DeepMind 5, Humanity 0

Hello all! Fear not: despite some extended breaks, I’m still in the game of reporting anything I hear regarding Chung Kuo, Roads to Moscow, or anything else from the pen of Mr. David Wingrove. A couple of days ago, when I saw in the news that a computer had finally beaten a human champion at Go (otherwise known to us Chung Kuo fans as wei chi), I made a mental note to post about here. And then, out of nowhere, a post appears from David in my inbox.

Without any further ado, here are some reflections from Mr. Wingrove about the significance of the event and its repercussions on AI in general. Full text after the break. And hopefully more news soon!

-Matt

_________

Anyone who has read CHUNG KUO over the years will understand the significance within the work of the Chinese game of Wei Chi (known in Japan and more commonly in the West as Go), the most ancient and probably the world’s most complex and difficult game.

As Major DeVore says in Chapter Twenty Eight of The Middle Kingdom, “A Game Of Static Patterns”

“He glanced at the machine again. It was a complex game, and he prided himself on a certain mastery of it. Strange, though, how much it spoke of the difference between East and West. At least, of the old West, hidden beneath the levels of the Han city, the layers of Han culture and Han history. The games of the West had been played on similar boards to those of the East, only the West played between the lines, not on the intersecting points. And the games of the West had been flexible, each individual piece given breath, allowed to move, as though each had an independent life.  That was not so in Wei Chi. In Wei Chi once a piece was placed it remained, unless it was surrounded and its ‘breath’ taken from it. It was a game of static patterns; patterns built patiently over hours or days – sometimes even months. A game where the point was not to eliminate but to enclose.

East and West – they were the inverse of each other. Forever alien.”

The quality – or suzhi – of certain characters in the books is often measured by their mastery of (or failure to master) the 19 by 19-space Wei Chi board. And for good reason. Since we have entered the age of computers – that is, effectively since the 1950s onward – no one had managed to come up with a programme which came even close to defeating the greatest Masters of the game. Indeed, it was believed that it would be a long time yet – estimates were given of ten to thirty years – before we finally broke through that barrier.

Only… yesterday, Nature magazine published an article called “Mastering the game of Go with deep neural networks and tree search”, claiming to have done just that. Just as Gary Kasparov lost to DeepBlue in 1997, so, it seems, Fan Hui – European Wei Chi Master – was beaten in five straight games by the latest challenger to the ‘Go’ crown, DeepMind, using a new software called AlphaGo.

Fan Hui, thrice-times winner of the European Go championships is not the world’s best player. That accolade is shared between China’s Gu Li and South Korea’s Lee Sedol, who have shared a rivalry on the Wei Chi board these last few years, having faced each other on at least 48 different occasions. It is Lee Sedol who will play DeepMind this coming March in what will be the ultimate man versus machine encounter.

Okay. So just how important is all of this?

I’d like to claim that it is very important indeed. In fact, I’d go so far as to say that the age of Artificial Intelligence is just about to begin – this coming March. For to have a machine defeat the finest Go minds of our time is really something else. That is assuming that it does, for Lee Sedol is a mighty 9 dan Master, whereas Fan Hui is a mere 2 dan. But I think the machine will win. My instinct is that the way it’s been programmed will enable it to defeat its human foe.

And then?

Well, and then we need to be careful. Very careful indeed. For we’re talking super-intelligence here: of the kind which has no moral basis. And who knows where that will lead?

I still play go on my computer. A programme called COSUMI, which costs nothing to play. I give myself a handicap and take my time, and usually beat it, but we’re not at a very challenging level and I tend to give up if I know I’m going to lose. I’m certainly no Tuan Ti Fo!

Which reminds me of an experience I had over thirty years ago now. It was late evening and my dear friend Rob Carter decided I needed to join a go club – one situated in London’s Covent Garden. I had a few games there against guys – for it was all male, I recall – and won them comfortably. Probably because Rob and I, at this stage, were spending whole evenings – two times, three times a week – playing go at our local pub.

Then a Chinese guy arrived. He watched the end of one of my games and asked me what dan I was. I smiled, enjoying the quite obvious irony. Game ended, he sat down across from me. Did I want a handicap? I considered then shook my head. He looked the part, but I had already sampled the standard of this club. And so we began.

Less than five minutes later he stopped the game. “That’s it,” he said in a no-nonsense fashion. “You’ve lost.”

I made to protest. There were barely 17 of my white stones on the board, and while I knew I’d made some poor choices, I didn’t think I was beaten yet. But he knew otherwise. Picking up both my stones and his own, using a strangely Han way of doing it, scooping them with a single finger into his palm, he showed me what I’d done wrong and why it was wrong, placing the stones back down from memory. That alone – the memorising of 34 stones – impressed the shit out of me, but the way he projected the play on was even more impressive.

He was, I understood from what I was told later on, a championship player with a low dan rating. A Master, if a little one. Years later, when writing CHUNG KUO, he would often come into my mind, whenever I approached the subject. Tuan Ti Fo was an extension – an exaggeration – of that un-named go player. Once glimpsed never forgotten.

Okay. That’s it for now. It would be great to hear what you make of DeepMind and the AI future. Oh, and here’s a silent toast to Marvin Minsky, master of AI studies, who died on the twenty fourth of this month.

More news next time out. And there is news. Exciting news, in fact. But enough for now.

David Wingrove

Friday 29th January 2016

11 thoughts on “DeepMind 5, Humanity 0”

  1. Thanks for mentioning this! I had no idea! I had thought that with the billions of variations branching off of a single early move in a match, we would need a quantum computer to defeat the world’s strongest players. I play Go online on flyordie, kgs, and GoPanda. What is your online rank?

    Oh, and i am definitely going to watch this man-vs-machine match online this coming March… if the machine wins, i’ll be watching a turning point in history in real time :O

  2. Hi, thanks for the update.

    Regarding this AI, I’m not sure what to make of it, tbh. I’d like to think humans can always do it all on their own, but it could be the case that an artificial super intelligence could be the only to way to overcome some of problems in physics.

    Humans may not be able to figure out a way of reaching the stars, an artificial super intelligence might.

    Humans may not be able to figure out a way of discovering immortality, an artificial super intelligence might.

    The possibilities are limitless.

  3. The power of artificial neural networks is impressive. I have been to a few conferences in the last couple years where I’ve seen some demos of some of the things they can do in the manufacturing world (specifically prediction of heat deformation caused by the welding process on metal structures). The accuracy and stability of the results is pretty good, but the speed at which it can work is amazing relative to the “traditional” finite element analysis methods.

  4. Hi. I read a year or so back that Mr Wingrove intended to publish the remaining volumes of the mighty epic independently, but have neither seen nor heard anything since. Does anyone out there know how things stand? I know a great number of fans would like to read the whole epic in its revised form….

    1. Hi John. My understanding is that that’s still his plan, but wrangling the rights from a publishers is a tall order, not to mention that independently publishing is tough, too. As soon as I here something, I’ll post it here right away.

      1. I’m not familiar with how the publishing industry works, but are you saying the publisher does not want to release the remaining books, whilst at the same time not want to let go of the rights to Chung Kuo?

        That seems ludicrous to me. Either put up or shut up would be my message to whoever (Corvus?) this publisher is.

        1. Yes, that appears to be the case. I agree that it’s ludicrous, but if they can generate revenue by reverting the rights back to DW, that’s an easy way for them to make money without printing a single new page. Keep in mind that this is me reading between the lines, but if it were as easy as we think it should be, then we’d have the rest of our story already.

Leave a Reply

Your email address will not be published.