Bot or strong human beats weak human more?

[Krazula]

Do you think that a strong human with a fibs rating of about 1950 will beat a weak human rated about 1400 more consistantly than a gammonbot will beat that weak human?

Please vote and take the time to explain your reasoning.

Behind the spoiler tag are my opinion and explanation.

Spoiler

I think the strong human will win more often. I think the main flaw of the bots is their inability to profile their opponents and exploit the errors that they are likely to make in their future decisions. The bot assumes its opponent plays exactly like the bot, this assumption isn't too harmful against a strong opponent since the strong opponent plays similiar to the way that the bot does. But this assumption becomes increasingly detrimental against weaker players because the weaker the player the less like a bot he/she plays.

If a particular weak player makes the error of taking too many cubes an aware human can exploit this easily by waiting to double, a bot will double as soon as it thinks the proper cube action is double-take or double-drop and since a bot won't lose its market often this will make the weak human's tendancy to take cubes often correct. The strong human that exploits this can wait until he/she has a bigger lead before offering the cube and can be less concerned about losing his/her market. The strong human will also keep the cube at a lower level resulting in more rolls per match and more opportunities for the weak player to make more errors. The strong human may also lean towards making a slotting play over a splitting play because they result in longer games on average, and the longer the game the more errors the weak player has an opportunity to make. There are other ways that the human can exploit the weaker player, but I think those are the biggest most common ways. 

I think that the equity gained in exploiting the weak player's mistakes is larger than the equity difference between the bot and strong human in both technical and thematic decisions.

[close]

[dorbel]

Spoiler

There is an underlying weakness in the question. A GammonBot (essentially gnu 2-ply) always makes the same plays in the same situations and uses the cube on the assumption that it has an equal opponent. These are its strengths and paradoxically, its weaknesses, because it would do even better than it does if it could assess its opponent's playing strength and vary its play, particularly its cube action, according to the strength of its opponent and the difficulty of the position. However, as it will already consistently beat every player on Fibs and probably break even against even the very strongest humans, playing even better is not really necessary.
A 1450 Fibs player on the other hand isn't so easily categorised. You can assemble a representative group at this level and they will all be weak in different ways. Not only that, but it is unlikely that even one individual will be consistent. Sometimes he will pass easy takes, sometimes he will take big passes, sometimes he will fail to double a race that is already a drop, sometimes he will suddenly flash out the cube in the middle of a complex and even positional struggle. Sometimes he will pass the position that he took yesterday, sometimes he will fail to double a position that he would drop from the other side! Al in all, the bots strategy is probably the best. Make strong checker plays, cube when you get into the window and reap the benefits, which against players at that level in thre point matches probably means winning 75-80% of the matches. Can a 1950 fibster do better than that by altering his play with checkers and cube to suit his opponent? Unfortunately, we can't say because 1950's fibsters are as diverse a group as the 1450s! They too will have a variety of strengths, weaknesses and inconsistencies and they too will be making mistakes with both cube and checkers even when playing at their best! Just making objectively strong plays will give them a good winning percentage of course and if they are intelligent and understand where and when they might adjust their play to their particular opponent, they can sometimes do better, but they will never approach the win rate of a bot.
This of course is just opinion. What we need is data. We need a cross section of 1450 players and a cross section of 1950 players. The 1450s will play 100 matches against the bots. They will also play 100 matches aginst the 1950s players. Then we will see, but I would bet heavily on the bots to have a better result. I can't imagine that there would be many willing to back the 1950s though.
 

[close]

[garp_02]

I'm happy to volunteer as a 1450 guinea pig 

[Zorba]

I assume we're talking about a player that averages 1950 on FIBS, by playing normal matches (3,5,7, maybe longer) against a variety of opponents, ergo a real 1950 FIBS rating.

Same for the 1400 player, and I'd like to add that I will assume it's someone who's been playing for a while, not a newbie.

I think that on average, the Gammonbots (GNUbg 2-ply) will do better in 7pt matches. However, if you select the 1950 players such that they are a group that claims to be good at exploiting weak opponents, and that they have at least somewhat critically assessed their own claims to do better that way (by bot analysis), then I think the 1950 players have a good chance to do at least as well as the 150-200 points higher rated bots.

You can give the humans more of an edge by letting them play longer matches (more chances to find out cube behaviour, 7pts is often too short for that), and letting them play against the same opponents several times, so that they can get a clearer pictures of that opponent's specific mistakes.

Of course there's also added information in live play, I don't have much experience with that but my guess is that certain players could benefit quite  bit from extra knowledge about f.i. opponent's emotional state (and even influencing it, of course). The time someone takes to make a move (or think about a cube!) can also give interesting information, but especially online it's not always reliable.

[Doc]

a human will do better 80% of the time. the bots are roll rigged (IMO) and don't make the right moves for a roll value.

I have been so far behind AI and won...as most of u have.

All I do is think outside the box and it screws the AI. The roll is the issue...not the AI

Lou.