Hi James,

Fundamentally an old idea: what about power measured as freedom/ or the force of will: 

The ABCs of power:

Operationalized  at the affective level: degrees of freedom in joy/passion

Operationalized at the behavioral level: degrees of freedom in mobility

Operationalized at the cognitive level: degrees of freedom in imagination

You do come up with interesting ones. For some contexts you could "give" your agents power. If x and y fight then x has more power and is therefore more likely to win. But here it seems that you want some behaviour to exist in which power is "manifested" and can then be measured. If I "have the power" to set the rules by which tax is paid then clearly I have "more power" than the poor bloody tax payer who just pays up or risks getting hammered. But is having the power to change the rules of cricket more or less power than being able to change the tax rules? That seems to be what you are hinting at and if so the examples suggest why there probably can't be a universal metric for power. If the rules are being used for "obvious" personal gain then power can be associated with metrics like wealth (for example) but not all power is like this (though you might build an abstract model in which wealth _could_ stand for power because the domain dealt with competition for wealth). But power also manifests in the actions that people take in the light of it. So I wouldn't call the CEO of the company I work in a fool (even if he is) because I know he could get me sacked. Power can also be used, for example, to evade responsibility (Berlusconi) or performance (yes men).

I guess what all this slightly rambly stuff is heading to is that you might need to define power not in terms of a metric but in terms of the ability of x to realise their goals (whatever they happen to be) relative to y in a particular social context. But that still won't quite do because if I persuade y to help me shift a piano then I don't have "power" (the power of persuasion?) over y. Power means that I got y to do what I wanted even though y continued _not_ to want to do it. (With the piano, y _did_ want to for payment later, to be a pal or whatever.)

So you don't "compute" power, but you observe it in different circumstances where agents have different capabilities (which may include things like institutional roles - "I can have you shot" - depending on your domain of application) and want different things. (x has a lot of power when something happens even if everyone else doesn't want it. y has little when something doesn't happen even though quite a few other people want it too.) This approach also crosses the boundaries of domain specific power. Can the Chairman of the MCC get David Cameron to change the tax rules or can David Cameron get the Chairman of the MCC to change the rules of cricket? We don't have to decide if cricket or tax is in some sense "a bigger deal" to have power over.

Your challenge may be to decide whether power should be an "attribute" of agents in your model or whether it is an observable "behaviour" that arises from things like coercion, control over status and so on.

Some 40 years ago, I published a definition of power that went something like this:  A has power over B if A can cause there to be consequences that B finds unacceptable but B cannot cause there to be consequences that A finds unacceptable.

However, this is just a binary relation and is not transitive.  I am quite certain that some further individual C could have power over A whilst B has power over C though the nature of the consequences would have to be different.  Example: A could force B to lower its prices because A has resources so much greater than B that B would necessarily lose a price war.  B could have the same power over C.  But C might own a patent that A requires in some production process but B does not.  So unless you restrict your model to cases where all agents are engaged in the same activity, it would not be prudent to rely on transitivity in the power relations.

You might be able to concoct some measure of power based on the numbers of individuals over whom A has power in the above sense which would make a head of government (e.g. the British prime minister) more powerful than the head of a sporting body (e.g. the head of the governing body of cricket) but there is still nothing to ensure that relationship is transitive.  If it isn't then I don't think there can be a power metric.

Whether there is a metric or not, it seems to me that you are going about this in the wrong way.  If your model produces outcomes in which some definition of power (such as, though by no means necessarily, mine) is useful in describing those outcomes, that would be good.  But to impose a definition of power on your model and to assume that power is an attribute of the individual seems to me to be misguided.  Power is a feature or determinant of social interaction.  Consequently, it is not possible to specify a quantity of power for any individual.  Power is probably not well represented by a number but rather by a set of specific social relations.  Individuals obviously have attributes such as wealth that can give them power but power remains a factor in social relations.  I don't see what is to be gained by giving it a number but, if you must, then I think you have to restrict the set of possible actions of your agents so severely that the model will not describe facets of any actual social environment.

One thing you haven't told us is WHY you want a metric for power.

Hi James,

As a followup on your clarifications...and its connection to the ABCs of power...

Perhaps agents can be imbued with power through a nonconformity and nonmutuality rule. At the extreme, these two 'tenets of power' may result in the following emergent properties:

Brownian, random individuals, wondering about aimlessly without will or purpose. Which is unlike what most people would associate with power at the agent level. On the algorithmic approach to the dynamics power,the repelling effect of individual agents may be an index of power(loneliness is often the price of power). Over time the spread of this entire mass, analyzed in terms of its "diaspora effect", may determine the resilience of that system (all eggs are not in a single basket). This hold if each agent is endowed with immortality.

In the case of willful, purposeful agents,some kind of fitness rule implemented as a modifier to the above scenario, may result in an interesting fitness landscape defined mainly by agents who are: the most traveled, charismatic and creative.

As an aside: rich lessons on agent-power issues can be gleaned by following the premise and plot of the "hunger games" franchise. Also a good compendium of power(taxonomy) can be found in Robert Greene's The 48 Laws of power-albeit more of the Machiavellian kind (suited for agents?).

http://www.amazon.com/The-Laws-Power-Robert-Greene/dp/0140280197

The amount of money the agent control the spending per year.    

Many thanks for range of interesting and helpful answers which I shall digest and respond to later.

Immediately a reply to Scott’s request to be told more what it’s all about!

I am writing a program (MoHAT-T) to generate and run ABSTRACT ARTIFICIAL SOCIETIES (within machine) looking for a match to the MAJOR characteristics of human society dynamics over the past 100,000 years or so as seen in the archaeological record etc. The search for the right model(s) needs effective heuristics, of course.

One of the major characteristics of long-term global human society is surely transient power hierarchies which have been prominent for at least 10.000 years – indeed all the way back to the great apes if you count family groups and dominance therein.

So my wider question is: given an “executing” artificial society, how can I ALGORITHMICALLY detect any powerful agents and power hierarchies that arise in a way that even a simulation-friendly social scientist will not instantly dismiss as an abuse of words. This is hard! [Of course, just running one’s eye over the output is a non-starter!]

For a little more context and detail see my downloadable RG publication (actually a working draft): “Five problems in the agent-based simulation of human society’s … trajectory”.

Cheers

Jim

I agree with Edmund. For such a simulation, power should not be a property of agents but an emergent phenomenon. So you can only observe/compute power by running simulations and detect some patterns. In this sense, power of an agent in some context might be the survival chance of that individual's preferences for that context. To give an example, let's suppose you have two different agent types Male and Female. Males prefer an ideal offspring value of 6 at average, but Females prefer this number to be 2. Males also prefer that the dominant strategy for food gathering is hunting but Females prefer this to be agrıculture.  Of course, these agents interact in many various ways and facets according to your model.

You run your simulations and consistently find that in the long term and for some set of  inital parameters, the average offspring valuee settles around 3 and hunting wins as the dominant food gathering strategy. Then, if I was a social scientist you can convince me that for such a model and for such parameters Females have more power for determining the ideal offspring value but Males have the power of determining the food gathering strategy. Does this make sense for your purpose?

To extend on my last example may be it is not Males or Females that determine a system behavior but some other parameters. So child death rate or some dietary parameters may in fact be the real reason for ideal offspring value or dominant food gathering strategy. So you need to check what happens if you keep all other parameters same but just swap Male and Female preferences. If Female's preference still wins then you can confidentally say determining X is Female's power.  

Power is a function of your location in a network of actors.  The links between the actors might change from one set of interactions to another.  Therefore you may have more power in one set interactions than in another.

I wrote a white paper on "Power in Networks" a while back, using two popular social network analysis metrics to investigate how power changes when the pattern of connections change amongst a set of actors.

http://www.orgnet.com/PowerInNetworks.pdf

Enjoy!

I apologize for the earlier missing message, which should have read:

First, I agree with Scott that power relations can be "intransitive" such that A>B, and B>C, does not necessarily eliminate some C>A, and this is known formally as an "intransitive tournament."

Second, I agree with Valdis' that what is known as betweenness in social network theory would be a useful measure of power, and it seems to be very similar to what Volkker describes as "brokerage."

Impressive project, I really am curious about it. For the sake of brainstorming, I would suggest that since power has many facets, and there are kinds of power such as legistlative, financia, reputational, traditional, psychological social types of power, i would suggest to start with a definition and a sound categorisation of power Types and their relative impact on the general Power, you are planning to compute. Some suggestions emerge in my mind. For instance number of friends and their Power levels can have an impact on one unit.

from a computational and algorithmic perspective, you may also consider the Sensitivity & Importance criteria applied to causal/relational models. the sensitivity of a given element in a model is the measure of how much change at the top is experienced by a small change at a given element. the most sensitive elements are naturally critical to the computation. sensitivity is an inherent quality of a model structure that depicts the relations and topology. The Importance factor is a matter of parameter values during evaluation and more related to the proportion of the top value (in your case power) contributed by a given factor/variable.

The concept of power then is a relative question i.e. assuming a social network has a power/influence of x in a larger context, then the sensitivity of contributory factors to x as well as their importance can be inferred from a model based depiction of the logical combination of influencing factors. 

good luck

I would suggest Spectral Ranking theory, used in many areas (e.g. Google's PageRank) for determining the most influential nodes in a network. It is a quite simple but powerful operational definition and there are many available implementations, depending on the desired algorithm and network type (Eigenvector centrality, Katz's index, etc.).

Hi James!

I would say that the concept of (social) power is basically linked to the idea of the ability to control the flow of "scarce" ressources in a given network. Hence, persons occupying strategically important positions - sometimes also called "brokers" (R. Burt refers to "structural holes" as the structural complement; see also his book on "brokerage and closure") might exercise power. Such actors possess instrumental social capital; they provide scarce ressources to others and receive obedience in return.

I published an article in Connections (that you can find on reserach gate) on a way to concepzualize different forms of social capital and, hence, to identify powerful broker roles in social networks based on a distinct concept of "betweenness":

"Measuring the Social Capital of Brokerage Roles"

Kind regrads, Volker

For instance, in a network with multiple agents with different wealth levels you could consider individual wealth relative to total wealth (moreover, you could also measure relative wealth with respect to the relevant network, that is a geographical area or a social group). For example, an agent can directly or indirectly (through credit) hires other agents for her purposes (for instance, production) and makes a profit and accumulate wealth. In this case we may say that economic power depends on relative wealth, because it is more likely that a richer agent becomes an entreprener while a poor agent plays the role of worker. In other words, richer people are more powerful in managing and remaining rich, due to the characteristics of the wealth accumulation process itself. This can be related with the emergence of a power law tail in the wealth distribution and the statistical properties of the network . Hope my simple story can be of some help. Best, Alberto

THANK YOU for the variety of helpful answers given to my original question. This brief reaction, from my particular point of view, I may elaborate in more detail later.

Overall the answers are diverse and most are not computational in content. Most are at the stage of seeking a definition of “power” in non-computational terms. But I cannot make progress in my own project (MoHAT-T) with a definition of “power” that has terms in it as ill-defined as “power” itself. [Elias Cooper makes this point.]

The answers given touch upon, amongst other things: the insufficiency of integers, transitivity, emergence, power In relation to network structure, and different types of power including over different groups.  I agree with almost everything said. All good stuff to think about but not always going in my direction…

I see three HEURISTIC REQUIREMENTS for the computational notion of power (and of a “power hierarchy” – see MoHAT-T project cited in my earlier answer) that I need:

1.       It must be algorithmic i.e. straightforwardly translatable into computer code

2.       It must have plausibility with respect to our primary concept of power

3.       It must be deployed in a computational context that is structurally and processually acceptably analogous to the real world

MY PRESENT LINE OF ATTACK

I am led to feel by the answers given in this thread that power is a complex concept. However, the online (Oxford) definition seems relatively simple:  

noun: power:

(1)   the ability or capacity to do something or act in a particular way

(2)   the capacity or ability to direct or influence the behaviour of others or the course of events.

Is this too naïve to work with? If it is not too naive, how abstract should we go?  Preferably abstract enough to rise above the confusing detail!

Hence my current attempt at a usable abstract computational/algorithmic interpretation of power:

Assume a set of agents

Agents have dynamic state, can perform actions, and can receive messages from other agents and send messages to other agents. Actions and messages are merely tokens i.e. they have no associated content or denotation but different messages have different effects on a recipient agent’s state. Actions when performed effect nothing in themselves, but their performance does contribute to the definition of power. [This corresponds to standard abstract mathematical definitions.]

A collective action is a set of individual actions by multiple agents in a particular temporal arrangement.

Then the POWER of agent A (available to it, not necessarily exercised by it) in the agent set at a specific time T0 is the number of distinct collective actions that A by itself can cause to occur within the next N time units, starting from the current network state at T0, each such causable collective action being achieved by A sending a suitable set of messages to other agents.

A relevant collective action is a collective action that appears in a pre-specified list, R, of collective actions. Importantly, the “relevant collective actions” of R can be used as a means of making the agents’ environment realistic (see requirement 3 above) but this is not simple.

By this definition, the power of A depends upon the total state of the agent set (except A itself) at time T0, on N and on R. So the power of A might be written as  Power[total_state(T0),N,R]

The foregoing can be seen, I think, as akin to but a generalisation of Xavier Vila’s “Mafia” approach.

SOME COMPLICATIONS AND UNCERTAINTIES

But what if in practice A WILL NEVER send certain of its available message combinations because its internal state cannot actually trigger the required message set? Is that a problem?

The agent can actually cause the collective action, in the light of other agents’ actual internal states, but it may in no sense “know” that it can or is doing so. “Know” is not defined. Is that a problem?

What if the “environment” implicit in the “relevant collective actions” list is NOT “realistic”? Ok -- it might still be interesting.

What if the power ratings of the agents are highly unstable in time? Is that a problem or is it sometimes to be expected?

What can be said about the power of veto (e.g. permanent members of UN Security Council) from this perspective?

All further thoughts on all this are most welcome – extending hopefully to an algorithm to recognise transient “power hierarchies” in a dynamic “artificial society” [See MoHAT and MoHAT-T as briefly described in my publication "Five Problems..." on RG].

Thanks again -- Best regards

Jim

Hi Jim,

My impression from your original question was that the network will evolve and at some point you wanted to measure (calculate the "size" of) the power of an agent and possibly compare it to some other agent's power. I think, this gave way to the answers which argue power is a complex thing and it is not easy to measure it. Your last comment is more about how you can define/implement power of an agent and how you can implement the ripple effect of each individual's power on system's overall behavior.

In that sense, I think your abstraction may work fine but I think power is not just about the number of actions available to an agent but also what those actions are. When you think about Role Playing Games for instance each player will have a unique skill and try to maximize this skill while playing. Or in board games, you may become more powerful by achieving the right to perform certain actions while the game progresses (number of actions available to you may remain constant though).

About the complications you mentioned:

* At some point T0, agent X may not use all available actions due to the state but I think if it can use them, it should use them. Otherwise being more powerful will not have a meaning. X may have insufficient knowledge but I am assuming it has a goal (or goals) and will try to achieve this goal through its actions. So at least each agent  should "know" what its goals are and have some notion of how (through which actions) that  goal can be accomplished.

* Power ratings may be unstable, I think this should be expected for a simulation like this. My real concern is power should be somewhat of a limited resource. How can you make sure that the total power doesn't grow too fast while power of individuals oscillate mostly incrementally?

* If you include the set of available actions as part of your power definition then you don't need to be concerned about veto. It should be just an action as others available or not to an inidividual.

Hopefully, these are not too far off from what you are imagining. May be I should first read your working papers before answering :).

You might draw an analogy from physics where "power is equivalent to an amount of energy consumed per unit time" (taken from wikpedia). 

Taking this as a starting point you could model the effect an action available to an agent has on others in the network. Where the 'power consumed' is the degree to which future actions of these others are limited by the actions of the agent.

Example: If an action affects no one its power is 0%. If an action fully determines the state (or next action) of an affected other agent then its power over that agent is 100%. By summing up the percentages over the number of affected individuals for each _independent_ action available to an agent you get a total power rating of that agent.

In my opinion this should further be scaled by the amount of power each affected agent has (affecting future actions of a powerless agent by 100% is less 'powerful' than affecting future actions of a very powerful agent by 50%) 

It should also be scaled by how long that action will affect another agent (killing an agent will affect it forever - which is a much more powerfull action than something that will just affect its actions for the next day, only)

>> To Izzet Pembeci (and all interested)

In his most recent answer Izzet raises a number of important points and I think the best way to respond to them is by explaining a little what is my immediate goal right now on this research front.

My immediate objective is to run an abstract artificial society (within a laptop, programmed in C) over a period of maybe 10,000 time units such that pyramidal power hierarchies (PPHs) - think chiefdoms  or extended households with a top member- come and go emergently (i.e. they are not explicitly programmed to be there – compare the EOS project – see relevant EOS publications and also my “Five Problems..” working draft both here on RG).

Note: I am thinking here of perhaps a thousand agents in a shared simple environment, and PPHs involving say  tens of agents at a time and each lasting maybe a hundred time units.

Clearly, definite decisions must be made about the agents’ architecture and processing and their joint (spatial?) environment.  For the agents I favour production systems because the simplicity of this architecture, although somewhat laborious, facilitates algorithmic manipulation of agents. NB production systems are potentially fully cognition capable.

More difficult is to decide the operational definition of a PPH – I’m only part way there – the aim is to devise and implement in the machine an algorithm that will detect (and report) ALL the PPHs that come into being during the lifetime of the artificial society. [There is potentially a massive computational load involved, but I am ignoring this at the moment.]

What kinds of criteria are relevant to recognising a PPH? My current thinking suggests: (i) a discernible set of agents with relatively frequent inter-agent message-passing, (ii) a pattern of inter-agent communication reflecting the standard notion of a pyramidal hierarchy, and (iii) power assessments that indicate that the agents at the top of the potentially recognised PPH do indeed have more power than those lower down.

Note that the meanings of the components of agents’ internal states (e.g. working memory elements in a production system) are unknown (unless very detailed analysis is undertaken), as are the meanings of messages being passed between agents. This is because such things are NOT directly specified by the programmer, but are emergent from the lowest level of the artificial society which, of course, is specified. Compare, perhaps, the human brain…

A final point; the foregoing may be characterised as a “bottom-up” approach. But it is also possible to work in a more focussed “top down” way, starting with PPHs located in time and space and which now are indeed specified by the programmer, and to investigate experimentally what the content of the agents must be to generate them.

I hope this is of some interest.

Best regards

Jim

If there is a strong correlation between agent A passing a message to agent B and then agent B subsequently either

a) passing message(s) to agents C (,D,E,...) (excitatory action)

or

b) passing less messages than without having previouly received a message from agent A (inhibitory action)

then you have a high power of agent A over agent B.

Check out 'reinforcement learning' algorithms from neural networks. The amount of reinforcement such a learning algorithm would give to a particular branch should correspond to the power of the source agent over the target agent.

In addition to the paper I proposed in my first contribution, I think that the use of "triadic role patterns" allows for a computational solution of "power" as defined by certain triadic positions (such as "brokers"). I remember a software (DOS stand alone) done by Ronald Burt, called STRUCTURE that provided as one option the analysis of a census of triadic role patterns. Of course, I don't know your data in detail and I can't tell if you can get information about connections on the level of triples...

Hi Jim, after reading your comment and draft paper I think the operational definition of a PPH will largely depend on how you will implement the rules sets of agents. If you can achive the following:

(a) rule sets will be shaped to display effective resource capture in non-trivial situations

(b) creating a rule set to generate a trajectory with the required properties

Then power assessment (iii) may be done by measuring how resource capture differed between individuals and how much their actual trajectories matched the intented trajectory of their rule sets. The required properties you mentioned may be also configured to have some indirect (i.e. emerging not explicitly programmed) consequences on the power of an agent. I also believe MoHAT should try to mimic humanity's natural social networks to some extent. Social network of an agent should evolve during the simulation (i.e. grow, shrink, nodes added/removed, link weights change) dynamically without the influence of the programmer but for each agent there should be a network to begin with or formed quickly at the beginning of agent's lifespan. Of course this social network will somehow effect the behavior of the agents. For instance, messages may be received/sent more proportionally from/to one's social network. Then you may detect (i) and (ii) by the help of this social network and what you have decided for (iiii).

So I guess what I am trying to say is before delving too much into the question of  how a PPH may be detected, it may be wiser working out the details of how the simulation will be driven, how the system will evolve in general. Once you figure out the structure of the rule sets, their overall role and the communication scheme between agents then definition and computation of PPHs can be done based on these.

El asunto del poder puede ser visto en el elemento de control, por tanto: el agente computacional ... ¿ejerce control "sobre" que y cuántos" elementos de la red? . Desde allí la def. operacional puede tener un camino.

Sugestions:

http://www.google.com.co/imgres?imgurl=http://image.slidesharecdn.com/conceptosbordieu-121106110455-phpapp01/95/conceptos-pierre-bourdieu-4-638.jpg%253Fcb%253D1352221551&imgrefurl=http://es.slideshare.net/villamarcos/conceptos-pierre-bourdieu&h=1020&w=638&tbnid=Uaf7K3WO5Zo7mM:&zoom=1&docid=xHEhytWGMAQBOM&ei=jsodVYPENpDdsATQl4DgCg&tbm=isch&ved=0CCQQMygLMAs

http://www.google.com.co/imgres?imgurl=http://image.slidesharecdn.com/bourdieupierre-razonesprcticassobrelateoradelaaccin-140523203831-phpapp02/95/bourdieu-pierre-razones-prcticas-sobre-la-teora-de-la-accin-13-638.jpg%253Fcb%253D1400895783&imgrefurl=http://es.slideshare.net/RonaldJosMarreroMedina/bourdieu-pierre-razones-prcticas-sobre-la-teora-de-la-accin&h=1051&w=638&tbnid=qFFwwnrh9QBWcM:&zoom=1&docid=HSVZ6aDItmpTbM&itg=1&ei=jsodVYPENpDdsATQl4DgCg&tbm=isch&ved=0CCcQMygOMA4

Many thanks for further posts which are both interesting and relevant. This reponse is worded as a reaction to Izzet P, but hopefully will be of interest to others, and it may clarify a little more the exact problem I am currently trying to solve.

Izzet wrote:

"If you can achieve the following:

(a) rule sets will be shaped to display effective resource capture in non-trivial situations

(b) creating a rule set to generate a trajectory with the required properties"

As regards these two requirements (a) and (b), I agree both are needed

Alas requirement (a) can easily become, in effect, a straightforward BOTTOM-UP requirement for a generally intelligent agent – which currently no one knows how to design (see e.g. my MIAP agent publication here on RG and also publications re EOS project).

Hence my current ongoing investigation is actually TOP-DOWN and asks the question – if the social trajectory has certain specified properties (e.g. transient PPHs) and agents do act to acquire the basic resources they need (simply-abstractly defined), what are the possible rule sets the agents can have? (One of) These rule sets are to be identified algorithmically!  And what cognitive and social interpretations can be put upon these rule sets (the “Psynigma” question)?

Requirement (b) is easily met by, in essence, creating a rule to handle every creation of every required working memory element at every time point on the trajectory. Of course this gives a potentiallvery large and wholly uninteresting rule set – I call it the TOTAL rule set.

This TOTAL rule set is the start point for my current TOP-DOWN approach – the next step is to ask the following question: What computationally precise reductions and modifications of the TOTAL rule set can take us to a minimal rule set (in an appropriate and precisely defined sense) which generates exactly the SAME TRAJECTORY as TOTAL?

I FIND THIS QUESTION HARD AND I HAVE ONLY MADE LIMITED PROGRESS WITH IT SO FAR... IS IT MISCONCEIVED?

As regards the remainder of your remarks, Izzet, I certainly agree that the power of an agent is a dynamic emergent property – but to be OBJECTIVELY measured – and I still like the definition of power I gave earlier but agree that there are a variety of interesting and important alternatives that can be considered. Similarly, social networks may be objectively detected and if they exist will be emergent and dynamic.

One might REQUIRE the property that every agent is part of a social network – but this possibility must comply with the answer to the core MoHAT question: what are the MAJOR properties of humankind’s sociocultural trajectory over the past 100K years (e.g. population increase, PPHs, endemic warfare …).

And everything must be pitched at a high level of abstraction or nothing will be computationally tractable! Of course, this may or may not prove impossible to do in a way that is relevant to human prehistory and history. We have to find this out...

Best regards

Jim

Further to my answer here of 17 days ago, current work is aimed at implementing a “proof of concept” demonstrator. An artificial society, ArtSoc, using production system agents (i.e. a PS-MAS + minimal spatial environment), is to be exhibited (in machine, programmed in C) in which two spatially distinct pyramidal power hierarchies (PPHs) emerge in parallel and are then merged by the joint action of a subset of their agents with consequent increase in cooperation. The PPHs will be based on the definition of “power” I sketched earlier [without prejudice to this definition being later replaced by something more generally convincing!].

ONGOING SUCCESSIVE IMPLEMENTATION STEPS

·         Craft a history for the ArtSoc in which two distinct PPHs are specified

·         Generate an equivalent joint rule set, R, (for each agent) from the history via minimal trigger identification

·         “Reduce” the rule set R to R*, with added generalisation and explore cognito-socio interpretations

·         Implement a notion of joint agent action sufficient to provoke PPH merging

·         Run demonstrator

Some of this gets complex and, in particular, involves heuristic search!

PRESENT STATUS (April 25, 2015)

Design provisionally complete, C coding ongoing

When the demo is running, i plan to upload the code so that anyone interested can knock holes in it...

Best

Jim

I'm prepared to believe that Python is a neat language. But I am more familiar with C, there is useful existing code available, and there are the usual syntactic sugar v efficiency arguments. Not a big issue, really...

Hi Dr James

I don't know if i well understand your query and correct me if not, but i suggest that power/energy can be computed relatively to the result you obtain. I mean if an agent energy is destructive its energy is not helpful and more the extent area is important the less is its energy, contrairy to a constructive energy: more the extent area is important and more is its energy.  

Am i outside the area??

Hi Hakima Mellah

I think you introduce an interesting consideration concerning the difference between destructive and constructive energy and hence different types of power. This is not part of my own current thinking. Perhaps there are others who will wish to comment.

JIm

Hi,

Power can be operationalized as a "stable" propagation.

Don´t forget the process of accumulation of capital (over time). 

http://www.morganclaypool.com/doi/abs/10.2200/S00527ED1V01Y201308DTM037

Because the concept of “power” is a social construction, you could go crazy trying to write a definition that satisfies every possible (and impossible) considerations. I think you should listen to yourself and stick with the simplest: “the capacity or ability to direct or influence behaviour on others”; Which thinking about it, fulfills many considerations; e.g. the “person about to detonate a bomb that will collapse a dam and flood the city” has no “power” unless “he” communicates with another person and force him to do something under the threat of flooding the city. Otherwise, the power of that person will no be materialized until he detonates the dam, forcing all the people in the city to do something: “run or die”.

So, I think, you should embrace the concept of “power” as a result of an interaction between two agents. In a simple way:

When the agent A sends a message to agent B (A telling B to DO activity F); if B decides to obey (do such activity), then A has power over B. The measure of “power” of an agent on its social network will be determined by the amount (number) of agents over whom it has power.

The simulation process should then include an agent X randomly choosing an agent Y. Then X ask Y to do something. If Y obey then ('Power' of X) P(x)+=1. After that X will repeat that procedure until it ask the same to all agents in it's social network. This behavior will be executed concurrently by every agent included in the simulation, and when they have all gathered the information of his “capability of power”, it's the end of T1 (time/lap 1). The second round (T2) could include a strategy that involves the results of T1, thus emulates a “status quo” typical in real societies. I may think of some ways to do so, but it should match the structure of your models.

Does this approach makes sense to you. Did you already use a similar/different one. Only after writing this, I realized that the last post was written a year ago.

Anyway, hoping this helps.

Daniel.

Hi James,

I propose to refer to concepts from Social Network Analysis in the sense that a powerful actor is somebody who controls social contacts to others by being the (asymmetric) connecting link between otherwise unconnected actors (this can also be operationalised through triadic role patterns; see my paper on "Measuring the Social Capital of Brokerage Roles" in Connections) - such an actor should of course be connected to persons that are themselves highly connected through asymmetric links (see Bonacich's concept of "Eigenvector Centrality"). Regards, Volker

My thanks to Daniel and to Volker for their recent answers to my question - both answers are helpful and thought-provoking.

Daniel proposes: "The simulation process should then include an agent X randomly choosing an agent Y. Then X ask Y to do something. If Y obey then ('Power' of X) P(x)+=1. After that X will repeat that procedure until it ask the same to all agents in it's social network." 

I fully understand this line of thought but in my computational context (nearly everything emergent!) there is a difficulty -- the meanings of the messages between agents in the system are NOT known but are to be discovered. So it is not (immediately) possible to say that one message is a request, say, or another an information provision. Meanings are potentially discovered by reference to a distinction between agents and their collective environment.

My ongoing attempt to assess an agent's "power" (alas, slow moving and as yet clearly inadequate!) is embedded within my MoHAT/MABEEA project computer program as "height". (see MoHAT project here on RG)

Thanks again for your interest and help

Jim

Dear Prof. Doran,

People often quantify that by means of graphs and flux, sort of similarly to steady state analysis in Markov decision process, trying to calculate how much 'flux' flows from one person to the other.

I think maybe you can define an orthogonal measure of power by each agent (could be one, but should be orthogonal, like in CDMA codes for cellphones). Then you can model the interactions as a graph and calculate the steady state average. When you get the result you can decode how much of the component of each agent does make it to the average. That would be a measure of 'power' of the agent.

I hope that can help.

Cheers,

Edilson

on the power within a social network here's my offering for a definition;

"the ability of an individual to influence the members of the network and achieve the specific outcomes"

the degree of influence can be measured by the proportion of the network who respect and follow the specific individual and the second impact factor can be a probability density function based on the likelihood that a given request/demand/concept generates the expected outcome that the instigator wishes to bring about in the network

cheers

Ali