Wednesday, June 20, 2007

Speculation on why things to wrong - part 2

I'm watching it rain over downtown Baltimore as this small plane flies by, and it struck me how much that plane is like the loop I just described in the last post.

Nothing holds it up except the very thin stuff we call air, but that turns out to be enough. Nothing makes it move except that continued cycles of the "propeller" push a little bit of air with each spin, but do 2500 spins a minute, so that turns out to be enough. A little effect times a huge multiplier can turn into a large effect, sometimes.

This is good news for us, because about all we can ever do is have small effects. At issue is what makes them add up to amount to something or not.

Anyway, here's a few "afterthoughts" on the prior post.

There seems to me to be only a small number of middle managers in US corporations, or mid-level officers in the military, or mid-level analysts in intelligence agencies, who can "tell it like it is", without adding or subtracting anything, without distorting the message to fit their preconceived notion (possibly right, possibly not) of what upper management's preconceived notion is. This means our big collective cybernetic loop has a lot of distortion in it.

It's tricky - we need mental models in order to make sense of the world, so they help and we'd be blind or effectively catatonic and overwhelmed if we didn't have these oversimplifications to cut the fire-hose down to a water-faucet trickle. And we want the models to be almost constant, or more precisely, "dynamically stable." They should change gracefully and smoothly as new information comes in that shows we need to update our thinking, but not so fast that the cart overturns.

That kind of stability, and information processing that is not devastated by one bad component, comes from parallel operation, not serial operation. If information has to go through a chain of links, and there is only one path, it is hostage to the worst or weakest link. If we use strings of symbols (words) to hold our knowledge, and we use serial processing (math, logic) to
"connect the dots", we are similarly hostage to and limited by the worst step. It only takes one mistake, anywhere along the line, in doing a math problem to get the final answer wrong. The mistake doesn't somehow "go away"as we do the subsequent steps.

You've probably never thought about it much, but if you had to buy that as a product, you'd think twice. What are the odds you'll get every step correct? Most people would say,not much.
What are the odds that everyone in a company will get every step right? not much.

So, aside from dropping physics and math and changing our major to something less technical, what can we do about this? Like driving on a road that's 1 lane each way and high-speed with no divider, we're at risk from the worst driver on the other side, regardless how safely we drive.

Well, there is an alternative way to work. There's probably many alteratives, but the one I'm familiar with and have a US Patent in involves "image processing" - or storing information in 2-dimensional (or higher) images and having basic operations (akin to addition or logic) that let us take several images and "connect the dots", same idea as before.

Except this time, this way, with images instead of words, we are almost entirely immune to "point errors." The picture of George Washington on that dollar bill looks like George and serves its role even if it is damaged, or we spill grape soda on it, or draw a slash across it with a black marker. The image is "robust." The image's tiny elements, or "pixels" hold each other up in some almost magic way, so they don't all fall down when one of them is bad, the way a string of pearls would fall apart if you pulled out the thread.

Hmmm. This looks like a much stronger way of representing knowledge and processing it then.

Unfortunately there is almost a religion around serial sequential symbol processing and strings of words as a way of knowing and thinking and drawing conclusions. This was populalarized by
Newell at Carnegie Mellon as a basis for computing, and so our computers have the same property -if one stupid little thing goes wrong, the whole output is messed up.

So, again I must head off to class. where I'm going is that there is a better way. The "loop" that I claim has to be at the core of any corporation, or any living or pseudo-living thing, carbon or silicon based, should best be implemented with PARALLEL channels, not SERIAL channels. The communications should be multithreaded and, effectively, images, not numbers or words. What we want is our messages to be self-validating and self-error-correcting so that the small and inevitable errors "go away" for real, like damage to George's portrait.

This turns out to be way cheaper than trying to get each individual chain to be "perfect."
We actually don't need each chain, or each manager, or each analysis to be "perfect" at the six-sigma level .... we only need the TOTALITY to be perfect.

Whoa, you say. You just said you can't get perfection if the pieces aren't perfect.

Well, I didn't lie, but I was talking about serial, single-threaded processing -the kind that a string of pearls represents. One cut to the thread and you have 200 pickup, not a necklace.

There's a better way, is what I'm getting to -- "massively parallel" channels of information flow, so that we are effectively using the same trick images use to self-heal.

Not just parallel among people, but parallel among levels of the organization, so that efforts are aligned, mutually supportive, and mutually "entraining." That both makes it harder to make a mistake in the first place, but easier to detect and repair it in the second place. It's very robust, and, surprisingly, very cheap and can be done, as Toyota has done, with silly little colored pieces of paper and such. It doesn't involve a $400,000,000 investment to accomplish.

But it does involve changing culture of information control and flow and how we listen to each other. That's both good and bad news.

Good news, because it's not technical and any country or people or economic status can do it.

Bad news, because it involves letting go of stereotypes and carefully locked down pathways for information flow that the serial model depends on.

So, either model X or model Y works. the question is, how the heck do you transition from oneto the other without dropping the ball or exploding the organization.


Sphere: Related Content

1 comment:

Cheryll said...

"So, either model X or model Y works. the question is, how the heck do you transition from oneto the other without dropping the ball or exploding the organization."

We're waiting...........