Thursday, July 6, 2017

Coarse rules and survivability...

As he did in his 2007 book, A Demon of Our Own Design, Richard Bookstaber returns to the story of the cockroach with his 2017 book, The End of Theory. The excerpt below really made me think, and as I read more of the book, I'm beginning to think just as his 2007 work was the antithesis to central banker comments along the lines of "At this juncture, however, the impact on the broader economy and financial markets of the problems in the subprime market seems likely to be contained." (Bernanke, 2007); his 2017 work will go down an the antithesis to the confidence of today's central bankers, portrayed recently by Janet Yellen's comment about believing that another financial crisis is not likely to occur in our lifetimes.

Now, onto Bookstaber and the cockroach:
The Omniscient Planner and the Cockroach 
If you were omniscient, if you could pierce through these limits to knowledge, how would you design a creature to survive in a world with radical uncertainty? That is, suppose that you have an omniscient view of the future and you know all the types of risks that a species will face and you can give that creature rules in order to give it the best chance of survival, not just in the current environment but in the face of all that will cross its path over the course of time. But you have one critical constraint: your rules don’t allow communication of any information regarding the unknown future states or solutions to those things that the creature would not be able to perceive in its nonomniscient state. (This is a bit like Star Trek’s Prime Directive.) 
Before setting down the rules, you might want to do some background work and see what the species that have faced this sort of world have done that have allowed them to survive. Species that have existed for hundreds of millions of years can be considered, de facto, to have a better rule set than those that have been prolific in one epoch but became extinct as crises emerged. So it makes sense to start there. 
If you take this route, you can’t do much better than to look at the cockroach. The cockroach has survived through many unforeseeable (at least for it) changes: jungles turning to deserts, flatland giving way to urban habitat, predators of all types coming and going over the course of three hundred million years. This unloved critter owes its record of survival to a singularly basic and seemingly suboptimal mechanism: the cockroach simply scurries away when little hairs on its legs vibrate from puffs of air, puffs that might signal an approaching predator, like you. That is all it does. It doesn’t hear, it doesn’t see, it doesn’t smell. It ignores a wide set of information about the environment that you would think an optimal system would take into account. The cockroach would never win the “best designed bug” award in any particular environment, but it does “good enough” and makes it to the finish line in all of them. 
Other species with good track records of survivability also use escape strategies that involve coarse, simple rules that ignore information. The crayfish, another old branch in the evolutionary tree that has been around in one form or another for more than one hundred million years, uses a winner-take-all escape mechanism: a stimulus triggers a set of neurons, each dictating a pattern of action, and one variant of behavior then suppresses the circuits controlling the alternative actions. That is, although a number of different stimuli are received and processed, all but one of them are ignored. 
These sorts of coarse rules are far removed from our usual thinking on how to make decisions because they ignore information that is virtually free for the taking. Yet if we look around further we see that coarse rules are the norm. They are not only seen in escape mechanisms, where speed is critical. They are also used in other decisions that are critical to survival such as foraging and mate selection. The great tit is a bird that does not forage based on an optimization program that maximizes its nutritional intake; it will forage on plants and insects with a lower nutritional value than others that are readily available, and will even fly afield to do so. The salamander does not fully differentiate between small and large flies in its diet. It will forage on smaller flies even though the ratio of effort to nutrition makes such a choice suboptimal. This sort of foraging behavior, although not totally responsive to the current environment, enhances survivability if the nature of the food source unexpectedly changes. 
For mate selection, the peahen uses a take-the-best heuristic: she limits herself to looking at three or four males, and then picks the one with the most eyespots. She ignores both other males and other features. A red stag deer also has a take-the-best strategy, challenging another deer for his harem by running through a range of behavioral cues until he finds one that is decisive, and stops at that point. The first can be done at a nonthreatening distance: the challenger roars and the harem holder roars back. If the challenger fails on this test, it’s game over. Otherwise, the challenger approaches the alpha male more closely and the two deer walk back and forth to assess their relative physical stature. If this showdown does not solve matters, they move on to direct confrontation through the dangerous test of head butting. 
The heuristic is a simple one, winner-take-all, where the first cue that makes a difference is determinant, but in this case the cues occur sequentially, going from the one that requires the least information (it can be done at a distance without even having a clear view of the opponent) to the most direct and risky. The underlying heuristic remains one that is as coarse and simple as possible, where the first cue that can differentiate makes the decision. 
Foraging, escape, and reproduction are the key existential activities, and we see that heuristics are at their core. And we also see a movement toward coarse behavior for animals when the environment changes in unforeseeable ways. For example, animals placed for the first time in a laboratory setting often show a less than fine-tuned response to stimuli and follow a less discriminating diet than they do in the wild. In fact, in some experiments, dogs placed in a totally unfamiliar experimental environment would curl up and ignore all stimuli, a condition called experimental neurosis. 
The coarse response, although suboptimal for any one environment, is more than satisfactory for a wide range of unforeseeable ones. In contrast, an animal that has found a well-defined and unvarying niche may follow a specialized rule that depends critically on that animal’s narrow perception of its world. If the world continues as the animal perceives it, with the same predators, food sources, and landscape, then the animal will survive. If the world changes in ways beyond the animal’s experience, however, the animal will die off. So precision and focus in addressing the known comes at the cost of reduced ability to address the unknown. 
It is easier to discuss heuristics as a response to radical uncertainty when we are focused on less intelligent species. We are willing to concede that nature has surprises that are wholly unanticipated by cockroaches and our other nonhuman cohabitants. A disease that destroys a once-abundant food source and the eruption of a volcano in a formerly stable geological setting are examples of events that could not be anticipated by lower life forms even in probabilistic terms and therefore could not be explicitly considered in rules of behavior. But the thought experiment of the omniscient planner provides insight into what heuristics are doing in our decision making as well. It is not that we take on heuristics solely because of limits in our cognitive ability to solve the problem with the full force of optimization methods. It is because some problems simply cannot be solved by optimization methods even absent cognitive constraints. Absent our being omniscient, we cannot apply optimization methods to the real-world problem we face, and if we try to do so, we simply have to make things up.