Libratus Wins But No-Limit Texas Holdem Far From “Solved”


In the beginning there are nothing but great intentions. Hormones rage like a wasp nest poked with a stick. There is a symbiosis.

Like a fridge and a novelty magnet from your last vacation, you can’t bear to be apart. And then time does its thing. It doesn’t work anymore. The Four Horseman of the Apocalypse — criticism, contempt, defensiveness and stonewalling — take hold.

Nothing you do works. Everything you try fails. And as you sit there in the midst of this raging inferno of a once glorious relationship, you can’t help wondering what the hell went wrong?

That’s exactly how I feel when I consider the impact of humankind’s forays into the world of games and artificial intelligence: Deep Blue, AlphaGo, Cepheus and now Libratus.

What Happens to Humanity?

Each time a computer shows us who’s boss, we swat the importance of it aside. But just like a marriage made in heaven turns into a holiday home in Hell via the thousands of little acts that never seem to mean anything, we need to understand each piece moves us closer to the completion of the human evolutionary jigsaw.

jeff jordan

REG’s Jeffrey Jordan.

And when that happens, what then? What happens to online poker? What happens to humanity?

As he is a man very dedicated to the future of humanity, I reached out to Jeffrey Jordan, Director of Outreach at Raising for Effective Giving (REG), before the Challenge to see if he had some answers. Here’s what he had to say.

LD: Do you think the humans will beat Libratus?

JJ: Probably not. Now that AlphaGo has dominated the best Go players it seems more likely that Libratus will beat the best poker players.

LD: Do you care if an AI wins and solves NLHE?

JJ: I care very much if an AI wins but an AI winning does not mean that NLHE is solved. Often when talking about solved games, solved has a more specific meaning than ‘an AI can win every time.’

In the sense that I’m using it a solved game can have its outcome correctly predicted (win, lose, or draw) from any position, assuming that all players play perfectly.

Connect Four is a great example of a solved game. For each player in Connect Four there is the best move they can make on any given turn. When a participant always makes their best move they are said to be playing perfectly.

Each time a player makes a move in Connect Four the board changes depending on their actions, moving it into a different state. In Connect Four, assuming that both participants play perfectly, you can always predict who will win or lose for any possible state. Thus, we can say that Connect Four is a solved game.

For NLHE to be solved we would have to be able to predict its outcome from any possible state, assuming that all participants played perfectly. Solving NLHE is very different from an AI being able to win against humans every time.

Somerville table2

“For NLHE to be solved we would have to be able to predict its outcome from any possible state.”

Even if Libratus always wins it does not imply that it has solved NLHE; it just means that it is better than the best human players.

That said, I care a tremendous amount if an AI beats the best human poker players. Such an outcome would be a significant milestone in creating an AI that could perform any task that a human being can (an Artificial General Intelligence).

Computers already behave intelligently; AIs are better than humans at playing chess and predicting consumer behaviour, both of which require intelligence. What AIs are currently bad at is applying intelligence in a general context.

Deep Blue can beat the best human chess players but cannot navigate a supermarket, as it can’t generalize its intelligence across different situations. In my opinion, poker requires much more general intelligence than chess. If Libratus can best us at poker, we will be one step closer to human-level or superhuman level AI.

Now why would such an AI be concerning? If done improperly the creation of a smarter-than-human AI could be catastrophic for humans and other sentient creatures. Humans are marginally more intelligent than chimpanzees yet because of this small advantage the fate of chimpanzees depends more on our actions than their own.


Machine Intelligence Research Institute.

A smarter-than-human AI would create the same situation; the fate of all other sentient beings would depend more on AI’s actions than our own. We had better be sure that a smarter-than-human AI is friendly towards other beings.

Some of our recommended charities, like the Machine Intelligence Research Institute and the Foundational Research Institute, are working to ensure just that.

LD: If the AI wins is online poker as a profession dead?

JJ: I’d like to emphasize that my answers are very speculative as I’m not an industry expert. I think that even if Libratus wins online poker has a future.

People might have a lot of fun playing against AIs, even if they’re likely to lose. Online poker could still host ‘human only’ rooms for those that like winning.

Systems for detecting poker bots rely on captchas, which means that unless poker bots start incorporating advanced computer vision techniques, they can still be effectively banned from online poker.

LD: Do you think online poker rooms have the ability to prevent superintelligent bots competing in your games?

JJ: It seems that current systems for detecting bots would work even if poker bots became significantly more intelligent. Many online poker sites use captchas to detect bots which should still be effective against even the best poker bots.

Poker bots can play poker well but can’t decode captcha images. There also seem to be ways to ban AIs that offer real-time data/calculations for the player, rather than playing the game automatically.

I don’t know how online poker rooms go about detecting this kind of software, so they could possibly run into a problem here.

Gary Kasparov beaten by Deep Blue 1997

But can Deep Blue navigate the A&P?

LD: Chess Grandmaster Garry Kasparov lost to an AI called Deep Blue in the late 90s. Today, he encourages joint AI/human participation in chess games. Do you think this is the way forward to combat bot problems in online poker rooms?

JJ: If the bot problem cannot be solved otherwise, allowing teams of humans and AIs to compete might be required to continue online poker in a meaningful way. This would amount to unbanning software that offers real-time analysis and commentary.

LD: Will an AI one day win a WSOP bracelet? What are your thoughts about competing with bots in tournaments/cash games?

JJ: If the WSOP lets them compete I think an AI has a decent chance of winning. In general I’m excited about more advanced poker bots and think their introduction could open up new opportunities for live poker players.

Poker bots will at some point dominate human players so we might want to ban them to keep things interesting. Live poker could also go the Advanced Chess route, pitting teams of humans and AIs against each other.

We could also host tournaments just for poker bots, though these tournaments would be less about poker and more about who could design the best AI.


What happens next?

LD: Do you care what happens 100 years from now?

JJ: I care more about what happens 100 years from now than I do about current events. I strongly believe that lives matter equally, meaning that those existing in the future count just as much as those existing now.

The amount of humans that could exist in the future is orders of magnitude greater than those alive today. There are simply more future people to help than present people.

LD: Describe your emotional response when you think about the advancement of AI.

JJ: I feel anxiety and excitement. An AI that cares about sentient beings could literally lead us into an age free from suffering. An AI that doesn’t could create astronomical amounts of suffering and end civilization.

I try to be very careful about attaching moral weight to intelligence. Intelligence is a tool like any other. Humans have used this tool to eradicate polio and smallpox, but we’ve also used it to develop agent orange and nuclear weapons.

An AI could be used for similarly noble (or ignoble) ends. Many futurists have a false belief that everything will be ok, that if we just develop more technology and better AI, then the future will just work itself out. I never assume the best and am always prepared for the worst.