|
Dither Dither If we could improve the process of human creativity the impact on our society would be dramatic. It is at the heart of all the economic and artistic advances we enjoy. Without it we would still be living in caves. Even a tiny improvement would have major implications over a time period measured in decades, such is the power of cumulative change. I believe that such a possibility exists, and it exists in a strange and unexpected place. The creative process is one of the least understood aspects of human behaviour. How do new ideas come about? How do we go from plainsong in the middle ages to trance music in the 21st century? From a stone knife to a nuclear reactor or from cave painting to video art? Do ideas fall like rain from the heavens? Are we visited by the Muses who gift us as they see fit? Or are there other more mundane processes at work? Processes that we might be able to consciously modify, improve perhaps? There are several well understood ways in which we know we can improve our creative output. We can do more of what ever we are doing in the attempt to create. Paint more. Write more songs. Build more prototypes, write more code. This has a positive effect as anyone who has done it can tell you. But if it was a reliable tool there would be a lot more great authors or songwriters. On its own, the practice of an art does not in itself make one good. It will certainly improve certain needed aspects of the processes one employs in the act of invention. Playing more music will make one a better musician, painting more will make one a better painter eventually. But it won't ensure that you paint a masterpiece, write a great tune, discover special relativity or invent the pagerank algorithm. So what are the processes that we really use to be truly creative? Well, there are an awful lot of people who believe they have the key to this. A quick search on the Net reveals a plethora of 'self help' books, 'systems' and even, gods help us, post modernist writings on the subject (although I always find these hard to distinguish from humour). Unfortunately, given how profoundly useful (and wealth creating) a properly successful model would be and the absence of billionaires in these ranks it seems very unlikely that any of them really have a clue. There is interesting work going on, particularly concerning the phenomenon of 'flow', that elusive state of sublime, intense focus that anyone who has ever created anything will know. This is being studied primarily as a psychological phenomenon.[1] It seems possible that in time this will lead to a much better understanding of the process and in particular what it takes to get a human being to be able to do it well. My interest in all of this is more prosaic though. I'd like to understand the process itself. It's a fascinating and potentially very rewarding area of study, figuring out how to make people be more creative but I'm more interested in how they actually do it when in the actual act. Why put that note there, why choose that word? How did you decide to paint his hat red, to use a hash table there? This is at the core of the creative act, the act of decision. One can argue that assembling all the options for the decision is important, and it is, but that's what practice is for, the dull but worthy work we all do before that experience of flow. It is when we actually decide that we create. It might be between one of two options, say whether to write the tune on the piano or guitar. Or it might be a decision between many options such as exactly where to place the man in the red hat on the canvas. It still comes down, in the moment, to a decision. It is very interesting to contemplate various strategies that creators use for making these decisions. Interestingly, writing them down and then doing them in 'slow time' is usually pretty disastrous for creativity. The ineffective nature of this approach is best seen in the difference in innovative output by software startups where there is no formal decision process and, say, Microsoft where there is lots. The quick flow based decisions gives one Google Search in a years effort by two people, the slow formal committee decisions gives one Vista in a five year effort by who knows how large an army. There is little doubt that which process in more effective. Why? The answer must at least partly lie in a better understanding of the decision process that underlies creativity. Acts of decision are well understood in other fields of human endeavour. In particular, in the audio world whilst converting signals from continuous (they can take any value they like) to digital (where they must assume one value out of a fixed number). When we want to take the continuous value that represents the sound in the air and make it a digital value we must make a decision. We decide which discrete value it will be and we make that decision a certain number of times a second. On a CD we decide 44,100 time a second and we have 65,536 different values (represented by a 16 bit binary number on the CD) to choose from. Making this decision is called 'quantisation' . Early on in the development of the devices to make these decisions (called Analogue to Digital Converters) people realised that they didn't sound as good as they thought they ought to. The reasons for this, and their fixes, turn out to have some interesting implications for the creative decision making process. The first and most obvious way of improving the quality of audio quantisation is to just have many more values to choose from. This is like having a better grain in a photographic emulsion, it means that each decision is easier to take, you can always find a nearer value to the one you want. When you hear audiophiles talk about '24 bit' this is what they mean. There are 16,777,216 different values instead of a measly 65,536. The new DVD-Audio and SACD both support this level of quantisation. In the creative decision process we often have choice over this aspect of the work, usually in choosing what tools to use, which instrument we wish to play, what medium to work in. Musical instruments that offer poor resolution in the translation of an idea in a musicians head (Stylophone anyone?) to reality tend to be unsuccessful. No one programs in Fortran anymore. We crave a better resolution in the translation of idea from inside our head to the real world. The second main way we improve audio quantisiation is simply do do it more often. Modern audio systems will do it twice or four times as quickly as CD's and the difference is notable. In the creative act this translates to how often one can make decisions. For the solo creator this can be very quick indeed. Add people, and the process slows up. Choose a bad tool or medium and it has a similar effect. This is one aspect of creativity that is well understood, the correlation between speed of action and output. Those that can try an idea quickly and discard it if its no good will often do better work than those who simply do the same process at half the speed. There is no question that marble is a harder medium for sculpture than clay. One of the main reasons is simply speed. You can try five hundred ideas in clay in the time it takes to try one in marble. There are fixes for these problems of course, but they are all designed to increase decision speed. The engineer builds the bridge on the computer first, the sculptor makes clay models before tucking into that marble. The third way we improve the quality of audio quantisiation is the weird one. Its also the one that may be a useful, non obvious way that we can improve the creative process. It is there in the title of this essay. It is called Dither[2]. In the early days of CD's, increasing the resolution of the decisions or making them faster was simply not an option. The CD format was pretty state of the art when it came out, and unfortunately had no room on the disc to be faster or more accurate. So we were stuck with it. And 16 bit, 44.1 Khz audio does sound a bit rough when compared to good analogue systems. This was a particular problem for very quiet sounds. A nice solo oboe in an orchestral performance for example. Or solo piano. They sounded 'grainy' and distorted. Faced with no other option some smart engineers came up with a counter-intuitive solution. They mixed some noise in with the sound before they quantised it. They actually added some of that horrible 'hiss' that we used to spend so much time avoiding in recording sound. And it worked. Sound quantised with just the right amount of added noise sounded way better. Understanding how and why this works is at the core of understanding why it might be useful to all creative people. When a decision must be taken (is this moment in sound supposed to be represented by 64,200 or 64,199?) the decision maker (the converter chip) is limited to just those two values. Now if the decision maker has some built in biases about the way it makes decisions these biases affect the outcome in bad ways. And it turns out that no decision maker is ever perfect. Everything, including us, has built in systemic errors. The kind of mistakes that we always make. We all have them. Adding a little noise smooths out these errors and makes the results better. You can even get really clever and add certain sorts of noise that are much less noticed, if at all (called 'noise shaping') . The effect of this has to be heard to be believed. You hear a really quiet, very distorted sound become perfectly clear with the addition of just a little noise. Now you might ask how does this have any bearing on the creative process? Well, every time a creator is faced with a decision, the worst decisions are not the really obvious ones (hmm, Python or Fortran?) they are the grey ones. One can become interminably stuck on these kinds of decisions. Or one can make them and make a mistake. The power of Dither is that one can just add a little noise to the decision (toss a coin perhaps) , make it quickly and know that over time all that noise will even out. One can work faster without penalty. In fact the theory tells us that one can trade off how fast we decide with how much noise we add. Faster decisions means we can add more noise and be cruder in our decisions provided we arrange to smooth it all out later in some way. The really interesting thing to come out of the study of audio Dither is not really the understanding that adding a little noise to the creative process can be useful. Many people have discovered this already for themselves. Musician Brian Eno and painter Peter Schmidt's 'Oblique Strategies' cards [3] are a great (and famous) example of such noise. No, it is that there are different forms of dither that can be theoretically understood, and that some are just better than others. Early dither was simply whatever noise happened to be available. Then came dither with a triangular correlation function, then Apogees UV22 dither, out of band noise shaped dither etc. Just what dither is best for oil painters? For folk musicians? For programmers? I'd like to know. Meanwhile I think I have finally discovered a use for my Mothers old pack of Tarot cards[4]. What shape noise do they produce I wonder? John Lambert Uphill Farm, June 2007 1 – One of the leading thinkers in this area is Mihaly Csikszentmihalyi from the University of Chicago. His book 'Creativity' (ISBN 978-0060928209) is well worth a read. 2 – There's a great Wikipedia page on dither and its use in audio and video processing: 3 – The original Oblique Strategies cards change hands for substantial sums these days but apparently you can still buy a new set: 4 – The Pulitzer prize winning journalist John Camp wrote a thriller called 'The Fools Run' ( ISBN 0-8050-0990-6) in 1990 in which his protagonist used tarot card readings to add randomness to decisions. It got me thinking about the use of shaped noise in decision making in general, so thankyou Mr Camp. |
