12-10-2021, 01:57 PM
Hi Michael,
Yes, this is something we have encountered in several games, and unfortunately is something that is unavoidable. We are bound by what was written by the people who observed or recorded these games, and very often these kinds of details are not reported. What we have in the database is what is is in the accounts, and for many of the games we have recorded the data about who played it (age, gender, social class, etc, and even the kinds of spaces it was played). The fact that it is not always available presents many challenges, but it is still useful. We can only deal with the data that are available.
Of course, AI does not play like humans. Games are a social act; much of the way that people play is bound by the sociability of their actions rather than the actual rules of the game. That is to say, an AI can play to avoid losing at the expense of ever making a valid attempt at winning—always using legal moves for thousands of moves—, whereas a human would be expected to play to win, or at least to progress toward an end of the game without playing indefinitely with no progress. The AI would stall for all eternity to not lose. While the AI values winning, most human play values the social aspect of play, and there are social pressures against stalling or playing in an antisocial way. Social rules are of course culturally contingent, and are indeed situational, and may vary between games. The question of etiquette is difficult to model with AI, and this is very very rarely, if ever, recorded by ethnographers, who would need to spend a lot of time studying game players in particular, whereas usually the games are just a side note to the work they are doing. Such is the nature of games research. So, rather than "drunk and blind-folded players," what we are dealing here is the more intangible style of play that may not be reported as rules, that kind that we expect in the modern world where games are designed to produce a clear winner. Even chess imposed kinds of etiquette as rules (turn limits without capture, etc) by organizations to produce winners in tournaments, where the social aspect of play is devalued and winning is the only objective. These were not part of the traditional game, and even today are not always observed or even known by casual players, because winning isn't the only objective of play.
We have methods of manipulating the AI agents when we do our playout experiments to compel them to play in ways that value certain kinds of actions, such as changing the thinking time, changing the search depth, adding heuristics, etc. Our analyses will contain hundreds of playouts per game, so we will be able to see if such possibilities are indeed something that are common in games—i.e., if these rules always lead to a kind of broken-looking game. We can then see if there are many games that have a bare set of strict rules which rely on etiquette for completion. We can consider this when making reconstructions, by comparing the games we are reconstructing, and their known ludemes, to those that are more complete. We expect similar games to play similarly, and expect that they will play as well as other games which are from nearby cultures (both chronologically and geographically)—although, maybe we will find that games within a certain culture vary widely in the ways they are played. These are all things we can investigate. If relevant games often have this kind of situation where they seem to be more driven by etiquette than hard rules, we can take this into consideration for what kind of metrics we expect a game to generate. We also take into consideration the social aspects of the games, where available, based on the kinds of games also played by those demographic groups/in those spaces.
We will never know if any of our reconstructions are correct. We can only deal with the data given to us. And indeed, individual games are played with many variations in different places, sometimes even down to the household, so there is no way to be able to say that a reconstruction is the only possible one. We can only say that a reconstruction fits the profile of what we know about a particular game, those similar to it, those we know from that region or time, or by people belonging to similar social groups. Furthermore, the reconstructions are not an end point; they are the beginning of a conversation about what to expect from games from various cultures and how to reconstruct them, and what methods to use. So much is unknown about games, and there are many assumptions out there about how games are played, what makes a good game, etc, and we aim to measure and challenge some of these assumptions in our work.
Yes, this is something we have encountered in several games, and unfortunately is something that is unavoidable. We are bound by what was written by the people who observed or recorded these games, and very often these kinds of details are not reported. What we have in the database is what is is in the accounts, and for many of the games we have recorded the data about who played it (age, gender, social class, etc, and even the kinds of spaces it was played). The fact that it is not always available presents many challenges, but it is still useful. We can only deal with the data that are available.
Of course, AI does not play like humans. Games are a social act; much of the way that people play is bound by the sociability of their actions rather than the actual rules of the game. That is to say, an AI can play to avoid losing at the expense of ever making a valid attempt at winning—always using legal moves for thousands of moves—, whereas a human would be expected to play to win, or at least to progress toward an end of the game without playing indefinitely with no progress. The AI would stall for all eternity to not lose. While the AI values winning, most human play values the social aspect of play, and there are social pressures against stalling or playing in an antisocial way. Social rules are of course culturally contingent, and are indeed situational, and may vary between games. The question of etiquette is difficult to model with AI, and this is very very rarely, if ever, recorded by ethnographers, who would need to spend a lot of time studying game players in particular, whereas usually the games are just a side note to the work they are doing. Such is the nature of games research. So, rather than "drunk and blind-folded players," what we are dealing here is the more intangible style of play that may not be reported as rules, that kind that we expect in the modern world where games are designed to produce a clear winner. Even chess imposed kinds of etiquette as rules (turn limits without capture, etc) by organizations to produce winners in tournaments, where the social aspect of play is devalued and winning is the only objective. These were not part of the traditional game, and even today are not always observed or even known by casual players, because winning isn't the only objective of play.
We have methods of manipulating the AI agents when we do our playout experiments to compel them to play in ways that value certain kinds of actions, such as changing the thinking time, changing the search depth, adding heuristics, etc. Our analyses will contain hundreds of playouts per game, so we will be able to see if such possibilities are indeed something that are common in games—i.e., if these rules always lead to a kind of broken-looking game. We can then see if there are many games that have a bare set of strict rules which rely on etiquette for completion. We can consider this when making reconstructions, by comparing the games we are reconstructing, and their known ludemes, to those that are more complete. We expect similar games to play similarly, and expect that they will play as well as other games which are from nearby cultures (both chronologically and geographically)—although, maybe we will find that games within a certain culture vary widely in the ways they are played. These are all things we can investigate. If relevant games often have this kind of situation where they seem to be more driven by etiquette than hard rules, we can take this into consideration for what kind of metrics we expect a game to generate. We also take into consideration the social aspects of the games, where available, based on the kinds of games also played by those demographic groups/in those spaces.
We will never know if any of our reconstructions are correct. We can only deal with the data given to us. And indeed, individual games are played with many variations in different places, sometimes even down to the household, so there is no way to be able to say that a reconstruction is the only possible one. We can only say that a reconstruction fits the profile of what we know about a particular game, those similar to it, those we know from that region or time, or by people belonging to similar social groups. Furthermore, the reconstructions are not an end point; they are the beginning of a conversation about what to expect from games from various cultures and how to reconstruct them, and what methods to use. So much is unknown about games, and there are many assumptions out there about how games are played, what makes a good game, etc, and we aim to measure and challenge some of these assumptions in our work.