As Professor Tuomas Sandholm high fives his young PhD student Noam Brown are we witnessing a Franz Ferdinand moment?
There’s no gun.
Nobody is dead.
But after their Artificial Intelligence (AI), Libratus beat four human players over 120,000 hands of Heads-Up No-Limit Hold'em; I can smell the stench of cordite 4,000 miles away.
Jason Les (-880,097), Jimmy Chou (-522,857), Daniel McAulay (-277,657), and Dong Kim (-85,649), were all soundly beaten by a machine that was simply too good for them. The final defeat was by a margin of 1,776,25 chips or 14 big blinds per 100 hands.
“We really got a beat-down,” Said McAuley after the defeat.
It's a landmark moment for AI. The machines had beaten the very best humans in Chess, Atari, and most recently Go. But a machine had never defeated a person in a game of incomplete information, until now.
Libratus was able to learn from its mistakes, daily. All four players said the machine got better throughout the 20-days with weaknesses falling away like the wings of a fly heading for one of those blue zappy things in a restaurant.
And therein lies the problem.
There is no ceiling to this thing.
The advancements made in AI, helped by this historic moment, will end very badly for the human race. Either AI will become so intelligent that they will destroy us in the same we destroy a convoy of ants that dare venture into our house. Or, we will use the AI to destroy ourselves.
People are wondering if this spells the end for online poker?
It could spell the end of the world.
The spectrum of intelligence is unimaginable. We have no conceivable idea what will happen when AI develops a level of intelligence that supersedes our paltry set of atoms. And the rate of progress is irrelevant. Someone has given Sisyphus the day off. The boulder is rolling. There is nobody to stop it.
I know what you are thinking.
What a fatalist.
And you might be right, that’s why I reached out to six professional poker players, and Berlin Bear Bill Perkins, to ask what they thought about Libratus and AI’s role in poker and beyond.
Will the humans win?
Before the match had begun, I asked the panel to pick a winner, and 70% of them thought the humans would have the slight edge over the AI. None of them thought there would be a landslide either way, with most panellists stating that the victory would be marginal.
"I thought the Humans would win and I still think irrespective of this first round the algorithm will have a flaw that is exploitable once deciphered," said the Global Poker Index Berlin Bear, Bill Perkins. "My hunch is that the program will not be able to vary enough the shifting of ranges and bet sizing of a human opponent. That's not to say a very strong AI player will not be built that would rank very high. I think one of the reasons the developers are tight-lipped about how the program works is that it is exploitable."
“I would expect the humans to win, as long as they are mentally adroit enough to try to exploit the patterns that they see emerge from the AI, while still not losing their grasp on the fundamentals in an effort to "fool the computer.” Said Niman ‘Samoleus’ Kenkre.
“I didn’t think AI had progressed enough,” Said the World Series of Poker (WSOP) bracelet winner, Andrew Barber.“ But that was pretty silly considering the pace of AI development and the famous picture predicting the Singularity (the point at which computers surpass humans in intelligence) in 10-15 years.”
Matt Ashton was the one member of the panel who never doubted that Libratus would win.
“The humans won’t win,” said the former WSOP Poker Player’s Championship winner, Matt Ashton. “I remember before the match thinking that the AI would be much improved compared to last year and thought it might be the favourite and was surprised people were laying prices of over 3-1 against the AI on 2+2 and Pokershares.”
Ashton is talking about the match between AI Claudico and four human players that took place in 2015. The humans won that game by 7,300 big blinds, but the research team said they had learned enough to make the changes necessary to defeat humans.
They were telling the truth.
Do you care if AI wins and solves NLHE?
57% of the panel were worried about a Libratus victory, but as both Terrence Chan and Brian Rast point out, the success doesn't mean that NLHE is solved. It just means that AI has beaten humans in a Heads-Up NLHE match.
“Winning does not mean solving NLHE,” Said poker pro and MMA fighter, Terrence Chan. “A true HUNLHE equilibrium solution is still quite far off, and I'm sure the people at Carnegie-Mellon would admit that.”
"There's a difference between AI winning and solving NLHE," said the multiple WSOP bracelet winner, Brian Rast. "Also, please note that it's HU NLHE and not NLHE in general. The AI hasn't "solved" HU NLHE."
Bryan Paris has won $10m playing online poker, so he does care that Libratus has won.
“Of course I care,” said Paris.“It could make my job obsolete.”
Brian Rast also fears for the future of online poker.
“I already believed that online poker’s days were numbered, and this is just reinforcing my belief.” Said Rast.
Matt Ashton and Chan think there was always an air of inevitability about the growth of AI and the future of online poker.
“Humans beating AI here might give more confidence of slightly longer longevity for online poker, but it's felt inevitable for a while that it would happen.” Said Ashton.
“While it's relevant to me as someone in poker, computers being better than humans is an inevitability, so whether it happens now, next year, or in 2020 is not hugely relevant to me.” Said Chan.
Andrew Barber believes that the AI victory has much broader implications than just the profession of an online poker player.
“I wanted it to be a harder game to solve for selfish reasons, but I want to use this opportunity as a teaching moment,” Said Barber. “If anything, this shows that computers are coming for a number of jobs requiring decent levels of intelligence, so we need to start having a discussion of what we do with those who are displaced. It’s looking quite likely that there won’t be enough jobs to go around.”
I agree with Barber.
According to the Bureau of Labor Statistics, there were 1.6 million truck drivers in America in 2014, more than half of a percent of the nation, with the earning power of 0.3% of US GDP.
How long before humans no longer operate trucks, planes, ships, coaches, trains?
If AI wins, is online poker as a profession dead?
Of the panellists, Brian Rast, was the only one to answer definitively that the end of online poker is nigh, although Rast points out that he believes online poker's troubles began long before Libratus came along.
“Online poker is dying, but this was already happening,” Said Rast. “This is a public landmark of a transitional process that is already underway.”
Several panellists believe that Heads-Up online poker is in trouble, but in the whole, the enterprise will survive.
"At the very least, high stakes will certainly die," said Bryan Paris.
“Forms of Heads-Up, yes,” said Andrew Barber. “I believe that three or more player games are unsolvable, but I could be mistaken.”
The ‘is online poker dead?' question relies heavily on the online poker rooms ability to prevent the proliferation of bots as a result of advancements in AI. A point that Niman Kenkre makes well.
"It depends on how well the sites can detect and police bots using their sites." Said Niman Kenkre. "Some countermeasures would not be too difficult to implement if the sites valued their integrity in this matter. But it certainly wouldn't help. Obviously, online poker is not flourishing as it was some 7-8 years ago, so if AI became more proficient, it would be another blow though not necessarily terminal."
Both Terrence Chan and Matt Ashton believe there is life in the old dog, yet.
“Absolutely not,” replied Terrence Chan when asked if online poker was dead. “There are games other than 200bb HUNL. Even looking back to when the University of Alberta solved HU LHE completely, people still play LHE online.”
"Certainly not just yet," said Matt Ashton. "I also think multiplayer mix games are likely to be difficult for AI to overcome for a long time and maybe not worth the processing power it would take, that said how much action there will be is another question. I think games will likely continue at a particular stake that people are not concerned enough about bots existing or losing money to them, in the games to stop playing, and I don't think it's impossible a few people can make a living in those games. But it seems likely there will be very little room for the profession in future."
Bill Perkins can see a world when humans will gladly compete with the bots, in the same way, that Chess Grandmaster Garry Kasparov turned his attention to creating human/AI partnerships in Chess after his loss to the AI Deep Blue.
"I think there are bots already online that are better than average and can play ring games and do well with human supervision." Said Perkins. "I don't think online poker is dead though it will change as the proliferation of stronger and stronger AI "solves" poker. Perhaps people will knowingly play the bots, maybe the value of being a pro will go way down in the online world as amateurs get real-time feedback on how to play. Online poker will transform."
Will an AI one day win a WSOP bracelet?
With the World Series of Poker (WSOP) recently announcing three online bracelet events at the next series, coupled with Bill Perkins view that we could one day see humans willingly playing bots, could we see AI winning a bracelet?
The overwhelming response was ‘no,' although there were some interesting comments regarding the use of AI in live tournaments.
"An AI could easily win a WSOP bracelet if a live player were allowed to use it during play," Said Paris. "Maybe if Google glasses become powerful enough? Shouldn't take too many years."
“I would bet a lot of money against an AI even being allowed to play in the WSOP before say, 2025," said Terrence Chan. "Poker for the conceivable future will likely continue the way it has right now -- the brightest and most intelligent players will use software tools away from the table to continue to learn more about poker so that they can implement even more advanced strategies at live and online tables. Since computers aren't allowed at the table, the best players will be the ones who are best able to integrate the knowledge gained from the software into their squishy human brains."
Perkins gives some good points on the advancement of ‘computer assisted' play.
"I don't think they will let Bots play at the WSOP, but what we are seeing is computer assisted play," said Bill Perkins. "For example, after each hand, you could look at a Snap Shove Chart for your position to help you make decisions for the next hand….the next step is for Snap Shove Charts to incorporate others chip stacks and historical ranges for your guess of them to give you your call fold range, etc.
"Pros are already playing great standard poker adjusting for others ranges and moves…right now the beginner can play just as good as a pro with stacks below 15bb in a tournament using Snap Shove Charts, which is allowed while not in the hand. Successive iterations of this will further integrate human-machine collaboration.
"I imagine one day an app that lets you take notes very easily of the play at your table or you voice note them. The voice note is converted into data, and then during a break or after each hand you have updated charts on what to do player specific that are easy for the human to memorise."
Niman Kenkre likes the idea of facing off against a bot.
"I would have no issues competing against a bot," said Niman Kenkre, "I think it would be kind of cool. It's interesting to reflect on how to adjust and optimise for a bot."
Matt Ashton would prefer to see the bots facing other bots.
"I don't like the idea of competing vs. bots," said Matt Ashton. "I think the bots should just have their separate competitions and let the humans with their inferior ability battle it out between themselves."
The Libratus creator said that NLHE was the ‘last frontier’ for AI vs. Humans in game theory. The whole purpose of creating Libratus is to design a general intelligence AI that can be thrust into the world. Are you worried about this?
Moving beyond poker, how many of our panel are, like me, worried about the impact that AI will have on the world as a whole in an existential way. It seems they were a divided bunch with an even split between those who are worried, those who aren't, and those who can't decide.
Terrence Chan prefers to focus on the good that AI can do in the world.
"There's always some measure of worry that the robots will take over the world and kill us all, Terminator-style. But the history of human civilisation is that technology has improved humanity, in spite of virtually every significant technological advance being seen (at that time) to have huge potential negative consequences," Said Terrence Chan. "The printing press, the steam engine, the car, nuclear fission -- these things were all seen as potentially harmful and in many ways were, but I would argue that they have made humanity better overall."
Brian Rast has mixed feelings.
"The potential for more powerful AIs to solve problems is immense, and we should be happy about the possibilities this brings humanity," said Brain Rast. "But on the flip side, there are problems. The same power could be used for bad purposes (as with any powerful technology). And AI itself could present humanity with an existential crisis. There's a pretty popular singularity theory - the idea is that if and when AI becomes sentient, it will immediately become significantly more intelligent than humans, and could decide to wipe us out if we aren't careful with how this is done and how it's programmed."
And the doom mongers:
"I am very very worried about AI being thrust into the world for some of the obvious reasons and some of the not so obvious," said Bill Perkins. "When a new technology is introduced, it tends to improve lives, but the benefits accrue to the wealthy or super wealthy creating more and more wealth inequality.
“AI is one of the most rapid wealth inequality creators. Wealth inequality ultimately leads to violence, chaos, crime and instability. If you think of a poor person today in the USA versus King George III the poor person today is better of by many many metrics - perhaps all. What drives the crime and despair is the relative wealth.
"AI can radically shift wealth distribution. Let's say AI solves ageing for specific sets of people who get to live to 200 or 500 - that type of relative deprivation is catastrophic. This is the issue I have with the singularity and AI - many, many people die, or chaos happens as we approach the singularity such that we never get there or few survive and see it. Humans need to be proactive about this issue as we move to a society that will eliminate all the driving jobs, many of the manufacturing jobs, some of the programming jobs, etc. In the interim, how do these people obtain resources? Crime seems like a rational response in the current construct."
Matt Ashton and Andrew Barber are equally as concerned.
“I don't fully comprehend the potential dangers of AI, but in the short term I expect almost only positive things will come from AI improvements,” said Ashton. “At the same time, I can envisage it going too far eventually and AI maximising some functions that conflict with most people's current morality.”
“I am extremely worried,” said Andrew Barber. “There are many in Silicon Valley and elsewhere who are worried about the existential risk presented by AI.”
Do you care about what happens 100-years from now?
For most of us, we might not even be around to see humans devolving into apes as AI takes our crown.
I asked the panel if they cared about the fate of the world, 100 years from now, and the answer was unanimously affirmative.
“I do care what happens 100 years from now," said Bill Perkins. "There is a very real chance my children will be alive to enjoy or suffer what is going on in the world, and there is a remote chance I will be too.
“One of the biggest issues humans have is caring about other humans not directly in contact with them. It’s one of the main reasons so much suffering continues in the world. Since time and distance (relativity) it’s the same issue; the real question is do I care enough about what happens 100 years from now. The same way you would ask do I care enough about starving children or children in war zones.”
"Yes, but it's hard to project current problems forward," said Niman Kenkre. "I have a lot of confidence in the human mind and soul; I think that humans will continue to innovate to find solutions to problems that grow and arise in the future."
“Yes, I care quite a bit. It's one of the reasons that Global Warming is one of the most important issues to me.” Said Brian Rast.
Andrew Barber believes we have got bigger problems than thinking about the world in 100-years time.
“I think we need to worry about getting there first," said Andrew Barber. "People discount the future way too much. Some serious concerns on the horizon need addressing e.g. climate change and "killer AI"."
How does AI make the panel feel on an emotional level?
I asked the group what they felt when they thought about AI?
“I am an optimist, so more hope than fear, but I believe that it will dramatically transform society in ways that people can't imagine currently." Said Bryan Paris.
"The gut reaction might be negative, but overall I feel positive about it." Said Terrence Chan.
"I am both excited but more afraid," said Bill Perkins. "Humans tend to misuse technology and the new things they learn whether it be in explosives, biotech, chemical, physics, etc. AI has many, many wonderful applications but it is a tool much like a knife or nuclear power except it's much much more powerful and much much more far reaching."
"Trepidation and scepticism," said Andrew Barber.
Most believe gambling will still exist, albeit in a very different world.
“Online poker will certainly die off," Said Bryan Paris. "I assume people will still bet sports and go to casinos. Live poker may survive as long as they can keep people from taking powerful computers to the table with them. People will have more time to gamble in general as most jobs will be obsolete."
“Gambling is one of those things that exists because it hits very primal parts of our brain that we have trouble controlling,” said Terrence Chan. “In 50 years it's possible that we'll have brain implants where we can dial up or dial down our dopamine output, and we won't need gambling or sex or food to hit those pathways, but if not, gambling will still be around.”
“Maybe, eventually the AI will conclude gambling is a net-negative activity for the world and influence people to get rid of it.” Said Matt Ashton.
“The pit seems to be pretty immune to people getting smarter!” said Andrew Barber.
"I think that nobody will be playing poker online for any significant money," said Brian Rast. "It will be played more as chess is played today. I believe that poker will still be played live, and in tournaments, people will not be allowed to use helping technology.
"I think that "gambling" outside of poker will still exist in the more traditional forms, such as casinos and softer forms such as slots, because AI has no effect on people deciding to bet their money in -EV games. That's just bad human decision-making which will always exist." Said Brian Rast.
Bad human decision making?
I can’t help but think of Brian’s last words.
Is this what Libratus is?
Has the team at Carnegie Mellon made a bad decision?
It was Sam Harris that reminded me that electronic circuits function a million times faster than biochemical ones, meaning AI can ram in 20,000 years of ‘work’ every week. How can we even begin to fathom what this means to the human race?
But I want to thank my seven panellists for giving it their best shot.
Bryan Paris, Terrence Chan, Bill Perkins, Niman Kenkre, Matt Ashton, Andrew Barber, and Brian Rast.