Wednesday, November 8, 2017

"The Australian": Review of "Deep Thinking"

Garry Kasparov vs. Deep Blue: Chess Match Lost to Technology
The Australian
12:00AM 
November 4, 2017
PAUL BROEKHUYSE 
Reporter Sydney 

Twenty years ago Garry Kasparov lost a six-game chess match to an IBM computer called Deep Blue. 

It was a memorable moment, as much for the Russian grandmaster’s hysteria in defeat as for being an apparent milestone in computing and the end of an era of human intellectual dominance. 

Now the former world champion has written Deep Thinking, a detailed, behind the scenes account of the encounter and an interesting overview of the relentless charge of technology generally. 

The book is timely not just for the 20-year anniversary. It comes at a time of increasing euphoria/terror over technology’s invasion of even white-collar workplaces, the rise of robots, autonomous vehicles and no doubt other scary things we haven’t even heard about yet. 

Kasparov details the early years of Deep Blue’s development, including matches he won against it (which no one remembers, of course). 

He says what started as a scientific experiment, an attempt to emulate human intelligence in silico, eventually morphed into a single-minded and sometimes nasty effort to beat him and boost IBM’s profile (and stock price). 

Initial efforts to create a human-like intelligence through pattern recognition, rules of thumb and something like “understanding” were soon abandoned when it became clear that “brute force” — faster and faster computers to evaluate vaster and vaster numbers of moves — simply worked better. 

There was a price to pay. Such calculating monsters could “evaluate” positions by adding up how many pieces each side had or measuring the territory they controlled but had no real grasp of chess and told us nothing about human cognition, Kasparov laments. 

He was bitterly critical of IBM during the match but attracted little sympathy. Reading this more detailed account from a self-confessed poor loser, it’s easier to understand his whining. The champion was denied copies of Deep Blue’s games before the contest, even though its programmers had access to a lifetime of his moves. Grandmasters typically study their opponent’s previous games before a match, looking for weaknesses to exploit, so this information vacuum was a serious disadvantage. And it was even worse with Deep Blue, since last-minute programming changes could effectively turn the machine into a different opponent. 

During the match, Kasparov also was denied access to logs of Deep Blue’s calculations, and occasional glitches meant the computer had to be reset, altering its thought processes. 

Arcane complaints, perhaps, but other revelations are more disturbing. 

Miguel Illescas, one of several grandmasters secretly recruited by IBM to help with programming Deep Blue, later revealed how the IT company had played the man and not just the board. 

In particular, the length of time Deep Blue took over its moves was adjusted to unnerve Kasparov. 

Instead of spending a standard three minutes on a move, it was programmed to occasionally take much longer, to make its nervous human opponent wonder what it was up to. 

“This has a psychological impact, as the machine becomes unpredictable, which was our main goal,” Illescas later admitted. That’s gamesmanship worthy of a human and it certainly rattled Kasparov. 

But the human couldn’t return the psychological favour to his unflappable silicon opponent. 

Artificial intelligence seemed to be morphing into artificial bastardry. 

Kasparov also questions IBM’s haste in dismantling Deep Blue after the match (it now sits idle in not one but two U.S. museums). If Deep Blue had fairly won the greatest intellectual contest in history, why not reveal its electronic entrails for the edification of all of us? 

The match itself was very interesting and not as lopsided as some believe. Kasparov won the first game in fine style. 

In the second, the computer played well and achieved a dominant position, but blundered badly towards the end, giving Kasparov a chance to escape. Instead, a demoralised Kasparov resigned in a drawn position. 

The next three were drawn, though the human came close to winning a couple. 

Then came the final game, undoubtedly the worst of Kasparov’s illustrious career. He played a passive opening, ostensibly part of his positional, anti-computer strategy, but blundered horribly on move seven. Deep Blue responded immediately with a standard sacrifice, something even an average club player might know. The move completely paralysed Kasparov’s position and he resigned soon afterwards. 

Technology 1, Humanity 0. 

Much has been made of Kasparov’s blunder. 

Some say it was a simple mistake caused by the extreme stress of the occasion. It’s possible he simply got the order of moves wrong (his error on move seven is actually the standard play on move eight). But Kasparov has always been an expert on openings, so this seems unlikely. 

Also unlikely is Kasparov’s own explanation. He says he reasoned that computers like to hang on to their pieces and are reluctant to play speculative sacrifices, which is true. So Kasparov claims he reckoned that Deep Blue would baulk at playing the decisive sacrifice and meekly retreat his knight, giving Kasparov easy equality. 

That’s just bizarre: a world champion playing a beginner’s blunder in the hope a computer would wimp out in a weird game of electronic chicken. 

And the background is even more bizarre. It turns out Deep Blue’s programmers specifically instructed the machine to make the sacrifice (by changing its so-called “opening book”) shortly before the match began — though whether it was the day before, or a few weeks, is disputed. 

And Illescas admitted that IBM was spying on Kasparov: knowing that he spoke with his assistants after games, they replaced his security guard with one who spoke Russian. 

“I knew what they spoke about after the game,” he says. 

Illescas says this wasn’t important, a mere anecdote, but clearly it isn’t, and Kasparov is right to be appalled.  

Ironically, computer databases and “engines” are now a standard part of the grandmaster’s toolkit, and Kasparov spends the rest of the book considering the role of technology in chess, and more broadly. His musings are surprisingly sunny, but not revolutionary. 

He’s optimistic about the future, while sympathising with those who will lose their jobs. Mercifully he doesn’t take complacent refuge in the lazy mantra about new jobs replacing the old, understanding full well that the lift operators who lost their jobs in 1920s America probably didn’t turn around and become electrical engineers working for Otis. Instead, they and their families simply suffered. 

There will be progress but there will be pain, and he rejects attempts to disguise or sidestep this. The world, says Kasparov, “needs new jobs to build the future instead of trying to bring back jobs from the past”. Indeed. Kasparov knows better than most what it’s like to lose your job to a machine. 

Paul Broekhuyse is The Australian’s chess columnist.