Games that challenge the human brain – and teach computers to think

273

Lucinda Robb

THE WASHINGTON POST – Thanks to the pandemic, everyone is playing more games than ever – and not just for entertainment. Something else compels us to test our skills with games. Or to put it another way, why can’t we stop playing Wordle?

You might find some clues in Oliver Roeder’s illuminating new book, Seven Games: A Human History. Roeder, a senior writer and puzzle editor at FiveThirtyEight, has written a fascinating group biography of some of the most popular games of all time. In doing so, he offers powerful insights into why we play games and what we can learn from them.

You’ve probably played at least one of the seven games profiled: checkers, chess, Go, backgammon, poker, Scrabble and bridge. Roeder chose them in part because of their longevity. An early version of checkers is mentioned in Plato’s Republic, and a prototype of backgammon was found in a thirrd-century Nubian tomb (by comparison Scrabble, invented during the Great Depression, is the new kid on the block).

Each game gets its own biographical chapter that digs into what makes it unique. Also included is a rough sketch of the rules of play and the skill sets needed to win. We’re introduced to legends of the games like the humble and devout Marion Tinsley, who lost only three times in a 40-year period of playing competitive checkers. Roeder himself competes in the World Series of Poker and the Scrabble Players Championship, for which he memorises nearly 35,000 words.

The real payoff from the group biography concept – the connective tissue that ties the book together – is the way Roeder illustrates the impact these games have had on artificial intelligence. This gets to the heart of what makes games so valuable, both for humans and for computers.

Games allow humans to make decisions and experience agency in a limited context. As Rice University philosopher Gwen Bradford observes, “Every time you play a game you are choosing to do something that is more difficult than it has to be.” When we strive and come up with solutions, it gives us an important sense of achievement (hence, the appeal of Wordle).

More specifically, games are made up of arbitrarily imposed rules. Following those rules allows us to practice making choices and solving problems in a controlled environment. Which also makes them a perfect tool for teaching computers how to think.

Roeder has arranged the chapters in his book according to “a rough menu of intelligence” that correlates – broadly – to the difficulty computer programs have in mastering each game. This ranking is based on the complexity of each game, the range of skills required and the element of chance.

Thus checkers, the most straightforward game and nominally the “easiest” (but deeper than it gets credit for) comes first, whereas Scrabble, which Roeder memorably describes as a sort of “brainy heptathlon”, is sixth.

Rather surprisingly, the famously cerebral game of chess is only second. Although endlessly complicated for humans, chess is not a terrible challenge for a computer program with enough processing speed. Armed only with the rules of the game and no other human input, AlphaZero, the best chess program in the world, played itself 44 million times and discovered on its own the Queen’s Gambit, the English Opening and the Sicilian Defense.

The brilliance of the top-ranked human chess player in the world, Magnus Carlsen, pales beside computer-generated competitors. More than 90 computer engines are ranked higher than Carlsen. Check out Top Chess Engine Championship to watch elite computer programs play at a level that humans have never matched.

Because they are completely governed by rules, checkers, chess and Go can all be mastered by computers. But that doesn’t mean they are simple. Checkers alone has 500,995,484,682,338,627,639 different positions. The number of positions for the infinitely more complex Go is so immense that it has 171 digits. For comparison, Roeder tells us that the number of atoms in the entire universe is a number that is 80 digits long.

Yet as daunting as these numbers are, adding an element of chance is the real game changer (pardon the pun). Computers lose their advantage over humans when they play a completely random game like rock, paper, scissors.

In backgammon, poker and Scrabble, the roll of the dice, or the draw of a card or a tile, introduces a level of luck that makes it exponentially harder for humans or computers to predict the next move. It is simply not feasible (for now) to use raw computing power to muscle through enormous decision tree branches and pick the best options. Machines can tackle aspects of the game, but they can’t solve it.

This forces scientists to find new ways to harness computing’s tremendous potential. For backgammon, IBM researcher Gerald Tesauro created a neural networking program that looked instead for patterns, mimicking human learning. Not only did TD-Gammon, as he named his program, eventually dominate the backgammon scene, it would go on to yield real-world benefits in a variety of other applications such as elevator traffic in tall buildings and job scheduling for the space shuttle.

As computers improved their play, so did humans, who now study their games and learn from commercial and free programs such as eXtreme Gammon, PokerSnowie and Quackle. This has helped to democratise the games. You no longer have to live in certain locations, or be able to afford to travel, to play and study a game at the highest levels. You can just go online or consult an app on your phone. This is particularly true of poker, where the wide availability of programs known as solvers helps a new generation of would-be pros pursue optimal betting strategies.

But there may be downsides. The colourful, larger-than life poker players with big hats and jewelry are slowly being replaced by guys in sunglasses and headphones who don’t interact with anyone at the table. They may be more successful, but they aren’t fun. Traditionalists fear that if the social aspect of the game fades, so will the enthusiasm of the casual gamblers who trust their luck and lose reliably enough to bankroll the winners. What the long-term impact will be is hard to tell, but gambling is absolutely essential to poker. Think about it: If people didn’t bet in Texas hold ‘em, then why in the world would anyone play?

Bridge, the last game and the one Roeder describes as “most ‘human,’” is perhaps the most complex from a skill set perspective (it even has its own special languages for bidding). It is also the lone game of the seven where artificial-intelligence programs still trail human players. While its popularity has fallen dramatically from its heyday in the 1930s and ‘40s, being good at bridge is still viewed by many on Wall Street as a predictor of real-world success.