KGS Issue - Bot Plague
At KGS, it is possible to run Go playing robots like e. g. GnuGo or self programmed ones. Go is one of the hardest challenges to game programmers, and a testing field like KGS is a nice support to facilitate program development.
However, over time, the number of robots has considerably increased. There are often a dozen or more bots simultaneously offering games nowadays. Many of them are nothing but clones of the same version of GnuGo. I wonder if it's appropriate to give (or suggest to wms) some aid to people who prefer a clearer open games view. Just as an idea, we could have clickable column headers which would allow the user to sort the list in various ways, with the user type (i. e. human or robot) being a small distinct column. What do you think?
Javaness Personally I don't understand what the problem is. People are having fun playing around with computer Go and GTP, why is this socially harmful?
blubb: Thanks for your reply. My point is, if the count (actually, the bot fraction of open games) increases further, it will start to become an annoyance. I, for one, would not check a "censor all bot offers" checkbox, since I am interested and involved in computer go by myself, and occasionally want to take notice of what happens. I would like to be able to sort them away, though (e. g. to the bottom of the list, to find all human offers condensed at top).
Poketen?: I have never seen so many bots on KGS, but I agree that it would be nice to put them at the bottom of the list automatically. This would be easier for 20-30k when not enough bots and looking for one, or too many bots and looking for a human (note that in this case you can just create a game yourself..)
Here's a screenshot of the "Open Games" tab as it stared at me (blubb) today:
Gresil: There's a much more irritating kind of bot plague: a pool has developed of players who play bots exclusively or nearly exclusively. These players' ranks are inflated many stones above their true skill level. They invariably lose to players of similar nominal rank who have acquired their ranks playing other people. There should be a sign comparable to ~ that is acquired for playing against bots too much. It is frustrating to have one of these people as opponent, and it is very hard to tell such a person by simply looking at his game list as bot opponents aren't visually indicated.
atila: In order to avoid the distortion in human users ratings, I suggest a modification in the rating system: All games played by recognized robots should not affect human opponents ratings at all. These games should only count to calculate/modify robot ratings. I don't think such modification would be very difficult to program, and it certainly would serve to the accuracy of human players ratings.
Steve: If games between humans and bots are not rated for the humans, they should not be rated for the bots either, since the ranks calculated for the bots will not be based on games which are not similar in nature to all other ranked games on the server. I would suggest that only a certain percentage of one's rated games may be against a specific bot. If that percentage is already exceeded, the bot should refuse to play a rated game against you (or the ranking system should ignore that game). For example, if the percentage was 5%, the bot will see if you have more than 1 game in your last 20 rated games against him (or ask the server this). If so, the game can not be rated, if not, the game can be rated. Extending this idea for human vs. human play may help remove the 'isolated pools' rank inflation/deflation issue.
Fusojigen: How are the ranks of bots decided? - by playing other humans on equal terms. If a player is strong at fighting, what do you do? You adopt a strategy that best exploits his weaknesses while subduing his strengths. Should this be considered for the respective players rank and objective measure of strength? What is unfair for these bot-loving players, or more importantly, players who occasionally play bots. If they are good or bad against bots, does this not reflect their objective strength? The problem is that the population from which the rank of only-against-bot-players is estimated does not accurately represent the true population of players. Some players are bad against bots and would be ranked far below their real in this kind of situation, while others far above. The same would happen if a player would select any other too restricted and benefital set of games. Examples would include handicap games, fast/slow games, games as black/white, against a particular group or type of player. I believe we are discussing a much larger problem even though marks would be okay for this particular instance.
Peterius: I had played speed games against gnugo, a bot, off of KGS for a long time before returning to KGS. Because gnugo is so good at tactics, I have gotten better at strategy to compensate and compete with it but my fighting skills have weakened through lack of use. The past couple days I have started playing on KGS again and I find that my rank has improved. My fighting is still weak but I am aware of enough tesuji to make an attempt and my strategy is better than many of the other players at my rank.
What it amounts to is that I have a different playing style than many players which probably adds to the diversity of the pool of play styles. In any situation where a person plays the same type of game over and over again, they will become weaker against other styles of play. If you play your friend every day, you will get better at playing your friend and he, you, but you might find that you are very weak when you play other people. Bots are no different. I have become unnaturally good at strategy for my poor, unused reading and fighting skills.
Taking bots out of the ranking system in any way, I think, is a bad idea. They are simply players with particular styles like any person. If someone is purposely inflating their rank by playing a bot over and over again, first of all they could simply do this against another fake account, playing themselves, and second, as soon as they play anyone else, their weaknesses will show through and their rank drop.
But really this is a separate discussion and maybe should be moved to a different page. In terms of a feature to distinguish bots and people in open games on KGS, that seems useful though perhaps low priority in terms of enhancements. Though the screenshot proves the legitimacy of the "bot plague" claim, I have never seen such a preponderance of bots.
Tapir: A minute ago i kibitzed a game between a bot-player and a bot. The bot (Leela) was ranked 4k, the player was ranked 6k after playing exclusively to bots. The bot played a über-cosmic game taking the whole center, touching the edges but granting most of it + all 4 corners to black and still won. What disturbs me is that while the bot seems to have had a "plan" :) the player had not. Even in detail his play was adopted to bot mistakes & strategy - getting used to a lot of minor mistakes himself. - 2nd thought: If the bots are trained through players adopted to bots both will have a hard time to improve. I don't mind for the bots, but it is sad for the players.
Andurian?: I'm a beginner and I've been playing an abysmal bot called "butterbot". As I play it my rank goes up and its handicap improves, but I'm pretty sure I could beat it with a 20 stone handicap...it does stupid things like playing a stone into atari and not bothering to keep me from connecting my groups. So if I keep playing it my rank will rise far above what my actual skill level is. It might already have... I'm probably developing bad habits, but its hard not to play it again - its fun capturing lots and lots of stones. :) Anyhow, it makes sense to me to limit the amount of rank increase one can gain from bots - people are the real measure of skill.
kokiri: i think there will always be ways to game the rating system, and i think that excluding bots would throw the baby out with the bath water - it's really valuable to see how bots rank with we flesh-based computing machines. If some people covet a number more than rounded game, then w/ever.
If a human beats a bot, the bot's rank should get weaker. If a weak player takes advantage of a bot's weakness, then the bot's rank should plummet. If a human player can inflate his rank by beating bots, then there must be something wrong with the rating system, not with the human player's behavior.