There is an old maxim in the world of artificial intelligence: The smaller the domain, the smarter the A.I.
Translated, it means that artificial intelligence looks a lot more intelligent if you give it less to think about. As an example, if I have a “HELP” app that can answer questions about how to screw in a light bulb, it’s going to look absolutely brilliant (no pun intended) because there’s virtually no question it can’t interpret and handle, at least when it comes to changing light bulbs.
On the other hand, an app that purports to have a conversation about your symptoms and then give you a medical diagnosis is a tougher challenge, because the subject is so broad.
In both cases, the A.I. processor is dealing with “perfect information.” Everything about the situation is known, nothing is withheld, and the only task at hand is to bring together all the available data and interpret them according to a set of fixed rules.
Chess is another “perfect information” scenario, because there is nothing whatsoever hidden. Everything about the status of a given game is known. All you have to do is look at the pieces on the board. While it’s true that there is an advantage to understanding something about your opponent’s style of play, if the chess-paying machine is powerful enough, that doesn’t matter, because the machine’s combination of advanced algorithms and brute-force ability to consider millions of moves is just too overwhelming. Which is why, at least when it comes to chess, the human-vs-machine debate has ended. (Spoiler alert: The humans lost.)
Watson, the Jeopardy-playing computer from IBM, had it a little tougher. Jeopardy is a kind of reverse Trivia, in which the answer is given and the player has to figure out the question. That means the player, whether machine or human, has to deal with the brain-scrambling nuances of language before it can even start to look up specific information. It’s an incredibly difficult exercise in semantic analysis.
But, once again, there’s nothing hidden. The way a clue is phrased is unambiguous; all the information is on the table.
Which brings us to poker.
Poker is different. Not only does it involve information that exists but is not available, there are actually three separate classes of hidden data. The first is all of the cards you can’t see but your opponents can, because they’re in their hands. Then there are the ones that no one can see because they have yet to be dealt.
Then there’s the big one, which is what’s inside your opponents’ heads.
What makes poker truly different from most other A.I. challenges is that it’s a game of deception. If it were just a matter of throwing some money in the middle of the table and seeing who had the best hand, it wouldn’t be much of a game. In fact, it’d be no different from coin tossing or dice rolling.
Instead, betting on each turn of the cards allows players to use a wide varieties of strategies to deceive the other players. This is known as “bluffing.” The most common bluff is to bet big when you have lousy cards so your opponents fold and hand over the pot. But if I’ve got great cards, I want you to think I have a lousy hand so you’ll stay in the game and keep throwing money in the pot. If I don’t yet have a good hand and am banking on making it better, I might want to slow down the betting until I see what I end up with, or actually boost the betting to either drive out others who don’t have good hands yet or get more money into the pot in case I hit.
A really accomplished player might even deliberately lose a few small pots in order to make the others think he’s a bad player, or stay in with bad hands so the winner is forced to show his hand and provide clues as to his style. To make it even worse, a player who stays to the end of the betting but loses doesn’t have to show his hand to the other players, depriving them of extremely valuable information as to the player’s style. And worst of all: Most smart players constantly vary their styles of play, making it very difficult to tell what they’re holding.
The net of it is, poker is a bottomless pit of missing and misleading information. Build a computer that can deal with that and you’ve got my attention.
Turns out, somebody might have. There’s a match going on between four top professional players and a poker-playing computer from Carnegie Mellon University called Libratus.
What makes this such an exciting concept is that the computer not only has to try to figure out if and when it’s being deceived, it has to purposely try to deceive the other players. This elevates artificial intelligence to an entirely new level of sophistication that comes close to actually earning the moniker “intelligence” as opposed to brute-force, algorithmic processor pounding.
As of right now, Libratus is kicking everyone’s butt. Badly.
However, there’s a catch. It seems that the machine’s behavior is changing radically from day to day, leading some to speculate that its human overlords are tweaking its parameters during the overnight breaks. The Carnegie Mellon professor overseeing Libratus has declined to comment, which is pretty much a sure a sign that the machine is getting human assistance.
Some might find that comforting. I find it only a temporary glitch. It’s just a matter of time until the machine is taught to understand how the tweaking is being done and do it itself. Stay tuned…
Lee Gruenfeld is a Principal with the TechPar Group in New York, a boutique consulting firm consisting exclusively of former C-level executives and "Big Four" partners. He was Vice President of Strategic Initiatives for Support.com, Senior Vice President and General Manager of a SaaS division he created for a technology company in Las Vegas, national head of professional services for computing pioneer Tymshare, and a Partner in the management consulting practice of Deloitte in New York and Los Angeles. Lee is also the award-winning author of fourteen critically-acclaimed, best-selling works of fiction and non-fiction. For more of his reports — Click Here Now.
© 2021 Newsmax. All rights reserved.