On Deciding . . . Better 3.0

We’re too snow and cold bound in Baltimore for new image capture, so I’m wandering through Umbria again with these D300 images from 2009. The Oz 2.0 techniques are giving me a second chance to see and enhance the light in the captures.

Chaos Theory was developed by exploring dynamical systems in computers. Its worthwhile considering this idea of a system itself in our exploration of deciding better.

Systems theory is an approach to studying a collection of parts by considering them as a whole. Each part has its role to play and some how interacts with other parts of the system. Consideration of systems is vital to understand where uncertainty comes from in our world or in simple clockwork universes like computer programs.

There’s is a much stronger tradition in empirical science of reductionism, understanding the functioning of a system by taking it apart to study its component pieces. The contribution of the part is considered by examining the interaction of that part with other parts of the system. If the functioning of those parts is understood, an overall picture of the system is built up. For example, the biochemical processes of a cell can be understood by analyzing the metabolites and how they are processed by cellular enzymes.

Reductionism has proven an extraordinarily powerful way to understand the world. For the most part, when an enzyme is blocked in a cell, its product disappears and its precursor builds up exactly as expected. You don’t need to know much about the function of the metabolites or the habitat and behavior of the animal. The function of these components is likely to be the same in locust, rat, cat and man.

A powerful reductionist approach is to study simple systems. The study of memory in the brain of humans or even rats remains much too complicated to explain as a system. It was possible to trace the system to particular brain areas like the hippocampus by studying brain injury in man and experimental lesions in rats. But after establishing that a rat without its hippocampal formations can’t remember how to rerun a maze, how can you figure out the circuitry within the hippocampus that stores that memory? And even if you do, how do you trace that function in the full functioning of a rat in a maze with its visual input and motor output?

Eric Kandel took the approach of finding a much simpler system to study memory, choosing the sea slug Aplysia as an experimental model. This classic reductionist approach provided important insights into how connections between neurons are changed by activity and eventually many of the same mechanisms were found to be operating in the rat brain. Eventually manipulation of these mechanisms in rats demonstrated that they were critical for memory formation.

Reductionism often works well in science. It shows that a component or mechanism in one system serves a similar purpose in another system even though these systems may be too complex themselves to understand fully. This can serve as valuable information if it turns out that manipulation of this one particular component has a consistent effect on the functioning of the overall system.

I’ve just finished re-reading Mitchel Wladrup’s Complexity: The Emerging Science at the Edge of Order and Chaos to followup on the Chaos discussion. About halfway through I realized that the book is now 20 years old. Perhaps because it was written so close to the founding of the Santa Fe Institute and was based primarily on interviews with key figures in that exciting flowering of ideas it still provides a vivid read.

I was struck by how little impact these big ideas seem to have had on the usual way of seeing the world. Perhaps chaos, complexity and emergence have entered the language, but dreams of improved prediction tools or appreciation of principles like unintended consequences don’t seem to have been achieved. There was a feeling when the book was written that we were on the verge of new forms of artificial intelligence and new approaches to economics that would help us understand the interconnectedness of the global economy. We have Google and mobile devices like the iPad. Sadly, it seems like more of the same only faster and in more places.

I had hoped that tools of decision theory, modeling and simulation would change difficult research and development projects like drug development. In my current job I get a pretty fair overview of the industry on a daily basis and can report that little has changed.

Insights from behavioral economics and advances in cognitive science that have had less impact one we way we see the world. There’s a constant stream of media reports about the science, but little evidence that these fundamental insights are informing our discussions about human behavior and ethics.

My original impulse when I started writing On Deciding . . . Better in late 1999 was to be at least one voice discussing what I thought were important implications of decision theory and Bayesian approaches to probability theory. Over the years, I’ve explored the sources of uncertainty in the world and most recently the emerging insights of Cognitive Neuroscience. I admit that mostly I write for myself, to get ideas out into better organized form and critically review them for myself.

My view of the value of writing and publishing on the net hasn’t changed in the last decade. I have a free, universally accessible publishing platform for my ideas. I’ve been fortunate over the years to actually have kindred spirits interested enough to read and comment on my efforts. I’ve been further encouraged over the last few months finding how Twitter as a microblogging environment provides a new venue to widen that circle of interaction like a virtual interdisciplinary conference.

The world of ideas is still vibrant. Its bigger and noisier than it was in 1992 or at the founding of the Santa Fe Institute in 1984. Certainly its bigger than the world of physics was at the time of either Einstein or Newton. However, I’m brave enough to suggest that like our world, those worlds also were ruled by a power law dictating the impact of ideas.

This is some of the last snow. Today it looks like a Dr Suess book out there with droopy trees everywhere

There’s a difference between predicting and knowing the future.

Predicting decreases uncertainty but does not eliminate it. We’ll call that level of certainty “knowing the future”. Predicting involves beliefs about the future state of the world and should be probabilistic, dealing in likelihoods of events, often describing a range of possible outomes.

I can make an excellent, accurate prediction about the card that will drawn from a deck. It will be one of the 52 cards in the deck.

Trivial? Not really. I’ve used my knowledge of the nature of card decks to constrain the possible range of outcomes to only those that are possible outcomes.

With more information about the particular deck I’d might be able to narrow the odds of various cards being drawn. For example armed with the exact order of cards in the deck and a historical dataset describing how often each card in each deck position is chosen, I could actually know which card is the single most likely to be drawn. Perhaps if we studied how people tend to draw cards we’d find that the center 25% got 60% of the draws, increasing the probability of those cards being selected over the top and bottom of the deck. Probably I’d be able to provide a list of the cards from most likely to be chosen through least likely.

Just because I know which card is most likely, it doesn’t mean that if the mostly likely card is not drawn in any particular trial that my prediction is wrong or inaccurate. The prediction only allowed me to strengthen my belief in some cards being drawn over other cards.

In selling any prediction service, If I only get one chance to test my ability to predict there’s no way to prove why I happened to be right or wrong. If my knowledge of the order of the deck and human behavior in selecting cards improved my ability to predict the card selected from 1 in 52 to 1 in 20, I’d be more than twice as good at picking the card to be drawn in advance, butits still overwhelming probable that my guess on any particular trial will be wrong.

It makes for a lousy magic trick but a very good way to make better decisions.

I’ve probably photographed my front walk more than any other subject in the last few years. There’s a nice combination of textures in the brick and garden elements particularly when there’s been rain or snow.

Today the world here is buried under a few inches of snow with more on the way tonight. If I want to do some more practice with coach Vince (Versace), I’ll need to use some archive images.

In this deterministic world, one of the most important sources of uncertainty is ignorance. In the Schrödinger’s cat thought experiment, we don’t know whether the cat is alive or dead. The source of that uncertainty at the macroscopic level is ignorance. The chamber is sealed and we can’t look inside of it.

From a macroscopic perspective, it’s no different from guessing what’s inside a sealed envelope or any other situation where some state of affairs exists but lack certain knowledge.

From a probability perspective, since we understand the physics of the cat in chamber system well enough, the probability of the cat being alive or dead is known. One could put that probability into a decision tree with a high level of confidence. If the probability of a decay event is 0.5, then the cat is not half dead and half alive, it has a 50 −50 chance of being alive. The event is determined, we just don’t know the answer until we open the box.

I want to completely blur past, present and future when it comes to ignorance since for decision making becaus it makes no difference. If the cat was unlucky enough to die in the first 5 minutes of being in the chamber, then the event occurred in the past we remain uncertain in the present. If we’re making a decision at the 30 minute mark the cat has a higher probability of being alive at present than if we make it at an hour just before opening the chamber, but since we won’t open the chamber until later, it doesn’t matter to us making the decision at 30 minutes. The event may or may not have occurred; ignorance obscures equally from the start to the finish. This is true of the future as well. If we substitute a coin flip to occur in an hour with the same 0.5 probability of success, the fact that the event occurs in the future makes no difference.

Let me also point out that knowledge without communication doesn’t dispel ignorance and thus has no influence on decision making. If we assume that the cat knows that it’s still alive inside the chamber, it makes no difference to us outside of the chamber. The outcome when the box is opened remains uncertain to us. Similarly, if I have a tape or transcript of an event in the past as long as I can’t time travel to influence those events my knowledge has no effect on decisions in the past. I know the text of Lincoln’s Gettysburg Address but that doesn’t affect the choices he had in writing it in 1863.

In the same context, consider the effect of an omniscient God who controls events in the world. Whether its a God that created a clockwork universe that plays out over time or a God that intervenes in events in the time stream, it makes no difference. Whether its a God who knows the full history of creation from the inception of the world to its end or a God that steers events with some ongoing intent also makes no difference. The knowledge and power of this God has no influence on our decisions unless knowledge is granted us about the world in the past present or future. Without knowledge, the veil of ignorance is itself enough to create conditions of uncertainty. As the world is presented to us, our ignorance requires us to make probabilistic decisions based on belief.

Knowledge has positive value in decision making. The more we approach the god-like state of knowing past, present and future, the better we can make decisions. Either probabilistic states can be made to collapse into certainty, like opening up the chamber and determining the state of the cat, or probability estimates can be refined, like checking on the cat halfway through its time in the chamber. Information that is irrelevant to the decision doesn’t help refine the probability estimate. Telling me its snowing outside can’t help refine the probability of whether the cat’s alive when the chamber is opened.

Making decisions under conditions of uncertainty is no different from making decisions under conditions of ignorance. Choosing ignorance over available knowledge is a completely different case

I’ve been writing more than photographing in the last few months. I’ve set up the MacBook Air as a fast and light photo processing station.

I’ve moved Aperture and Photoshop CS3 over to the new machine. Nikon Capture NX2 didn’t make it because of serial number issues, but I’m fine with that for now. Based on Thom Hogan’s recent recommendations I feel pretty comforatable with RAW conversion in Aperture with a Vincent Versace influenced OZ 2.0 post processing in Photoshop.

The speed of the workflow is perfectly fine as Derrick Story described soon after the release of the new Airs. I was a bit hesitant about ordering this base 2GB model, but would have had to pay too much of a premium for the 4 GB. And it was suggested to me that with the SSD serving as scratch disk and memory paging there wasn’t going to be a huge hit compared to a conventional HD system.

I like the Tamron zoom with Nikon D7000 combination for image capture and I’d prefer to practice with my favored tools- a DSLR and fast lens, rather than try to get by with a compact. Switching to the Sigma DP1 hasn’t worked because of the fixed wide lens. Its a great travel camera, but not flexible enough to grab a shot like this one from a car window in very dead overcast light.

There’s really nothing very mysterious about uncertainty. We’re soaking in it every moment of our lives. We generally don’t know what’s happened in the past, we don’t fully know the current state of affairs in the world and we have even less of an idea of how events will unfold. While some have looked to the uncertainty that’s part of the current quantum physical models of the universe, there’s really no need to invoke such esoteric mechanisms. Its part of the nature of being human that we experience the world as uncertain. Yet it appears that in a material world working according to Newtonian physics, there is no real uncertainty. The current state of the universe determines the next state of the universe.

The prospect of a “clockwork universe” is frightening. In the most extreme conception, we live in a universe which began its existence with a set of initial conditions that then is playing out the predetermined events implied in those initial conditions. Whether those initial conditions were intelligently set by a creator or “just happened” doesn’t matter for us,we are each actors playing out our parts in a predetermined script. Theoretically, a sufficiently powerful being could have access to all of the initial conditions and the ability to calculate all of the interactions. Prediction would be easy enough by simply running the calculations needed to predict any future event.

I’m going to unilaterally cut off debate on this interesting topic (author’s privilege!) by assuming that this is exactly the kind of world we live in. I’ll ignore quantum effects because at physical level the world appears to be mostly deterministic. Actually, I believe that the uncertainty at the atomic level makes the clockwork universe conception untenable. It appears that events at some physical level like radioactive decay are truly probabilistic. As long as these random events have effects at a macroscopic level (producing a click in a geiger counter for example) then if the world were started over again with exactly the same initial conditions, it would turn out differently because these random events would be different the second time around.
We can use the famous thought experiment of Schrödinger’s cat to illustrate this.

Schrödinger wrote:

One can even set up quite ridiculous cases. A cat is penned up in a steel chamber, along with the following device (which must be secured against direct interference by the cat): in a Geiger counter, there is a tiny bit of radioactive substance, so small that perhaps in the course of the hour, one of the atoms decays, but also, with equal probability, perhaps none; if it happens, the counter tube discharges, and through a relay releases a hammer that shatters a small flask of hydrocyanic acid. If one has left this entire system to itself for an hour, one would say that the cat still lives if meanwhile no atom has decayed. The psi-function of the entire system would express this by having in it the living and dead cat (pardon the expression) mixed or smeared out in equal parts.

It is typical of these cases that an indeterminacy originally restricted to the atomic domain becomes transformed into macroscopic indeterminacy, which can then be resolved by direct observation. That prevents us from so naively accepting as valid a “blurred model” for representing reality. In itself, it would not embody anything unclear or contradictory. There is a difference between a shaky or out-of-focus photograph and a snapshot of clouds and fog banks.

Schrödinger has successfully introduced the odd mathematics of quantum uncertainty into the realm of our experience. There are interesting questions about whether the cat is conscious and thus an observer of its own state and whether there is a difference between the unknown and indeterminacy that are beyond the scope of the current discussion. We want to focus on the fact that the quantum event of decay, an event that can’t be predicted only observed to have occurred, can dictate the fate of the cat.

If there was no random decay, the cat walks out of the chamber and continues to influence the world, shedding fur and ending the lives of mice and birds. If the cat is dead when the chamber is opened, there is a different set of events because of the random quantum event of radioactive decay.

But as I think Schrödinger is arguing, to say that reality itself is blurred seems ridiculous. The macroscopic world operates under deterministic Newtonian physics where theoretically every event is mechanistically caused by the immediately preceding state of the system. My next thought is determined by the current state of my brain, not some mystical non-physical spirit that is free of the physics of our deterministic world. There is no quantum level entanglement like our blurred cat where the next thought both exists and does not exist.

We’ve come to the first philosophical barrier in Deciding Better, the nature of free will in a deterministic universe.