3 ms·

> For an individual, the best option is betrayal ... Well, this applies to prisoner's dilemma games if you know exactly how many times you will "play" the game

> For an individual, the best option is betrayal ... Well, this applies to prisoner's dilemma games if you know exactly how many times you will "play" the game with a given other player, including the case where you will only play one time. But for repeated prisoner's dilemma games, where you don't know what games is the last against a given player, a tit-for-tat strategy massively dominates an always-defect strategy. Humans have evolved in environments where repeated prisoner's dilemma situation is common, and have evolved emotions such as empathy and consciousness for interactions with cooperative agents. We ALSO have evolved (to a lesser or greater degree) the ability to turn such emotions off when dealing with people who attempt the always-defect strategy (such as psycopaths). In fact, psychpathic behavior only becomes viable in populations that have been using tit-for-tat for so long that they've "forgotten" how to punish defectors (ie almost all agents switched to always-cooperate). But once you introduce a few always-defect agents into the population, the tit-for-tat agents will soon dominate again. At least for the brain wireing for interacting with members of our own "tribe". When dealing with people outside our "tribe", it's much more likely that we will only play them once, and in such cases, always-defect may indeed dominate. However, there is also a tribe-level game, where the tribe can act as agents. And in tribe-vs-tribe games, depending on circumstances, both tit-for-tat and always-defect can be optimal strategies. Always-defect can, ultimately, lead to genocide, while tit-for-tat can lead to alliances, trade and intermarriage. People who think that always-defect is always the optimal mode of behavior tend to be people who either don't consider it for their close relations, or who have very few close relations. For instance, someone who lives alone in the city and is either unemployed or tend to have purely transactional work relations. What such people tend to forget, is that there is an artificial presence in the city, called the police, that can impose a level of safety that can remove the need for high trust, pro-socal relationships. In a society with police, all you have to do is to know what behavior lands you in prison, and otherwise you can be a selfish bastard. As soon as you remove the police, loners like that are free game for "predators", and only those who have a "tribe" have any kind of protection. Such a "tribe" can be a clan in an Afghan mountain area or it can be a gang in a high-crime city. While a gang member may have little to no empathy or conscience when dealing with outsiders, maintaining good relations with other clan members can be critical (Depending on what specific gang it is.)

everforward 3y ago
I suppose I should have clarified that always defecting is optimal if you're optimizing purely selfishly for material goods. It won't make anyone feel fulfilled and loved. > But for repeated prisoner's dilemma games, where you don't know what games is the last against a given player, a tit-for-tat strategy massively dominates an always-defect strategy. If we're considering repeated prisoner's dilemma games, you also have to consider the gain/loss of power between the various players. The classic prisoner's dilemma consists of two equally impotent players. Over a series of games, a player could aggregate enough power to change the nature of the game. They could aggregate enough power/resources to alter the parameters of the game, like using those resources to offer a substantial reward for cooperating with them or confessing. Inversely, they could use that power to enforce penalties on other players for betraying them. The question then becomes whether an always-defect player can accumulate enough resources to change the rules of the game before the other players switch to tit-for-tat. The answer in the real world appears to be "mostly yes". Uber appears to have chosen "always defect", but they got enough power through their defections that we've become effectively powerless to penalize them for defecting. Theranos is a counter-example where they were eventually punished, though they made the mistake of playing the prisoner's dilemma against already powerful players. Without getting too political, Trump has a history of defecting on business relationships and the man was elected president. > At least for the brain wireing for interacting with members of our own "tribe". When dealing with people outside our "tribe", it's much more likely that we will only play them once, and in such cases, always-defect may indeed dominate. Within the tribe, the power levels tend to be flatter, though, making it easier for an always defect player to seize enough power to not have to play the game anymore. I.e. if the player can become moderately wealthy, the resources they offer their tribe would likely outweigh their history of defection. Company towns are an example of this; virtually all the residents agreed the company was awful, and yet they stayed (for a long while). > As soon as you remove the police, loners like that are free game for "predators", and only those who have a "tribe" have any kind of protection. Such a "tribe" can be a clan in an Afghan mountain area or it can be a gang in a high-crime city. While a gang member may have little to no empathy or conscience when dealing with outsiders, maintaining good relations with other clan members can be critical (Depending on what specific gang it is.) Sure, but that's not the prisoner's dilemma anymore, so always-defect is probably not an optimal choice. A crux of the prisoner's dilemma is that a player can benefit by choosing to defect when the other player chooses to cooperate. If cooperating results in the best individual outcome and the best societal outcome, it's not a dilemma anymore. Cooperating is the obvious best choice. This is part of what makes always-defect viable. Most situations do not mirror the prisoner's dilemma, and cooperating is often the best individual outcome. If you look at an always-defect player, they might be cooperative in 9/10 or 99/100 situations because only 1/10 or 1/100 were prisoner's dilemmas where they could benefit by defecting. I don't think "always defect" is globally optimal, only for prisoner's dilemmas. Always-defect globally implies doing so even when it is clearly a sub-optimal choice (like driving your car into a brick wall just because the police said not to). The parameters of the dilemma scope it to only situations where defecting confers gain.

everforward 3y ago

I suppose I should have clarified that always defecting is optimal if you're optimizing purely selfishly for material goods. It won't make anyone feel fulfilled and loved. > But for repeated prisoner's dilemma games, where you don't know what games is the last against a given player, a tit-for-tat strategy massively dominates an always-defect strategy. If we're considering repeated prisoner's dilemma games, you also have to consider the gain/loss of power between the various players. The classic prisoner's dilemma consists of two equally impotent players. Over a series of games, a player could aggregate enough power to change the nature of the game. They could aggregate enough power/resources to alter the parameters of the game, like using those resources to offer a substantial reward for cooperating with them or confessing. Inversely, they could use that power to enforce penalties on other players for betraying them. The question then becomes whether an always-defect player can accumulate enough resources to change the rules of the game before the other players switch to tit-for-tat. The answer in the real world appears to be "mostly yes". Uber appears to have chosen "always defect", but they got enough power through their defections that we've become effectively powerless to penalize them for defecting. Theranos is a counter-example where they were eventually punished, though they made the mistake of playing the prisoner's dilemma against already powerful players. Without getting too political, Trump has a history of defecting on business relationships and the man was elected president. > At least for the brain wireing for interacting with members of our own "tribe". When dealing with people outside our "tribe", it's much more likely that we will only play them once, and in such cases, always-defect may indeed dominate. Within the tribe, the power levels tend to be flatter, though, making it easier for an always defect player to seize enough power to not have to play the game anymore. I.e. if the player can become moderately wealthy, the resources they offer their tribe would likely outweigh their history of defection. Company towns are an example of this; virtually all the residents agreed the company was awful, and yet they stayed (for a long while). > As soon as you remove the police, loners like that are free game for "predators", and only those who have a "tribe" have any kind of protection. Such a "tribe" can be a clan in an Afghan mountain area or it can be a gang in a high-crime city. While a gang member may have little to no empathy or conscience when dealing with outsiders, maintaining good relations with other clan members can be critical (Depending on what specific gang it is.) Sure, but that's not the prisoner's dilemma anymore, so always-defect is probably not an optimal choice. A crux of the prisoner's dilemma is that a player can benefit by choosing to defect when the other player chooses to cooperate. If cooperating results in the best individual outcome and the best societal outcome, it's not a dilemma anymore. Cooperating is the obvious best choice. This is part of what makes always-defect viable. Most situations do not mirror the prisoner's dilemma, and cooperating is often the best individual outcome. If you look at an always-defect player, they might be cooperative in 9/10 or 99/100 situations because only 1/10 or 1/100 were prisoner's dilemmas where they could benefit by defecting. I don't think "always defect" is globally optimal, only for prisoner's dilemmas. Always-defect globally implies doing so even when it is clearly a sub-optimal choice (like driving your car into a brick wall just because the police said not to). The parameters of the dilemma scope it to only situations where defecting confers gain.