Complexity Blog http://complexityblog.com/nucleus/ en-us Nucleus CMS v3.2 © Weblog http://backend.userland.com/rss http://complexityblog.com/nucleus/skins/default/images/cblogo.gif Complexity Blog http://complexityblog.com/nucleus/ <![CDATA[It's Now Cliché to Criticize <i>Homo Economicus</i>]]> http://complexityblog.com/nucleus/index.php?itemid=111 homo economicus model is an unrealistic and generally unhelpful one. In fact, its failures are so widespread and obvious these days that it's cliché to even mention them.I've criticized rational choice theory, utility theory, and decision theory elsewhere (recently here) and I extend my criticism to the pseudo-novo techniques that attempt to incorporate various "visceral factors" such as hunger or fear into their analysis. My problem isn't that they attempted to incorporate them; clearly such features need explanation. The problem is that they have been so clearly added as ad hoc fixes to a dead theory to keep in churning. The analogy to the epicycles of Ptolemy's planetary model is strikingly strong. This research is stillborn; the death-throws of a sinking colossus of methodological history. But this post isn't about how bad that research is. This post is about not needing to motivate those considerations anymore.

Rational choice theory had a good run. It started mild in purpose and gained greater and greater momentum as various conceptual technologies and then mechanical technologies allowed steady improvement in the models and their applications. In the past forty years computers have exploded progress in the field to the point that the marginal benefit of adding another rational widget to the framework is exceeded by the cost of investing the time and effort to find room for one. Combine this with explosive growth in behavioral and neural psychology that has served to provide mountains of concrete counter-evidence to the obviously false (and never seriously purported to be true by reasonable scientists) claims of rational choice theory and we have a group of desperate social scientists grasping for something to do and hoping that combining new irrational or arational widgets to their old framework will keep them useful (or at least busy).

But that framework is the wrong framework for incorporating these new factors. Arguably the belief-desire framework was never the right framework for modeling decision behavior, but it was the best available for a long time. Now we can do better. But before we can move forward people have to realize that people are not homo economicus, that this is too obvious to bother mentioning, and that we need to come up with genuinely new methods to model humans as they are. ]]> Commentary http://complexityblog.com/nucleus/index.php?itemid=111 Mon, 08 Oct 2007 23:49:56 -0700 <![CDATA[As True as True Can Be]]> http://complexityblog.com/nucleus/index.php?itemid=110 true in a fictional world.The first point is a Humean truism: since our "knowledge" of the world is delivered through our senses, all statements about the world must actually be about a fictional mental model based on data from our perceptions. Naïve realists make a pretty good case that the things we say about the world are really about the world and that insofar as I can knock this table this table really exists. Within a reductionist framework we can go even further than the Humean in denying the naïve realist position. Even granting that our perceptions are reliable revealers of a real world, my perception of a table object isn't a perception of an existing thing. This table is made of atoms; in fact we say it is constituted by atoms to underline the fact that there isn't both a table and a collection of atoms under my computer. If we are talking about interior design then we'll probably say that tables are existing things, but that's just the ontology for the furniture/room model. If we're talking about the real world's ontology then we can't double count the atoms and the table – the table does not exist in the real world if atoms do.

Complex systems theory is within a reductionist framework so we have secured that the set of objects that exist must all be at the same level of organization. Furthermore, we are used to thinking of objects as coherent behaviors of constituent parts; i.e. emergent phenomena. A statement about the body, (say) is actually a statement about aggregates of organs which is actually a statement about cells and all the way down. If feels strange to say that it is really true that unicorns are white because there aren't any unicorns "in the world". Whatever it means for a statement about a fictional object to be true it doesn't mean that it is instantiating in the real world. Now consider the statement "My body has two arms". My body doesn't really exist, it's just an aggregate of cells and aggregates of cells can't have arms. We make a conceptual leap from the aggregate of cells to the body, but the body doesn't exist as a physical object. Just like the unicorn exists in a model of mythical creatures and stories, the body exists in models of biology and most of the folk models that allow us to live our lives.

So a scientific claim is true if it is coherent with the model under consideration. That's all, that's what 'true' means. The models are fictional worlds; they posit objects and interactions that may have absolutely no bearing on reality. There may not even be a reality. It's irrelevant to the models and to whether the statements of the model (theorems or corollaries) are true…except claims in the model that the model corresponds to reality. Claims about matching reality are either nonsense (if there isn't a reality) or false (because reality can't actually be the way it is described in the models…tables (and people, and atoms, and hurricanes, and…) can't exist in the "real world". That's as far as I'll take this now, but there are lots and lots of implications of this result for the notions of causation, inference, mereology, emergence, machine learning, etc. It's a big result I guess, but it's pretty obvious too.]]> Philosophy http://complexityblog.com/nucleus/index.php?itemid=110 Tue, 02 Oct 2007 00:52:32 -0700 <![CDATA[Tournament organizers make it right]]> http://complexityblog.com/nucleus/index.php?itemid=109 click here to read 'How my tourney experience turned into a fiasco'). Here we are a year later, and the tournament organizer has not only apologized for the snafus, but has taken several of my key suggestions into account for the next tournament. Kudos to James Fowler for making the situation right. Read his words here. Dr. Fowler email me on September 23, 2007 and offered these comments for this blog:

Hi Ken,

I just recently found your blog on your entry in our Tournament of Party Strategies in which you point out that we programmed your strategy incorrectly. Please accept my sincerest apologies. You are right that we implemented your strategy incorrectly. It is too late to change the paper, but we have now added the following text to our onlne appendix at http://jhfowler.ucsd.edu/tournament_appendix.pdf:

"NOTE: (9/1/2007) This strategy was not programmed correctly in the tournament, due to a mistake James Fowler made when translating code sent by Ken Zick. Fowler inadvertently left out the first satisficing stage of the author's strategy. Zick emailed Fowler about the problem, but his email did not reach him because Fowler had changed institutions. Given the success of other satisficing strategies, we think the correct implementation of Zick's stratgy would have made the strategy perform much better."

Michael Laver and I plan to have a second tournament in 2008, following publication of the results of the first tournament in the Journal of Conflict Resolution (the same journal where Axelrod's tournaments first appeared) and we would be happy to pre-enter the *correct* version of your strategy in the first tournament (if you are willing to take the risk of us programming it again!). We have taken your suggestions to heart, and improvements for the second tournament will include hiring additional programmers (I did all the programming myself last year) and a plan to show authors the code for their strategy prior to running the tournament.

All my best,
James Fowler, UCSD
http://jhfowler.ucsd.edu


I'm impressed with this openness and sincerity. I look forward to a great event in 2008. ]]> Methodology http://complexityblog.com/nucleus/index.php?itemid=109 Fri, 28 Sep 2007 16:46:47 -0700 <![CDATA[Role of Prompts in Scientific Theorizing]]> http://complexityblog.com/nucleus/index.php?itemid=108 same fiction when they discuss the model and can hence argue or agree about what is true in that fictional world. But those parts that are not explicitly specified will admit to variation based on the background models of the imagining person. That variation allows exploration from many frontiers of the existing canon.

And just as a child moving a toy truck from room to room is an effect of a child's imagining a real truck behavior in such a way, the theorems, equations, diagrams, and other representations of scientific theories are initially the result of somebody "seeing" the world in a certain way. These initial items then prompt others to visit the same imaginary realm and create more representations that fit that realm. There is a natural feedback pattern here.

Now turn to the example of children building a snow fort. Initially a simply wall of snow provides a focus for imaging a great structure. As more and more people get involved and improve the edifice over time it constrains the imaginings more and more. A well developed snow fort may have four sculpted turrets thereby making it difficult to imagine it having six…harder than if there were actually none. If the game they want to play demands a fort with six turrets then they have to destroy their hard work and perhaps build those turrets. Sound like science to you? Sure does to me. Removing, editing, and/or replacing large chunks of theories is a great deal of hard work and it always confronts a tremendous degree of resistance (from those who still want to play the four-turret game – what I call "recalcitrant has-beens").

The next trick is to figure out how to use this analogy to motivate these sorts of changes in scientific theories. Using insights from information design, visualization, and human-machine interface we might be able to develop more appropriate and adaptive prompters for the complex systems model of the world. It too is a (hopefully useful) fiction and too many are having trouble imagining its world, or resist for fear that there be dragons here. ]]> Philosophy http://complexityblog.com/nucleus/index.php?itemid=108 Wed, 12 Sep 2007 20:01:43 -0700 <![CDATA[Physical, Virtual, and Hybrid Networks]]> http://complexityblog.com/nucleus/index.php?itemid=107
On the other extreme of this scale are the purely virtual networks. Social networks, the world wide web, airline and shipping volumes, political hierarchies, food webs, have no physical substrate to constrain their structure. The network structure is an abstraction of the flows of something (information, power, passengers, etc.), among entities in the model. So even if the entities are physical objects, and the things flowing are physical objects, there is no actual network structure to be found. The network is just the representation of interactions or potential interactions of the system's entities. Dynamics take the form of flows among entities and/or the addition and/or deletion of edges and/or nodes.

In between those two poles are the hybrid network models: road traffic, commerce (e.g. supply chains and money transfers), postal service, signals in a computer, blood flow, etc. What makes them hybrids is that the dynamics in the flows are constrained to the physically network structured interaction pattern. For example, while cars are not strictly restricted to roads, for most intents and purposes traffic can be considered as restricted to the road network. One could have a fully abstract model of people driving from point to point without worrying about the intermediate paths (like a model of airline passengers from city to city without worrying about the flight paths), but it wouldn't model traffic. One could also construct a model, such as a city planning tool, with feedback between the traffic flows and the existing roads to reroute cars and build new roads. Combining the fully virtual travel model to the fully physical road network produces an useful hybrid with a wider spectrum of dynamics.

Realizing these distinctions may help our ability to create measures on these different sorts of network models. Perhaps the purely physical and purely virtual are the easiest to start with. Hopefully the hybrid models can gain some ground by combining these techniques, though other techniques will probably also be required. The distinction here is obvious but not obviously useful. Perhaps all that is needed is a distinction among changing structure, flows, changing flows, changing both, and feedback between structure and flows regardless of what it’s a model of. Hopefully drawing the parallels will improve our insight for what to study and how.]]> Methodology http://complexityblog.com/nucleus/index.php?itemid=107 Mon, 03 Sep 2007 23:27:34 -0700 <![CDATA[Heuristicism: Action from Rules]]> http://complexityblog.com/nucleus/index.php?itemid=106
As I have demonstrated elsewhere game theory is a simplified form of agent-based modeling: rule-following agents make actions contingent upon other agents' actions and the environment. There is just the one rule – maximize utility – and the burden of determining the utility function is pushed to utility theory (a modular part of the model that is also used by decision theory). Game theoretic models that replace utility theory with heuristics are exemplified by Axelrod, Grimm, Miller, Bednar & Page, etc. We can likewise replace the utility theory behavior driver in decision theoretic models. The motivation is that people, in fact, rarely perform the sort of calculations posited by decision theory and rarely have the information assumed by most of decision theory. People's behavior is generated by the interplay of many simple cognitive rules (Minsky, Dennett?) that are triggered by features of the environment. Modeling decision procedures in this way will produce higher-fidelity predictions and explanations.

The psychological support for humans' heuristic behavior is substantial and constantly growing (Dennett, Nowak, etc.). Of course, these are just psychological models and whether our conscious operation runs by adaptive rules or maximizing utility is irrelevant; the neuro-physical evidence increasingly lends support to the idea that what goes on in our mental theatre does not affect our behavior (Watson, etc.). That our behavior is more controlled automatically by physiological reactions, hormones, and the lower brain functions than the lofty operation of our frontal cortex means that even if the belief and desire model is better for modeling decisions, decisions do not affect behavior and hence will fail to make useful predictions. As Kant famously said, we cannot help but perceive our actions as being the result of the operation of our mind, but it turns out that this perception is illusory.

If the conscious mind doesn't affect behavior then there is no process by which the mind ought to make decisions. It can do whatever it wants so long as the body does the right thing. The evidence supports the claim that Joyce's non-necessary relation between decisions and actions is actually just correlation. So even if one creates an excellent model of decision-making one will not have really modeled anything except the cognitive wheels of the mind that never tough the world. In order to model what a person will do when confronted with a choice we should model the process with verisimilitude to how people choose actions – with systems of rules. While this heuristic approach bares some similarity to behaviorism, it is only the realization that predicting decisions is pointless if that doesn't help predict behavior. What we want our model to match in the final analysis is behavior.]]> Philosophy http://complexityblog.com/nucleus/index.php?itemid=106 Fri, 31 Aug 2007 21:56:26 -0700 <![CDATA[Morality as an Emergent Property]]> http://complexityblog.com/nucleus/index.php?itemid=105 Stanford Online Encyclopedia and Oxford's Dictionary of Philosophy for authoritative definitions. For those already familiar with some moral philosophy I offer the following caveat: this is quite different from what's out there, but hang in there.Behaviors Instead of Emotions or Beliefs

Humean concepts have so permeated moral philosophy and action theory that even accounts based on duty, reasons, or the incalculable value of human life are cached out in terms of the appropriateness of emotions, what beliefs are required to make the system work, whose goals are frustrated by certain actions, and so forth. Many of these theories explicitly list a freedom of will as a necessary assumption of the theory; at least most (and perhaps all) of the remainder has it as an implicit assumption. I have a paper that outlines reasons for wanting a moral theory that doesn't require free will that will eventually make it to this website, but the bottom line is this: we can't actually have the kind of free will that these theories depend on so to have a viable theory of morality it should not depend that kind of free will.

We also want a moral theory to be the kind of thing that evolutionary forces can act upon. Since we famously don't have direct access to others' minds, mating success can't depend on minds and we should focus on behaviors (acts rather than actions in the philosophical jargon) and other physical features of the system. If there is a strong correlation between mental states and physical behavior (and there very well may not be) then theorizing over mental states will be an unproblematic rough approximation – but we can do better. The trick is that even though morality is felt (and that's mental) it only can be selected for or against if that feeling translates into behavior. Moral feelings and intuitions are still what we have to appeal to and make sense of, but this must be done in terms of mechanisms that make sense in terms of biological explanations of human behavior and their origins.

Morality at the System Level

For morality to be an emergent property it must be a system-level property resulting from the interaction of parts for which morality does not apply. If one thinks that morality is a purely mental phenomenon and that all mental phenomena are emergent from neuron activity then that would do it, but that's not the avenue that we'll pursue here. On this systemic view, morality is a property of interpersonal interaction. At minimum that requires a group of people; in particular morality requires a reproductively viable population of individuals, enough to perpetuate the species. This claim is the conclusion of the realization that moral intuitions evolved and evolution operates on sufficiently large (and sufficiently differentiated) population. The first objection I would expect to hear is that even if the population dropped below the viable level the individuals will have the same moral responses and (for example) murder would still be wrong. But this is a completely different metaethical position; behavior features that evolved as improving reproductive fitness become irrelevant when the population is doomed to extinction. That is not a claim but rather a corollary of the systemic view of morality.

Stability of Moral Disagreement

One of the results of placing morality at the system level is that it would explain moral disagreement in a particularly useful way. Variations in moral feelings, reactions, behaviors, institutions, etc. are analogous to variations in beak shape, feather color, tongue length, and other physical features that selection pressures can act against. Feature variation does more than just provide differential fitness for cross-generational adaptation to the environment. Persistent differences in behavior sometimes take the form of specialization – niche formation – such that the persistence of variations actually has a beneficial effect on the fitness of one individual versus others and for groups with niche behavior versus generalist groups. So the persistence of differing moral behaviors may be a natural aspect of human biology and culture given their evolutionary origins. These differences can have genetic, environmental/developmental, and learned generating mechanisms; I'd expect they all play a role.

These are some of the major ideas that I'll be developing through my research in the evolution of morality. The complex systems approach has a lot of scientific and conceptual insight to offer the topic, but there is hard question to ask: what is the philosophical impact of the view? To some degree philosophical ethics needs to be informed by science, but one can draw a clear distinction between descriptive accounts of moral feeling and normative accounts. Might there be a required role in normative ethics for the more traditional philosophical approaches. I will argue that there is not, but that is a hard argument to make and many people will resist convincing. Stay tuned for developments. ]]> Philosophy http://complexityblog.com/nucleus/index.php?itemid=105 Wed, 29 Aug 2007 00:00:07 -0700 <![CDATA[Notes on N-Person Games]]> http://complexityblog.com/nucleus/index.php?itemid=104
Coalitions appear in a number of different N-person games, especially ones linked to voting behavior and collective action. In the simplest majority vote game, any collection of strictly more than half the players counts as a stable winning coalition of players. In more complicated games, such those for actually used to model parliamentary elections, there are configurations that produce situations where there are nonstable winning coalitions, and sometimes no predictable outcome at all. I want to ask, "How can we adapt classical game theory to capture situations with multiple coalitions as the outcome?" An example would be various policies of nations or regions within a country where groups can act in concert, but each group has its own set of endogenously defined payoffs. State gun laws in the U.S. is a possible application; each state decides its own policy, but the policies of neighboring states has a tremendous affect on the efficacy of any given state's policy. Still, the system may admit several stable regions with distinct, stable policies.

The state gun law type problem can be (and first will be) addressed under standard closed-form game theoretic techniques. To further enhance this model and bring it closer to the structure of the actual situation we would like to be able to incorporate how each state's expected payoff is affected by their particular neighboring states' policy. If the hypothesis is that people can cross borders to acquire guns in states where the policy is more fitting to their desires, then the amount that one state's policy affects another is a function of how easily people can cross borders to acquire guns. California's payoff will be more affected by Nevada's policy than Delaware's. These heterogeneous payoff functions can be accounted for using a network-form representation of the payoff structure. Current models of games on networks use local information to make choices of actions, essentially pairwise interaction over the network's edges. But my technique acknowledges that in order to act strategically every state's policy matters to each state: California's payoff may not be affected by Washington's policy, but since Oregon's policy affects California and Oregon's payoff is affected by Washington's policy, California needs to consider Washington's policy's affect on Oregon's policy to determine what policy will maximize its expected payoff. So complete information is part of this framework.

Niche behavior appears when some action A only pays better than alternatives when it is played against some particular action B. You can think of this as a particular motivation for specialization, but it is more than just coordination. A niche member's success depends on others not joining that niche and in this way it is like an N-person minority game with several stable and distinct minority groups. One of the motivations for exploring this area is to show that even homogenous preference relations over outcomes can produce stable distributions of specialized behaviors in certain circumstances and to determine what some of those circumstances are. More on these topics as they develop over the next several months. ]]> Social Science http://complexityblog.com/nucleus/index.php?itemid=104 Tue, 21 Aug 2007 02:53:42 -0700 <![CDATA[Institutions, Incentives, and Recycling]]> http://complexityblog.com/nucleus/index.php?itemid=103
To get that model to work we would need to change the incentive structure in drastic ways. As one example, we could set up the whole garbage collection and disposal operation so that trash companies bill their residential and commercial customers based on volume and weight. This would encourage people to both reduce their consumption and increase their use of recycling. Since certain people (e.g. affluent and lazy people) would still not bother, there would also need to be an incarceration penalty for egregious violations. I said it was drastic. I'm certainly in favor of all this, but those same lazy and affluent people are the ones in control of such things and so environmental protection will probably never happen in this (or any similar) way.

So how can complex systems help? Two of the problematic features of recycling, alternative materials, and reduced packaging waste are that 1) it takes a certain critical mass before the value begins to accrue and 2) there are many opportunities for defectors to profit. So the situation reeks with a "tragedy of the commons" flavor and it's even harder because utility increases as a step function (so marginal personal benefits of cooperation may be nil while margin personal benefits from defection may be sizable). There are complex-systems approaches to investigating these scenarios, but the real insights come in when one sees beyond such simplistic models.

The problem ties into the literature on the diffusion of ideas/technology and the instigation of riots. Sophisticated models of both phenomena realize that social networks play an ineliminable role in determining how individuals' actions affect one anothers' behaviors. The goal would be to utilize the fact that people care about being cool and fitting in at least as much as financial gains and losses and that by convincing certain "mavens" to put the extra effort into recycling others will simply jump on the bandwagon. People also care about convenience; despite the observation above that some people can't be bothered to put in any effort at all, adoption of recycling norms would be a great deal easier if the recycling itself were a great deal easier. We can also utilize agent-based modeling to plan the layout of buildings and streets and the locations of bins and depots to facilitate maximal compliance with recycling goals. These are, of course, just simple, straightforward examples of some things that can be done. There is no shortage of good ideas in how to build a cleaner, viable future.

Complex Systems hardly offers a silver bullet in solving our environmental problems. Though there are plenty of good ideas, there are also lots of bad ones; and they are often difficult to tell apart and they compete for the same funds and attention of policy implementers. Perhaps owing to its youth and immaturity as a field, Complex Systems requires a great deal of research in new metrics and techniques for validation. We (as a community) are working on these issues, but in the meantime funds are still being diverted into efforts that have proven their lack of worth. The political message here is that investing in a plan that might fail is preferable to continued investment into a sure loser: we should try new things since old plans have failed or come up critically short. The second message is for researchers to give up on their dead horses: the problem is beyond adding epicycles and tweaking parameters; take a risk and try something radical because to solve the problem we're going to need a radical change in something.]]> Social Science http://complexityblog.com/nucleus/index.php?itemid=103 Sun, 20 May 2007 23:59:57 -0700 <![CDATA[Fractional Topological Genus]]> http://complexityblog.com/nucleus/index.php?itemid=102
With respect to agent based modeling, a simplified version of the role of topology is to determine what happens to the agents when they reach the edge of the world. That's simplistic because for higher genus (which have probably never been used with an ABM) it also determines how many edges of the world there are. One way to conceive of the fractional genus would be a permeable membrane such that some part of an agent wraps one way and some part of the agent wraps another way; but that only makes sense if fractional agents do. If agents are atomic (indivisible) then the easy way out is to use the fraction as a probability of one genus or another, but then I'd just leave it as a probability instead of torturing math for no reason (besides revenge).

Another place where fractions naturally arise is when ratios come into play (that's why they are called rational numbers after all). If it made sense to create a new space by dividing one space by another then spaces with fractional genus would just come out of the process. But what sense does it make to divide a space by another? Sometimes spaces are represented by matrices, matrices have inverses, and so multiplying a matrix by the inverse of another matrix is like division and the resulting matrix would be like the ratio of those two matrices. That might lead down several paths to sensible measures of fractional genus, the next paragraph describes one I'm familiar with.

Some matrices code for graphs (as in networks) and these graphs have a graph genus property: the minimum genus topology on which the graph can be embedded and still be planar (no overlapping edges). Graph genus is calculated in such a way that is it always an integer, but it is forced. There is another way to calculate graph genus such that it agrees with the standard calculation for all integer valued graphs, but reports fractional values for graphs "in between" (see a future post "Graph Genus as a Measure of Planarity"). Fractional genus is actually meaningful here (has several important mathematical properties), but it may not be the meaning we thought or hoped we’d find.

But the network version of fractional genus might be quite insightful and it's just hard to recognize. Or it might be a little off, but along an interesting track to a useful and insightful interpretation of fractional genus. Or it might be useless garbage - a desperate swing in the wrong direction. I don't have the answer or even an excellent answer to whether the concept of fractional genus can be put to good use. But considering how broadly applicable topology is to conceptual and physical systems, it seems to be a potentially fruitful area of inquiry that I hope some others will endeavor to pursue. ]]> Methodology http://complexityblog.com/nucleus/index.php?itemid=102 Wed, 18 Apr 2007 19:12:06 -0700