[Suggestion] Narrowing the divergence between AI ratings and player ratings

Currently, my perception is that there is a big difference between the strength of an AI with a certain multiplayer rating and a human player with the same rating. I'm a not-very-good multiplayer player--current rating around 700. My last 8 or so matches have been roughly balanced between matches against humans, sometimes with ratings a little higher than mine, sometimes a little lower and matches against the AI, generally rated somewhat higher than me. I've lost almost all of my actual PVP matches--something like 3 out of 4. But I've won almost all or all of my matches against Gary--even when Gary's rating is 150 higher than mine, implying that Gary should win >66% of our matches. It's a small sample, so that could easily be biasing things, but other people on the boards seem to be saying similar things--that the ratings of the AI are inaccurately high. This has the net effect that playing games against both the AI and other players increases ratings, causing you to be matched up against inappropriate (too good) opponents, whose scores are more accurate--presumably because they either uncheck the check box to play the AI, or play at times of day when there are more opponents around.

If this problem is real, I suggest that the ratings of the AI opponents be lowered, or even better be allowed to float if they're not, so that they'll end up as accurate ratings. Of course, the problem could be that my tactics are better against the AI than against other players relative to other players with similar ratings. But I suspect that part of the problem is that the AI ratings are just too high, which produces unfortunate ratings change patterns.

 

I wonder if it would be feasible to have Gary et al. have a real multiplayer rating. That is every time Gary wins or loses a multiplayer game, his rating should be adjusted just like it would be if he was human.

 

Yup, Jon has said they do have multiplayer ratings.

 

Huh, if the multiplayer ratings for Gary and friends float, then they should be appropriate matches. That suggests that either I'm getting fooled by a small sample, or that my tactics are more effective against the computer than against a real human. <shrug> Thanks for the info!

 

I may or may not have occasionally inflated the AI rankings (and tanked my own) while trying out experimental decks which proved to be utterly mismatched for a particular AI style. >_> (Really, it sucks to build a fresh 3-warrior deck and get matched right away against Mom...)

Although I think this is a larger problem with rankings - my own fluctuates between 1000 and 1400, depending on whether I'm fooling around with new builds or even just get an unlucky string of matches against the wrong types of builds.