will never choose that option. In some interpretations, a potential change in the situational context is also considered: combining the two piles into one, after which the dichotomy disappears. Although there are different types of di chotomies, this is a mind-body dichotomy, essentially a philosophi- cal dichotomy. Recall that the most well-known dichotomy paradox was formulated by Zeno of Elea in ancient times: ‘ ity then takes on a finite and relatively low value, making it rational to make a limited bet, So, for players, there is a kind of inflation, or ‘even hyperinflation, of utility 2.3 Cognitive bias ‘The two games discussed are related to the psychology of the player. Let’s discuss an important characteristic of personality psychology: the tendency towards cognitive bias. A cognitive bias is a distor- tion in the cognitive processing of information. The term «bias» refers to a systematic deviation from logical and rational thinking compared to reality. Most cognitive biases have been described by scientists and many have been proven in psychological experiments. Some of these biases can actually be effective in natural environ- ments, such as those that housed human evolution, allowing for more efficient evaluation or action; while they prove to be maladap- tive ina modern artificial setting, Others seem to stem froma lack of appropriate reflection abilities or from the misuse of adaptive skills, in other contexts. We'll describe here some cognitive biases that, in my opinion, are the ‘most common in decision-making and during economic and mathe-30 Cuarrer 2 MatHemarics oF DECISION matical games. Exaggeration of the probability of specific cases. For example, generalizing from specific cases is an unreasonable transfer of char- acteristics from isolated or specific cases to broader sets. There are many types of this cognitive bias, the most classic being conspiracy theories. In games, it can be the belief, «I got lucky once at roulette today, so I'll get lucky again», The Baader-Meinhof phenomenon or frequency bias means that newly learned information that reappears after a short period is perceived as unusually frequent, Have you ever, just after starting to read this book, heard of something simi- lar around you? Reevaluation of the importance of specific cases. Hindsight bias or the «I knew it all along» effect is a tendency to judge after the fact that an event was predictable. It’s the tendency to judge de- cisions based on their final outcomes rather than judging the quality of decisions based on the circumstances at the time they were made (winners aren’t judged»). Clustering illusion — exaggerating the regularity of randomness. Reevaluation of your abilities. Egocentric bias involves a person, during joint action or group work, overestimating their contribu- tion and taking more responsibility than an external observer would have. The illusion of control is people's tendency to overestimate their ability to control events; for instance, this occurs when some- one feels in control of outcomes they clearly have no influence over («Come on, ball, land on red!»). Zero-risk bias is a preference for a controlled but potentially more harmful situation (due to its higher frequency) over the opposite due to an overestimation of control pos- sibility. In other words, a person believes they completely eliminate the risk (when, in fact, they don’t have full control), while statisti- cally, it’s about reducing just one risk to zero from the larger ones. For example, many people fear the complications of medical pro- cedures more than the disease and death resulting from untreated23 CoGNrtive Bias 31 illnesses. Dunning-Kruger effect means that those less competent in afield overestimate their abilities, while the most competent tend to underestimate theirs. Reevaluation of Importance of one’s own opinion, position, and/or choice, Choice-supportive bias is the tendency to retroac- tively attribute positive attributes to a selected option and its further justification. Attention bias refers to how certain information is pro- ‘cessed differently by the brain based on an individual's concerns or interests. For example: «l love red, so red will come up mote often!» Escalation of commitment is a tendency to remember your choice as being more correct than it actually was. Blind spot bias isa bias re- «garding the recognition of the impact of biases on others’ judgments while omitting to see the impact of biases on one’s own judgment. Methods to protect one’s opinion from such distortions generally include: Reduction of cognitive dissonance: you try to reinterpret a situation to eliminate contradictions, Texas sharpshooter fallacy is the selection or adjustment of a hypothesis based on the measurement results. This fallacy gets its name from a joke where a Texan marks- man draws a target around the point where his bullet hit. Confirmation bias relates to the tendency to validate one’s opinions with confirming instances and outright dis- miss instances that refute them. In what is called the expectancy observer effec, the experimenter can subtly communicate their expectations for the study's outcome to participants, leading them to alter their behavior to conform to those expectations.32 Charter 2 MatHeMarics oF DECISION Selective perception is the tendency to selectively interpret what one observes based on our interests, social situation, experience, and at- titudes. The backfire effect leads individuals presented with clear and logical evidence that contradicts or invalidates their beliefs to reject it and feel strengthened in their initial belief. Survivorship bias involves focusing on elements that have successfully passed a selection process to draw conclusions about all elements. ‘And, of course, the most common and frustrating cognitive distor- tion for many is called the «Curse of Knowledge». It's a cognitive bias that occurs when a person, communicating with others, uncon- sciously assumes that others have the same knowledge base to un- derstand. You're probably faced with this almost every day, from both sides, 2.4 The two envelopes paradox ‘The Two Envelopes Paradox is a classic problem in decision theory that leads to a seemingly contradictory result, ‘There are several versions of the para- dox. Most commonly, it presents the following decision situation: Two en- velopes each contain a check. You know that one check contains twice the amount of the other, but you have no information on how the amounts were determined. A host offers a par- ticipant the choice of one envelope, and the amount in the chosen enve- lope becomes theirs.24 THE TWO ENVELOPES PARADOX 33 ‘The paradox itself lies in the argument that follows: before the par- ticipant opens the chosen envelope, the host advises them to switch with the following reasoning. Let V be the value of the check in the chosen envelope. There are two possible scenarios: + There's a one in two chance that the other envelope contains a check twice the amount (thus worth 2V); + There's a one in two chance that the other envelope contains a check half the amount (thus worth V/2) ‘The expected amount obtained by switching envelopes would then be Enwiteh = 50% x 2V + 50% x V/2 = V + V/A = 5/4 x V, which is greater than VV. ‘Thus, the contestant would be better off switching envelopes, which, is absurd since both envelopes play the same role and the contestant, having not yet opened the first envelope, has no means to distin- guish between them, ‘The puzzle gained popularity thanks to Martin Gardner, who de- scribed it in 1982 in his book «Aha! Gotcha». Renewed interest in the paradox came after Barry Nalebutf published an article listing several probability paradoxes in the Journal of Economic Perspectives. After receiving many responses to this publication, he prepared a second article directly addressing the Two Envelopes Paradox titled «The Other Person’s Envelope is Always Greener». Could the same game be «more profitable» for each of the two part ners? Clearly not. Isn’t this a paradox because each player mistak- cenly believes their chances of winning and losing are equal?”34 Charter 2 MatHeMarics oF DECISION From Nalebut’s perspective, the first satisfactory explanation of his problem is given by Sandy Zabell in «Losses and Gains: The Paradox of Exchange». To paraphrase Nalebuff's explanation: Everyone believes that the amount they see doesn't matter because there's a chance that a larger amount is in the other envelope. This means one assumes the probability of their envelope having a higher amount is 1/2 regardless of what they see. This is only true if each. value from zero to infinity is equally likely. But if all infinite pos- sibilities are equally probable, then the probability of each value is zero. Then each outcome has no chance, which makes no sense. 2.5 Monty Hall Problem ‘The so-called Monty Hall dilemma is a well-known puzzle named after the first host of the American TV show «Let's Make a Deal». In this TV game, the host pre- sented participants with a choice| of three doors. Behind one door was a car, and behind the other two were goats. The car and goats were randomly placed beforehand and remained in their positions throughout the game. ‘After the participant made their choice, the host would always open. ‘one of the remaining two doors behind which, he knew in advance, there was no car. The contestant then had the option to stick with their initial choice or switch to the third door. ‘There are actually several possible strategi for Monty.25 Monty HALL PRoLem 35 + Infernal Monty: The host suggests changing the choice if the door is correct. + Angelic Monty: The host recommends changing the choice if, the door is incorrect. + Goaty Monty: From the start of the game, the host picks one of the goats and reveals it if the player has chosen a different door. It is therefore preferable to base the problem on an unambiguous statement, including the constraints of the host and described by Mueser and Granberg as follows: + There are three doors: one hides a car, and the other two hide goats. The prizes are randomly distributed. + The host knows the distribution of the prizes. + The player chooses one door, but nothing is revealed + The host opens another door not revealing the car. + The host suggests the contestant change their choice of door to open. ‘The host never opens the car door. So, if the player chooses a goat, door, the host will open the only other goat door. And indeed, if the player chooses the car door, the host will randomly open one of the two goat doors (possibly pre-determined by a random draw). ‘The question then arises, «Does the player increase their chances of ‘winning the car by changing their initial choice?», or in other words, «ls the probability of winning by switching doors greater than the probability of winning without switching doors?» ‘The overwhelming majority of players and respondents refused36 Cuarrer 2 MatHemarics oF DECISION to change their choice, even though it would have doubled their chances of winning. At the same time, people think that with the two remaining doors, the chances of winning are equal, and there's no point in changing their choice. If you think the same, don’t feel ‘embarrassed because you're not alone in this misconception. Below is a part of a famous statement of the problem, from a letter Craig F. Whitaker published in the «Ask Marilyn» section of Mari- lyn vos Savant's Parade Magazine in September 1990: Suppose you're on a game show, and you're given the choice of three doors: Behind one door is a car; behind the others, goats. You pick a door, say No. 1, and the host, who knows what's behind the doors, opens another door, say No. 3, which has a goat. He then says to you, «Do ‘you want to pick door No. 2? Is it to your advantage to ‘switch your choice? ‘The publication of this article in Parade Magazine had an immediate impact on the readership and sparked countless discussions among mathematicians, both famous and unknown, and anonymous enthu- siasts. Marilyn vos Savant received over 10,000 letters. As you can see, trolling thrived even in those times when it required much more time and effort than today, and even the cost of a postal envelope and stamp. ‘The talented Hungarian mathematician Paul Erdés also fell into the trap and even refused to make a decision until he saw with his own eyes a computer simulation of the experiment’s results. To be hon- est, it’s hard to believe, but the rumor still spread.25 Monty HALL PRoLem 37 Result ifyou | Result if you switch | don’t switch your choice | your choice Goat Car Car Goat Car Goat For a winning strategy, the following is important: if you switch your door choice after the host's actions, you win if you initially ‘chose the losing door. ‘This will happen with a probability of 2/3, because initially, you can pick a losing door in 2 out of 3 ways, But often, when solving this problem, reasoning goes something like this: the host always removes a losing door in the end, so the prob- abilities that a car appears behind the two unopened doors become ‘equal to 1/2, regardless of the initial choice. But this is not true: al- though there are indeed two choice options, these options (given the context) are not equally probable. This is because initially, all doors had an equal chance of winning but had different probabilities of being eliminated. For most people, this conclusion contradicts their intuitive percep- tion of the situation, and due to the emerging divergence between logical conclusion and the intuitive leaning, the task is called the ‘Monty Hall paradox. ‘The situation with the doors becomes even clearer if we imagine not 3 doors, but say, 1000, and after the player's choice, the host removes 998 leaving only 2 doors: the one the player chose and one more. It then seems more obvious that the probabilities of finding a prize behind these doors are different and not equal to 1/2. If we switch the door, we only lose if we initially choose the prize door, with a probability of 1/1000, We win in the case where our initial choice was wrong, with a probability of 999 out of 1000. In the38 Charter 2 MatHeMarics oF DECISION case of 3 doors, the logic remains the same, but the probability of winning when the decision is changed is 2/3 and not 999/100. Did you experience cognitive dissonance trying to grasp this para- dox? 2.6 And what about the pigeons? Surprised by the inertia of some models of the human mind, researchers Julia Schroder and Walter Herbranson [herbranson2010birds] decided to test the results on pigeons ‘who were successful before in a number of practical probabilistic problems. The birds did not disappoint expectations this time either, After some training, the pigeons empirically learned to choose the correct strategy, whereas people in the same experiment did not. Here's how the experiment was organ- ised. Scientists selected six common blue pigeons and gave each a choice between three shiny feeders. ‘The ini- tial selection with the beak followed; all three feeders turned off, and af- | ter a short pause, two started shining again, among which the pigeon made ‘an initial choice, The computer simu- lation replaced Monty Hall, removing ‘an empty feeder. Afterward, the sub- ject could choose again from the re- maining two. ‘The prize was food, and when the pigeon guessed the correct feeder, it opened, and the bird received a reward. This choice and the re-26 AND WHAT ABOUT THE PIGEONS? 39 ward the pigeon got upon success reinforced the incentive and gave a boost to learning. Then a new set of glowing feeders appeared, ‘The birds quickly learned to make the right choices by «caleulat- ing» their advantages, and within 30 days, the percentage of feeder switches went from 36.33% to 96.33%. Some birds reached absolute figures: they always changed their choice, ‘This was different with people. During the 30 days of experimenta- tion, progress was initially observed, but a trend could not be iden- tified, The observed increase in choice change went from 56.67% to 65.67%. However, the confidence interval boundaries indicate that the choice might have been determined by chance. Another series of tests was conducted in which the conditions of the Monty Hall dilemma were set in such a way that sticking to the initial choice became more profitable. The goal was to test the brain's ability to find optimal strategies even when conditions sud- denly change. In the second experiment, the location of the prize was determined only after the initial selection, The result confirmed the trend. The pigeon’s brain’s central pro- cessor calculated everything correctly. On the first day, the wrong strategy was chosen in 30.17% of cases, and on the last day (15th), only in 4.33%. The variation among young Homo sapiens, on the other hand, was not significant: on the first day, they changed their choice in 30% of cases, and on the last day, 27.67%.