Deep Blue was an Idiot

That titanic battle with the machines, the meat vs. the AIs, the fight to determine the future of intelligent life, the war we’ve been awaiting…

Well, you missed it. It happened back in 1997, back when smartphones were still CD Walkmans and the internet was all geocities. In ’97 we chose a Russian genius as the champion of humanity. But he lost, to a computer named Deep Blue.

From 1986, until his retirement in 2006, Garry Kasparov was the World Chess Champion, the peak human performer in pattern recognition, abstraction and interpolation. There have been another seven contenders for the WCC crown since Kasparov retired, but, seriously, post-Deep Blue, does anyone really care?

Happily, Deep Blue, the machine that defeated Kasparov, did not go on to conquer the human race. Deep Blue was an idiot. It was, at best, a savant. Not an intelligence. It had been designed to beat a human at chess, and nothing else. In fact, it was built specifically to beat one particular human, Kasparov, at this one particular test of skill. Away from the chessboard it couldn’t even tie a pair of shoelaces.

Yet still, a man being defeated by a creature not even capable of drooling was hugely significant. Mainly because this wasn’t a contest of machine skills, but human ones.

……

There is many a machine capable of out-performing a human with, say, a complex mathematical calculation. But can it make a soufflé? Or tell a ripping yarn? Or kick a football? Or appreciate a work of art? These are all uniquely human aptitudes. And chess, too, is one of these idiosyncratic skills.

You can’t just throw brute force mathematics at chess. A chess board four moves each into a game will show one of 288 billion possible positions by that point. Calculating the possibility tree for a full game is comparable to tracking the number of grains of sand on the planet. Even with the vastly accelerated calculation skills of today’s super-computers, this method is only realistic on geological timescales.

Chess skill is more about abstraction. It’s not just about analysing a position and looking for threats or opportunities. It’s looking for the threat of a threat. Or the threat of a threat of a threat. Etc… This kind of multi-level abstraction, meta on meta on meta, is very difficult for a machine, but is something we humans do all the time.

We have adapted this skill to cope with sensory overload. It’s how we process the world – we channel the sensory bombardment into conceptual chunks, so we can skip over them. This way our brains can see a face, not eyes ears and nose, not follicles and molecules, not atoms and electrons.

……

While Deep Blue might have only been capable of calculating 5 or 6 moves ahead (Kasparov, at the time, claimed he could do 15), it had other, non-human, tricks up its cold metal sleeve. It was programmed with every previous game of chess ever played at championship level, so it could make a statistical analysis of how likely any early move used in a previous match would eventually result in a victory. It bought a Big Data gun to a pattern abstraction knife fight.

In the previous match against Kasparov it had tried this method and failed. Kasparov beat the machine simply by accounting for the database and playing a move that would throw it. He chose an opening so rarely used in championship level chess there was only one record of it. He instantly took out a whole statistical arsenal, enabling him to dominate the game early. Kasparov swatted away this Artificial Artificial Intelligence (AAI) as easily as he could have beaten it at an arm wrestle.

But this was to be Kasparov’s last victory over the machine. In the next contest, the non-human data arsenal (combined with improvements to its more traditional computational powers) proved enough to give it the edge. In this highly-artificial combat, a highly-artificial AI scored a victory.

……

I was being deliberately facetious earlier when I wrote on the subject of AAI, dismissing it as the acceptable face of machine intelligence. AI development is not dead, nor has it wandered down a cul-de-sac. It has simply evolved beyond the fears of sci-fi. The focus now is not on developing consciounesses as such. Because, hey, what would be the use of that? Why would we want a machine prone to the chaos of emotion? Why would one ever create a lifeform we’d feel guilty about enslaving?

You might take a moment here to also ask these questions of whatever god you believe in.

What I facetiously call Artificial Artificial Intelligence is more kindly referred to as “Artificial Smartness“, and is the major current trend in AI. Non-general intelligences that are to, say, shopping, what Deep Blue is to chess. Or navigation. Or market analysis. Or medical diagnosis. Or truck driving. Or customer service. Or any form of data-processing-based intelligence you can imagine. These intelligences are already with us, and exceed human capabilities.

Once you’ve come to terms with how easily it may be to automate whatever craft or creativity it is you possess that gives your day job its economic value, next imagine taking these any of these specialised AAIs, these Artificial Smart-arses, and multiplying their capabilities by ten. Then a hundred. Then a thousand.

They will be no nearer to human, we will still be able to feel a superiority to them on that respect, but they will have powers beyond us in other ways. Alien superpowers.

……

The winner of Kasparov v Deep Blue, aka Human v Machine, was not the idiot Deep Blue. Even if it did beat the Russian at chess, the Russian, I would hope, might have been waiting for it in the car park. To settle matters with a fairer fight. The only real take-away from the titanic battle to determine the future of intelligent life, was that it was utterly pointless. That it didn’t need to be a battle. The smarter strategy was not to fight, but to learn to work together.

In 1997 the Internet was beginning to reach the homes of ordinary people. Mobile phones were taking off. Most everyone in the West had a home computer by then. The Kasparov v Deep Blue battle was really the end of an era, the dying gasp of a time when we might have considered computers as an enemy.

Our relationship with the machines since then has increasingly become a symbiotic co-dependency. They empower us with ever-more-mighty tools, and we help them past their social awkwardnesses and cognitive limitations with our incredible anthropic abstraction abilities. And this has served us both well.

Even in the field of chess, which we might reasonably have abandoned following the Kasparov v Deep Blue match, we have evolved. The worlds best chess players at the moment are no longer individuals, they are teams, human’s and AIs working in collaboration. Kasparov is a prime mover behind this, calling it “Cyborg Chess“, or sometimes “Centaur Chess”.

So while the sci-fi may still remain dystopic on the future of AI, it is the sci-fi that is struggling in its relevance, not AI research. We might continue to develop alien intelligences, without them ever superceding us. The cautionary tales have been heeded, and a path forward has been mapped that benefits us both.



posted September 22nd, 2016



«
Comments are closed. But you can still hurl abuse at me on twitter if it makes you happy.


Generative Art book cover Generative Art: A Practical Guide (Manning 2011)

The complete beginners guide to the philosophy and practice of using a programming language as an artistic tool. Foreword by Marius Watz.



Novelty Waves book cover Novelty Waves: A Short Book About Digital Art (Leanpub 2014)

Five long articles and two short stories exploring the current state of digital art. Foreword by Jeff Noon.

Name your own price.