FIBS Board backgammon forum

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.

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.