Editorial: Three Reasons Why We Shouldn’t Be Worried about “Libratus”

A couple of weeks ago in the Rivers Casino in Pittsburgh, a rather stunning occurrence took place. Over the span of 120,000 hands and roughly 20 days of play, a computer artificial intelligence (AI) called “Libratus,” developed by the Carnegie Mellon College of Computer Science, defeated a squad of tough poker professionals – Dong Kim, Jason Les, Jimmy Chou and Daniel McAulay – in heads up No Limit Texas Hold’em. “Defeated” is a bit too mild a word. “Devastated” would be a better one as, once the final numbers were tallied, “Libratus” was able to score a $1,766,250 win over the humans, a crushing defeat so thorough there was no room for any discussion.

Fear not, humanity. This isn’t the end of the world as we know it. The truth is that there won’t be a tableful of “Libratus” knockoffs sitting around the online baize at any point soon. Here’s three reasons why the human race is still the “top dog” when it comes to the game of poker…at least for the immediate future.

Did They REALLY Play Texas Hold’em?

As my friend and colleague Dan Katz pointed out in his rundown of the rules for the staged event, there’s some question as to whether what happened at the Rivers Casino could be held as a true facsimile of an actual Texas Hold’em game. With each hand played, the players and “Libratus” started with identical stacks (20,000 chips, with blinds of 50/100). No matter what had occurred on the previous hand, the stacks were reset each and every time. The logic of this is that it didn’t allow for either side to build up a stack to “bully” their opponent.

That in its very nature is a part of the game of Texas Hold’em, however. You use your stack as a weapon just as much as the cards and, as such, can make players with stronger holdings let their cards go to the muck rather than stay around. It would have made more sense to reset the scores if someone busted to allow for that “bully” factor to be involved rather than start each hand equal, which in no way is how poker is played.

Then there was the odd “all in” rule. If either the live player or “Libratus” were to go all in at any time during the dealing of the board, action on the hand was immediately stopped and the winning percentages were determined AT THAT POINT, with the chips in the pot then being divvied up by those percentages. This removed “luck” from the game in the form of “sucking out” on a better hand, but that is a part of the game of Texas Hold’em; sometimes your pocket Aces are going to get run down by that paltry 9-3 (and usually at the worst moment ever).

These rule changes were enough to allow mankind to question the outcome of the “Brains vs. AI” battle in Steel City. If they weren’t playing by the rules that humans normally play by, were they actually playing poker?

So You Think You’d Like to Own “Libratus?”

There is no question that “Libratus” is a feat of engineering and computer programming (done by humans, it must be added). The ability of the AI to not only build its own database (it played, per the Reading Eagle, literally trillions of hands against itself – the human analog of sitting on the bed and turning cards – before even entering the competition to build information) was stupendous, but it also would continue to learn about its opponents as the game was going on. This allowed for “Libratus” to do exactly what humans do on the felt – make gear changes that throw the opposition off their game, so to speak.

It isn’t going to be next week that a card sharp will have “Libratus” in their pockets, however. The AI was run by a supercomputer that, per the Eagle, costs millions of dollars to operate per year. You’re not going to see someone wheel a bank of computers up to the chair behind them to play anytime soon – and you’re not going to run into someone online who has a similar capacity sitting in their home (just as an aside – and one of the reasons the development of “Libratus’” AI is important – it gives artificial intelligence improvements to be able to better make decisions in an “imperfect” situation).

It’s Just Heads Up Texas Hold’em

Let’s not completely disavow the victory by “Libratus.” Over that many hands of play – and even with the weird rules in place – the monumental victory is obvious. But let’s also keep in mind that this is just ONE discipline of the game, ONE part of what makes poker a fascinating game.

First, it was heads up. You’re not going to see special tables set up in a casino for a mano y mano battle. Hell, they barely can keep such tables active in an online setting. It is a special game, heads up No Limit Texas Hold’em, and there are nuances that are utilized that would be devastating in other settings. This was pointed out by Kim, who noted to the Eagle, “Those guys (his fellow human players) don’t play our game type. They might play other kinds of poker, but even small-stakes heads-up players on the Internet would crush them.”

Second, the AI only had to deal with one opponent. The calculations utilized to make the choices numbered in the billions in that singular circumstance. There has never been the attempt to take any poker-playing AI against even a four-player setting, let alone a six-max table or a full nine handed cash game. With that, the number of calculations would be astronomical and could overwhelm the AI.

Finally, did the AI even try to take on the complexities of another game…Omaha Hold’em, for example? With more potential hands, would that hurt the AI’s computational powers? Let’s not even get into the potential of “wild” games or other variations (Hi/Lo?) that could muck up the AI’s strategy.

It isn’t the end of the world that “Libratus” and Carnegie Mellon’s brainiacs could defeat the human race. There’s still a great deal of space between a person having the AI’s power in their pocket to utilize and, furthermore, there’s more to the game of poker – long respected because of its “incomplete information” setup – than just Texas Hold’em. When the AI is ready to take on a full table of nine players – or, better yet, step into a 1000 player tournament – and win, then get back to me. For now, humanity is still the “king” over artificial intelligence when it comes to the game of poker.