There's a moral in here somewhere. Actually, I'm rather afraid to look for it. In 2017 Google's AlphaGo Zero challenged its predecessor to a world championship match of Go. Poor old Dad; the final score was 100-nil. This whitewash had nothing to do with increased processing grunt or slicker componentry. The key to AlphaGo Zero's victory lay in a simplified algorithm, which bypassed human training altogether. “We’ve actually removed the constraints of human knowledge and it’s able... to create knowledge itself from first principles, from a blank slate,” said lead researcher David Silver.
In order to understand the significance of this event, which some are calling a 'singularity', we need to go way way back into the distant past: 1995. This was the year that a software programme called Chinook successfully defended its world championship status at checkers/draughts. It came a shock to most people, however there was some consolation to be had: all of Chinook's smarts had been programmed into it by humans. It had been given a huge library of moves plus acres of code to confer the ability to calculate benefits. Two years later the game had changed. IBM’s Deep Blue beat grand-master Garry Kasparov at chess. This time humanity had no wiggle room - artificial intelligence was the key to our defeat.
There are about 10¹²³ possible games of Chess. If that number means nothing to you, it also happens to be the same number as the estimated amount of informational bits in the known universe. In other words, in computing terms this is as big as big can be. When numbers reach this kind of scale,s computers can't simply run through a series of options and choose the best one. They need smarts.
The next step was machine learning. Beating humanity at Go was always going to be a bigger challenge than beating us at chess. Our best estimates put the possible number of games of Go at about 10⁷⁰⁰. This makes chess look like Tic Tac Toe. It is a game that must be played strategically. And that's what we saw in 2016, when Alpha Go senior beat the world's greatest Go player, Lee Sedol, by a margin of 4–1.
Ah, 2016; those were the good old days. Back then humanity's best and brightest could still claim a single match. From now on, I suggest you bet on the robots. Because in 2017 AlphaGo begat Alpha Go Zero. The difference between the two wasn't one of processing power, it was a matter of simplification. AlphaGo had been fed a huge number of games to learn from. AlphaGo Zero was given the rules and left to learn the rest for itself. And so it did. It played itself 4.9 million times in three days, and then took on its elder. The result wasn't even close. The final score was 100-nil.
That's game over, fellow humans. After only twelve years of board games we've been well and truly left behind. We can argue all we like about the definitions of artificial intelligence and machine learning, but who are we to decide? Still, at least it has been a bloodless affair. Just, please... NOBODY challenge these guys to an arm wrestle!