Divisão de Observação da Terra e Geoinformática

Essa página está em modo somente de leitura. Você pode visualizar a fonte, mas não alterá-la. Informe-se com o administrador do Wiki, caso você ache que isso está incorreto.
=====Generative Social Science: Studies in Agent-Based Computational Modeling=====
|Joshua M. Epstein|

{{ http://leg.ufpr.br/~pedro/figures/epstein-generative.gif}}

====Introduction====

I hope the book demonstrates that the agent-based generative approach can be  explanatory even in [...] cases [...] where 
"the equilibrium approach," [...] is either infeasible or is devoid of explanatory significance. (p.xiii)

Every computational model is a computer program. Therefore it is Turing computable and then it can be described by a 
recursive function (gargantuan in size and imposingly complex, but they exist). Agent models are expressible as
equations. Which representation (equations or programs) is most illuminating?
Agent model is immediately intelligible as such. (p. xiv)

====Agent-Based Computational Models And Generative Social Science====

I use the term “computational” to distinguish artificial societies from various equation-based models in mathematical 
economics, n-person game theory, and mathematical ecology that (while not computational) 
can legitimately be called agent-based.

The agent-based model [...] is especially powerful in representing **spatially distributed systems** of **heterogeneous**
**autonomous** actors with **bounded information and computing capacity** who **interact locally**.

^ heterogeneity        | individuals may differ in myriad ways |
^ autonomy             | no central control over individual behaviour |
^ explicit space       | the main desideratum is that the notion of "local" be well posed|
^ local interactions   | interact with neighbours in the space|
^ bounded rationality  | bounded information and computing power|

Even if one conducts a statistical analysis over some distribution of runs - using different random seeds - each run is
itself a deduction. From a technical stand point, generative implies deductive.

[talking about sugarscape] when an initial population of [such] agents is released into an artificial 
environment in which, an with which, they 
interact, the resulting artificial society unavoidably links [lots of social areas]. 
Because the individual is multidimensional, so is the society.

Equations vs. ABM: equations only describe the global behaviour of the model, and it cannot conclude how the 
model evolves to that situation. //it devoid of explanatory power depict (?) its descriptive accuracy.//


====Remarks on the foundations of agent based social science====

one more feature of abm:

^ non-equilibrium dynamics | emergence of macroscopic regularity comes from decentralized local interaction |

The author uses Godel's incompleteness theorem to argue that not all equilibrium is computable. Also, following 
computational complexity, some of them would require too much effort to be solved. //These considerations, 
when combined with powerful psychological evidence, cast severe doubt on the rational choice picture as the most 
productive idealization of human decision-making, and serve only to enforce the bounded rationality picture insisted by Simon (1982).//

Empirical ABM can be seen as induction over a sample of realization, each one of which is a strict deduction, or 
theorem, and comparison of the generated distribution to statistical data. This differs from inductive survey research 
where we assemble data and fit it by aggregate regression, for example.

====Non-explanatory equilibria: an extremely simple game with (mostly) unattainable fixed points====

In many quarters, "explaining an observed social pattern" is understood to mean "demonstrating that it is the 
Nash equilibrium of some game". 

The game: players move in a 1-D space to achieve satisfiability. One moves 
to the rightmost cell if there is someone lower than him in his right. 

Most of the equilibrium points of the presented game are unattainable (as the number of players grows), given any 
initial configuration. If someone finds out in the real world one configuration which belongs to the equilibrium set, 

in classic social science she would argue that the players have followed  that game, what is indeed false, because 
that equilibrium is unreachable.

Implicit claims that equilibrium analysis is explanatory or predictive should be challenged and require the most careful 
defence. A successful defence of any such claims must include a demonstration of attainability on 
the scales of interest, by agents employing plausible rules.

====Coordination in Transient Social Networks: An Agent-Based Computational Model of the Timing of Retirement==== 

Two related theoretical issues:
  - the connection between individual rationality and aggregate efficiency
  - role of social interactions and social networks in individual decision-making and in determining macroscopic outcomes and dynamics.

Perhaps the main issue, then, is not how much rationality there is at the
micro level, but how little is enough to generate macro-level patterns in
which most agents are behaving "as if" they were rational, and how various
social networks affect the dynamics of such patterns.

In 1961, Congress reduced the minimum age at which workers could claim
social security benefits form 65 to 62. Collected data shows that it took
nearly 3 decades for the downward shift from 65 to 62. If agents were
rational, they wouldn't have taken that time. With social network and imitative
dynamics, very little individual rationality may be needed for society as a
whole ultimately to exhibit optimal behaviour.

Three types of agents:
  - **Rational:** retire at the earliest possible age allowed
  - **Random:** retire with probability p each period, once they reach the age
  - **Imitator:** retire when the majority of its neighbours is retired

====The Emergence of Classes in a Multi-agent bargaining model====

The paper studies the contrast between two types of norms:
  * **discriminatory norms:** allocate different shares of the pie according to gender, age, race, etc.
  * **equity norms:** do not so discriminate
How such classes can emerge and persist, given a norm-free, classless world initially.

Evolutionary game theory and ABM. Each person has tags (merely distinguishing features) and play the Nash Demand Game
(each player gets his demands if the sum of the two demands is not more than 100. players may demand 30, 50 or 70).

A **social norm** is a self-perpetuating state in which players' memories, and hence their best replies, are 
unchanging. The agent plays the best reply according to his memory, or a random choose according to an error probability.
There are two regions of the state space - one **equitable** (50,50), the other **fractious** (30,70 and 70,30) - that 
are very persistent: once the process enters such a region, it tends to stay there for a long period of time.

The authors extend the model to best replies according to the tag of the opponent. The model may evolve to different
strategies against each class. Various kinds of social orders - segregated, discriminatory, and class systems - can
also arise through the decentralized interactions of many agents in which accidents of history become reinforced
over time (a path-dependence dynamics).


====Zones of Cooperation in Demographic Prisoner's Dilemma====

The question: why rational choice leads to defecting in the PD?
why do we need a one-round memory to attain cooperative behaviour?

Demographic games seems an appropriate name for this class of models because
they involve spatial, evolutionary, and population dynamics. (p. 203)

Agents move around this space, interact with neighbours, and have offspring.
They are not fixed cellular automaton sites, as in Nowak and May 92, 
Feldmand and Nagel 93, and others). Each agent is an object whose main attributes
are vision, wealth, age and strategy. 
Behaviour: **choose a random unoccupied site within your vision; go there and
play your strategy against each neighbour.
Payoffs accumulate. Since our payoff matrix has negative entries, an agent's
accumulated wealth may go negative. In that event the agent "dies".**
If an agent's accumulated wealth exceeds some positive threshold and there is
an unoccupied site within the agent's vision, the agent has one offspring. (p. 204)

Perhaps it is worth emphasizing that, in adopting this assumption of a fixed
agent strategy, we are not claiming that human strategies are literally
hard-wired genetically. Rather, for modelling purposes, we are assuming that they
are culturally transmitted from parents to children with high fidelity, like
certain religious or ethnic affiliations, tastes and native language
(on cultural transmission, see Cavalli-Sforza and Feldman 81, Boyd and Richerson 85)
(p. 204)

**Cooperation can emerge and flourish in a population of tagless agents playing zero-memory
fixed strategies of cooperate or defect in this demographic setting.**

Variations: maximum age and mutation. some results are similar to predator-prey cycles
Cooperators function to separate defectors from one another.

For non-negative payoff, defectors spread while cooperators sharing until extinction.

Cooperation acquires selective advantage as selection pressures increase. Or, more 
prosaically, necessity is the mother of cooperation.

====Learning to be Thoughless: Social Norms and Individual Computation====

To me, [...] one of the more remarkable aspects of homo sapiens is how much of our social
behaviour is entirely thoughless.
Individual though is inversely related to the strength of a social norm.

The agent follows a boolean "norm". She looks to her neighbours in order to update her norm. 
If increasing the observation radius changes the statistics then she increases the radius. 
Otherwise, if decreasing the radius does not change the statistics, she decreases it.

If you had asked a 14th Century European if the earth were round or flat, he would have said
"flat". If, today, you ask the average American the same question, you will certainly get a
different response: "round". But I doubt that the typical American could furnish more compelling
reasons for his correct belief than our 14th Century counterpart could have provided for his
erroneous one. Indeed, on this test, the "modern" person will likely fare worse: at least the
14th Century "norm" accorded with intuition. Maybe we are going backward! There is a model
with a re-equilibrium after a noised period for capturing that basic phenomenon.

In the rest of the chapter, the author investigates different levels of noise.