Positions from AlphaGo's first win against Fan Hui

How AlphaGo Illustrates the “Warm Bath And Ice Bucket” View of Technology Progress

Positions from AlphaGo's first win against Fan Hui

Remember the last time you took a bath or shower and it started lukewarm but you gradually warmed it by adding more hot water, until it reached a temperature so hot you could never have got into it at the start? Isn’t it strange how we can be immune to subtle, slow changes all around us?

Then there’s the other extreme – the ice bucket experience, where you’re abruptly plunged into something so dramatically different you can’t think of anything else.

The warm bath and the ice bucket: that’s how technology progresses, too.

As an example of the warm bath, you could point to the improvements in computing power in PCs and smartphones. Every year, they’re faster. You don’t notice how much better until you have to use an old device. (Or, of course, upgrade from a years-old device to a brand new one. These days, the effect is less visible on PCs than smartphones.)

Warm enough yet?

Another, less familiar, example is the burgeoning field of artificial intelligence (also known as machine learning, deep learning, neural networks, expert systems and so on). AI has been the “next big thing” for decades; the burden of expectation was so great it couldn’t meet them. Where, in the year 2001, was HAL, the talking sentient computer from 1968’s film 2001?

And yet, bit by bit, AI has been improving. I realised something was going on two years ago when I wrote a story about an app called Jetpac which could examine Instagram photos and determine whether the people in it were happy, grumpy, and so on.

To do that, Jetpac analysed 100 million photos and was able to determine whether those in it were wearing lipstick (so must be “glammed up”), had moustaches, etc.

A fun story, but it was the underlying technology, which Pete Warden, then CTO at Jetpac, explained to me that made me realise AI was back on the agenda again. He had used a neural network (which mimics, in machine form, the way neurons in the brain work: certain stimuli are reinforced, others are de-emphasised, in a learning process) to do the analysis.

It wasn’t surprising when eight months later Google bought Jetpac. The fit with Google’s broader AI drive was so obvious.

Where’s that tech now? Almost certainly powering the recognition system in Google Photos. Isn’t the Photos recognition system clever? But equally, isn’t it so obviously a progression from the face recognition we’ve seen in apps for years? The temperature is rising.

In fact, the AI temperature is now so high that this week we may witness a key event: a machine winning against a human at one of the subtlest board games ever. AlphaGo, an AI program developed at Google’s Deepmind in London, learned how to play the Chinese game Go at a professional level – and then beat Europe’s best player 5-0. On Wednesday – Tuesday night in the US – AlphaGo takes on Lee Sedol, the game’s top player. (If you haven’t played Go (most people in the West haven’t) let me put it like this: it makes chess look crowded and trivial: the board has four times more points than a chessboard, and the number of possible moves is far, far larger.)

AlphaGo isn’t like Deep Blue; it isn’t programmed just to play Go. Instead, it has a “learning” system which was tuned to play Go by working through millions of games and learning what outcomes were best. It could probably learn to win at chess. The core program learned to play video games.

This is the warming bath: how did we get to the point where computers could learn to beat the best player in the world at a game where intuition and “feel” are seen as essential?

The ice bucket

By contrast, some technologies are ice baths – so dramatically different from what has gone before they upend our expectations. Virtual reality (VR) fits this well. Immersive VR is an utterly different experience from what has gone before and the potential for creating new ways of interacting are what have so many people excited about it.

To people who haven’t tried it, VR tends to be “that thing where you wear the stupid helmet”. But that’s because they haven’t experienced the ice bucket. In the past, trains were a similarly disjunctive experience, able to travel at absurd speeds. There were even fears that the velocity would make passengers’ bodies fly apart.

Are there other “ice bucket” technology examples? The original iPhone was a shock to pretty much everyone, even though the technologies it contained (notably the multi-touch screen) were already known. From January 2007, Google’s Android team sidelined their work on a BlackBerry-like device and focussed instead on a multi-touch product.

Your preference doesn’t matter

Ice buckets change the game abruptly; warm baths surround us and raise the temperature so we can’t imagine life before them. There’s no way to pick which is “better” – and we don’t get to pick anyway, because they happen quite independently of our wishes or expectations. But in truth, there are more warm baths than ice buckets. The gradual improvement of smartphone screens, battery life, chip speeds, mobile reception, mobile speeds, design improvements – they’re all slow improvements which you don’t notice until you don’t have them. For dramatic change, though, the ice bucket beats the lot.

Moment of truth

There’s an instant as you first experience a splash of water when you don’t know whether it’s hot or cold. The match between Lee Sedol and AlphaGo could be like that: an odd mixture of hot and cold, a “where were you when…?” moment. Garry Kasparov’s loss to Deep Blue in 1997 was an iconic moment, remembered by many. It has taken nearly 20 years for a computer program to be able to challenge the top human in Go, which tells you about the gap in complexity between the two games. Fewer people understand Go than chess; but everyone understands winning and losing. Computing’s advance is bringing us a moment when the ice bucket comes from a warm bath.

The first match between AlphaGo and Lee Sedol starts at 1pm Seoul time on Wednesday (4am GMT Wednesday, 11pm EST Tuesday, 8pm PST Tuesday). The match will last up to four hours. It can be viewed on Youtube; there will be commentary (which might not mean much to non-Go players) at Gogame.

Published by

Charles Arthur

Charles Arthur has been a journalist writing about science, technology and sports for over 20 years. He has worked at New Scientist, been technology editor at The Independent newspaper, and was most recently technology editor at The Guardian newspaper from 2005 to 2014. He blogs at The Overspill (http://theoverspill.wordpress.com).

23 thoughts on “How AlphaGo Illustrates the “Warm Bath And Ice Bucket” View of Technology Progress”

  1. “Fewer people understand Go than chess;”

    Oh please do try to remember that North America is not the entire world. In Japan and China, Go is culturally ubiquitous and everyone knows it and plays it.

    As to the meat of your article, I beg to differ. AI remains as elusive as ever. The seeming breakthroughs of recent years come about either because computers have become so fast that brute force techniques are becoming more and more feasible to apply to previously intractible problems (cf Google’s “boil the ocean” approach to autonomous driving, where rather than even trying to program a computer to have judgement skills, they just give the computer driving the car a ridiculously detailed map of every possible aspect of the road it is to drive on), or else because programmers have gotten more clever at choosing which aspects of a problem to have a computer grind mindlessly through, about which more below.

    Chess is a rational game – the player who thinks the most moves ahead wins. That’s extremely easy to model for a computer to play. The fact that the human can intuitively see which future moves can be disregarded and which need to be visualized becomes irrelevant because the computer is fast enough to grind mindlessly through all the possible moves until it finds the best one.

    Because it involves so very many possible moves, it’s impossible for humans to play Go by thinking so many moves ahead. Instead the best Go players rely on intuition, on pattern recognition and puzzle solving -like skills, on the “feel” of the board. None of which computer programmers have the slightest idea of how to program into a computer, no more today than they did 30 or 40 years ago.

    The programmers of Alphago simply changed the focus from producing a naive computer program that mindlessly grinds through possible moves to giving an expert system a knowledge base of 30 million moves from actual go games played by humans. With that massive knowledge base, they had an expert system that could play go passably well, and then they had it play against itself to produce an even larger set of go moves for the knowledge base, producing an even better expert system. After enough rounds of bootstrapping, they produced an expert system that could beat human players.

    In the world of AI, expert systems are old hat. They’re only as good as the dataset you feed into them, and they’re useless if circumstances stray beyond the realm of the dataset. The only AI breakthrough here is that we can now build computers with enough processing power and enough RAM to handle expert systems with staggeringly large datasets, that can generate answers to the question “what’s a good next move in this game?” in an acceptably short period of time.

    1. Yes, I recognise that Go is widely known in Asia. It might have been better to say that fewer people *in the west* understand Go than chess.

      As to AI, though, my point remains. Computer systems are improving at their ability to do things like “learn” and “mimic” and, in the case of Deepmind’s Alphago, outplay humans. One can say “it’s not really intelligent”, but that’s a form of pulling up the ladder. Is a fly intelligent? Is a cockroach intelligent? Is a dog? Is a cat? Is a gorilla? Where’s the dividing line?

      If a computer can do all the things we reckon are “intelligent” in a fly/cockroach/dog/cat/gorilla, then it the computer “intelligent”? We’re a mess of datasets ourselves; but human Go players can transcend their teachers, and it looks like AlphaGo has done that too. The line of “this computer is/is not intelligent” gets harder and harder to define.

      And my broader point is this: AI is a warm bath. It surrounds us. It’s improving, but we hardly notice it – except that there will come a point where we look back and say “that was once impossible for a machine, and we classed it as requiring intelligence; now it is done by a machine.” Such as beating the world’s best Go player. There’s nothing trivial about that.

      1. My broader point is that “progress” in AI is not due to any actual progress in figuring out how to make computers think like people. The implication of that is that while programmers are getting better at solving specific problems that in the past would have required human input, the solutions to those problems are very specific. Each separate problem needs to be solved anew: we are no closer to being able to program a computer to be a general purpose problem solver than we were 30 years ago.

        1. “we are no closer to being able to program a computer to be a general purpose problem solver than we were 30 years ago.”

          And simultaneously we have been striping away the human capacity for general information gathering and problem solving as we insist humans specialize, or as Buckminster Fuller used to say, instead of doing more with less, we teach people to do less with more.


        2. “we are no closer to being able to program a computer to be a general purpose problem solver than we were 30 years ago.”

          That’s not correct. AlphaGo, and the Deepmind learning system, are very much closer to being able to solve by learning. The underlying Deepmind network learnt to play video games from zero, just given the controls, the pixels of the screen and the score, with the instruction of maximising the score.

          Just because we don’t recognise how it solves something doesn’t mean it isn’t a solution congruent with what an intelligent person would provide. The route may be different, but the destination is the same.

  2. Great post. I was checking constantly this blog and I am impressed!
    Very helpful info specifically the last part :
    ) I care for such info a lot. I was seeking this certain information for a very long time.
    Thank you and best of luck.

  3. Hello! I know this is kinda off topic but I was wondering if you knew where I
    could find a captcha plugin for my comment form? I’m using
    the same blog platform as yours and I’m having trouble finding one?
    Thanks a lot!

  4. Appreciating the time and effort you put into your website and in depth information you offer.
    It’s good to come across a blog every once in a while that isn’t
    the same outdated rehashed information. Wonderful read! I’ve bookmarked your site and I’m adding your RSS feeds to my Google account.

  5. Hmm is anyone else encountering problems with the images on this blog loading?

    I’m trying to figure out if its a problem on my end or if it’s the blog.
    Any feed-back would be greatly appreciated.

  6. Your style is unique compared to other folks I have read stuff from.
    Thanks for posting when you’ve got the opportunity, Guess I’ll just book mark this

  7. Hi to all, for the reason that I am in fact keen of reading this webpage’s post to be updated on a
    regular basis. It consists of pleasant data.

  8. Awesome blog! Is your theme custom made or did you download it from somewhere?
    A design like yours with a few simple adjustements would really make
    my blog shine. Please let me know where you got your theme.
    Many thanks

  9. I am really impressed with your writing skills as well as with the
    layout on your weblog. Is this a paid theme or did you
    modify it yourself? Anyway keep up the excellent quality
    writing, it’s rare to see a nice blog like this one today.

Leave a Reply

Your email address will not be published. Required fields are marked *