Someone should do something similar like this one but with Subway Surfer. 
I've not played it, but did a quick google to get the idea of it.
This is possible, but I see a real problem here... part of the challenge of subway surfer appears to be that it is not a "perfect information" challenge, much of what makes it difficult is that the player doesn't know what will be coming up next. Bots would be able to "look ahead" down the track indefinitely, which would give them a huuuuuuuge advantage compared to humans. (It would make the moto bots' overwhelming of humans look trivial.)
Humans in any game (even GO) will be beaten by computers eventually if there is enough incentive (monetary reward in this case) to do so.
I think instead of a futile search for a "bot-proof" game, we should be thinking of a game/challenge that could be INTERESTING to solve through programming a bot, with many different possible approaches, where a "smarter" bot with fewer hardware/energy resources available could easily out-mine bots that have much more resources available but use more primitive approaches.
This game/challenge would also naturally grow in complexity over time in a manner that would make it difficult for bot designers to predict and quickly adjust to.
Ideally this game/challenge/problem would not even be something arbitrary but somehow contribute towards cutting edge AI and machine learning research in general. Which in turn would attract the world's leading researchers and the most talented programmers to work together in pushing the limit of machine intelligence.
How about that?