Machine Learning Ruins Blackjack

Blackjack, or 21, is a game that many enjoy wasting their money playing at casinos. For those who don't like to waste their money, or at least want to waste it more slowly than others, card counting is a time honored tradition for moving the odds away from the casino and in the players direction (blessed be Ed Thorp.) In other words it makes the game at least slightly enjoyable for those who like to win. But now a graduate of the University of Dundee, Kris Zutis, is going to ruin this small smidgen of fun:

A University of Dundee graduate has created a computer system with the potential to make the game of Blackjack fairer by detecting card counters and dealer errors.

Okay so catching dealer errors certainly makes the game "more fair." But detecting card counters? People who are eking out a minor advantage (and have to be aware of methods to avoid detection because casinos can kick them out not because of card counting per se, but because the casinos run the game) by using their damn brains are not acting fair? To be fair, of course casinos are already doing this so we should be nice to the grad student :)

And further, of course all is fair in love, war, and casino games. But this makes me wonder about arbitrage in the era of machine learning, each machine vying to outdo the other in keeping their profits locked up tight. My high margin classifier just gave me 21, yipee! Oh wait, this is already happening on Wall Street. Remind me again about the market making and liquidy arguments for blackjack.