Intro  Learning  Patterns  Real Life  Genetics  Classifiers  Biology  Neural Nets  Connectionism  Life  AI

Essays on Complex Systems Part 3 - Examples of Complex Adaptive Systems in Real Life

How corporations can evolve toward unethical or illegal behavior

A corporation conceived for profit, directed by shareholders whose financial exposure is limited and whose criminal liability is zero, has no built-in incentive to produce moral or legal behavior, except in the limited cases where corporate officials are liable for criminal acts.

Even there, the corporate entity has no natural life span, and thus can outlive all of its component people, and virtually any component is expendable.

So, for example, it can be beneficial to the corporation to develop or evolve an internal atmosphere where employees can only succeed if they break the law.

Employees who are caught are jettisoned and maybe prosecuted, but it becomes their personal responsibility, and their loss doesn't kill the corporation, doesn't damage the shareholders, (giving them no incentive to demand legal or ethical behavior), and even fines may not affect share price, since in some cases (if the controlling portion of an industry is involved) the cost will be passed on to consumers.

It probably doesn't even alter the internal atmosphere because it may only have evolved in the first place (meaning there may be no human to hold criminally liable) as the most efficient means of generating the greatest possible income.

Employees acting within that atmosphere are essentially adaptive agents in an environment where their behavior evolves according to the pressures of the environment.

The corporation may in fact produce employees prone to criminal acts that serve the corporation well, but who as a result of this service are individually doomed to "die" (via jettisoning and prosecution), a relationship between an entity and its component parts that exists in many places in nature.

Furthermore, employees that object will be pressured to leave either by their own volition or by being fired or by being passed over for promotion due to poorer performance than their more cutthroat competitors. However the exit occurs, those employees will be replaced by ones more accepting of that corporation's internal culture.

The implication is that if you want to change behavior, you must create rules that put the corporation itself in jeopardy, with a real threat of weakening or death.

As an exercise, describe how a corporation's components and activities map to those of a living biological system. Consider mergers, takeovers, ...

Shareholders are neurons, in that it is their collective behavior that directs the course of action, and that course, being collective, may be different from what any individual shareholder would choose.

But employees contribute to collective behavior, too, being capable of independent actions not directed by shareholders.

• Requirements for continued life
• What constitutes death?
• Instincts, Reflexes, conditioned responses, and reasoned responses to environment
• Thought and decision making
• Communications and Connectivity - how the parts are connected
• Growth and Repair
• Reproduction
• Metabolism
• Ingestion, excretion
• defense mechanisms
  • taking on debt to appear less attractive
  • camouflage? Adjusting ratios so company doesn't stand out in computer screens and searches.

Many corporation characteristics are very similar to those one individual person could have. What are some emergent properties or capabilities of a corporation, different from what could be possessed by one person, no matter how capable?

• The shareholder mechanism acts as a filter for motivations and behaviors. In the context of a voting block, shareholders may not consider factors that would affect them if it was their own reputation, etc, on the line, so what results is behavior from which certain possible considerations have been filtered out. But this relates more to shareholder psychology than truly emergent behavior.
• Outlives all of its components.
• Can outlive human antagonists or adversary litigants.

It seems like the above comparison of systems should be possible, which implies that if we start with a definition of a complex system, it should be possible for us to identify and describe them.

But we as people are also the component parts of this system.

As a theoretical question, is it possible for there to exist within a system levels of description that cannot be seen or comprehended by the components of the system?

For example, it is inconceivable that any individual ant has, or can ever have, any conception of the larger-scale organization of the colony that it is part of.

It is equally inconceivable that any neuron in a brain can be aware of the brain, or of the whole organism.

Similarly, is it possible that there exist levels of description within our social organizations that we are fundamentally incapable of perceiving or understanding specifically because we are the units that the system is composed of?

Or is our awareness that such systems exist sufficient to give us theoretical access to any level of description in any system, including those in which we are lesser units?

I've tended to view corporate behavior as being fairly uniform, shaped by environmental pressures common to all.

More likely, there is an entire corporate ecology, with different corporations filling different ecological niches. For example, corporate predators, who stampede a number of smaller corporations with rumors of takeover, watch for those that scramble least effectively with anti-takeover measures, buy them out at a discount, sell off the assets for cash, and move on.

Who would be considered the grazers? Who would be considered the plants, the original source of food? (Probably people!) What constitutes a fat, lazy, or sick corporation ripe for the taking? (Undervaluation,...)

And the Reverse...

If some useful insights can come from examining a business environment using concepts from evolution, what about the reverse, applying known principles from business (or other systems) to evolution?

As an example, for either system to thrive, there must be sufficient resources to make the building blocks.

As an example extrapolation, even if there were life on Mars, it would almost necessarily be primitive because the planet is not lush enough to have produced enough life to have generated much variation or develop much sophistication.

Misc.

Any complex environment may contain built-in pressures that cause agents acting within the system to adapt to it. As noted above, if such pressures exist, the agents will adapt to them.

A committee is a fairly simple system whose behavior is collective. Consider it as a brain in which the members are the neurons. What characteristics of committee behavior are similar to those in other complex systems? E.g. tendency to moderate or mitigate extremes?

Are the politics that occur within committees also applicable to descriptions of other complex systems? Probably so, as when some faction wants its views pushed through, but is blocked (inhibited) by another: neural subassemblies competing for dominance, but whose net output may be that of one, the other, a compromise, or some other unintended byproduct of the competition.

Hofstadter points out that intelligence exists in levels higher than the hardware; and that indecision or incorrectness at the intelligence level can exist even if the hardware is functioning perfectly (there are no indecisive neurons).

Consider the possibility that this hierarchy of levels is required for intelligence (it almost certainly is), and that there may be a minimum number of levels of removal from the hardware for intelligence to occur.

That's somewhat like saying there is a minimum level of complexity (which is also almost certainly true), but more than that, there may have to be sufficient buffer levels that the hardware considerations are completely irrelevant.

As an example, the cells in Adapt.cpp are completely unintelligent; they don't learn. One level higher, the collection of cells does learn, but that's as far as Adapt.cpp goes.

In systems complex enough to be described with multiple levels of description, are there any qualities that, in any such system, increase or decrease as you get farther from the hardware level?

Well yes, intelligence..., but does increased distance produce similar types of changes in all systems? That is, do the terms and methods you use to describe the system at the "next" level change in similar ways in all complex systems?

Figure out how to program some system (any system!) that can justify (support) more than one level of description. What is required? (Is Adapt anywhere close to this state? Neural may also be, since its flashing lights seem to form patterns.) Whatever the answer, there are these additional requirements for such a program:

1. Output that is observable. Intuitively comprehensible display of the system's output is very beneficial. Patterns are easier to recognize if the display (or reporting) methods naturally enhance or highlight them.
    
2. Practice at observing the output -- you have to become familiar with the system's behavior, same as if you were watching a species. You won't recognize different levels until you're thoroughly familiar. I would not notice any of the levels apparent in an ant colony without either being shown or by watching for a long time.
    
3. Practice, also, at thinking about and recognizing different levels of description in systems, in general.

-----

I think this is duplicated somewhere else:

Avoid spending time trying to build intelligence into programs: learning is the key -- acquiring new knowledge from experience.

Even a program as impressive as SHRDLU has the limitation that all its learning routines and methods are programmed in. If the programmer wanted an additional feature, he had to add yet another module to do it.

(Similar to discussion in Complexity of AI researchers ignoring learning: they'd just plug in a "learning" module later; pages 180, 183ff).

Actually, those are useful, since they are learning methods, methods of acquiring new knowledge, but more useful would be routines that could learn how to learn, so that the programmer wouldn't be required to figure out the proper routines: the program should be able to generate its own random flailing (as flexible as possible, even down to the level of churning out random bits: see section describing tradeoff between the size of the elementary behaviors -- flexibility -- and learning time) and then determine through feedback whether anything it is doing is bringing it any closer to its goal.

Evolution of Cooperation

Bicycle Racing

An easily understood example of where competition produces cooperation occurs in bicycle racing.

All the riders want individually to win. However, no one stands a chance of winning without using the benefit of the draft of other riders. It's virtually impossible to break away from the pack without the assistance of others.

So such cooperation as rotating the lead occurs whenever it is beneficial to the parties involved, but when any group, or any individual, stands to gain by defecting, then either a new alliance is formed or the cooperation dissolves completely.

Cereal price fixing, in the news at one point

Price-fixing situations don't require overt agreements. They can evolve, requiring only that companies coexist long enough to adapt to each other and evolve cooperation. (Such as the "cooperation" between opposing forces in the trenches in WWI, not to shoot to kill.)

Cereal: what factors allow cooperation?

1. Long history of pricing strategies practiced by stable cereal companies: company A develops a very good idea of what company B will do in response to a new pricing strategy. It's not even much of a stretch to imagine that company A might lower the price 25 cents, wait for company B to do the same, and then raise the price back again. It wouldn't take too many cycles of this for both companies to realize, independently!, "Hey, let's just both raise our prices together!" without any agreement, or even any overt conversations, at all. (Not all communication is verbal!)
2. Children aren't price-conscious and can be insistent.
3. Brand loyalty, unwillingness to switch for price.
4. Loyalty to cereal as a staple for breakfast.
5. Prices still low enough that consumers willing to buy.
6. Relatively fixed demand? Lowering the price doesn't increase sales much, or induce brand defection.
7. Small number of competitive cereal companies.
8. Cost of entry to the cereal market (advertising) prevents start-up competitors.

What changes in the operating environment would either prevent or destroy such cooperation?:

1. Tighter personal budgets.
2. Relax patent protection to allow duplicating each other's products.
3. Subsidize would-be startup competitors.
4. Define price-fixing by its end result, not requiring provable agreement, and regulate.
5. Define "agreement" as communication in any form, verbal or otherwise, that leads to cooperative practices, and try to prove a "communicative" pricing history.
6. Destabilize the environment (see below): sure, but HOW, without the above measures, all of which, except maybe 4 and 5, are unacceptable, with 5 very difficult? People are clearly willing to pay the prices charged.

Solution: Do nothing. Breakfast cereal isn't a necessity; let the prices rise. When they're high enough, people will stop buying, or other companies will brave the cost of entry.

Meanwhile, if you think their profits are excessive, buy stock in the company, which nobody who ever complains of excessive corporate profits seems to do.

The recognition that in an iterated (familiar) situation cooperation may be an inevitable result can be used to prevent cooperation where you don't want it.

In WW1, the cooperation (which may have been beneficial to the specific troops, but certainly wasn't for the warring parties), could have been prevented by recognizing that the cooperation resulted because the two parties faced a mutual dilemma and had the time to grow accustomed to each other's ways, so that any adaptations not to kill each other made both sides more comfortable and would inevitably be reinforced.

The solution would have been to move the troops around more frequently, to prevent iterated contacts under similar conditions.

As another example of cooperation where you would probably rather not have it, you would expect campaign contribution records to provide excellent ammunition for politicians to throw at each other, but they almost never do it, because it is in all of their best interests not to raise the issue too much, since they are probably all vulnerable in that area.