Proceedings
of the Nineteenth Annual Conference of the Cognitive Science
Society (1997). pp. 560-565.
Mahwah, NJ: Lawrence Erlbaum Associates.
Click here to get a postscript-version (142kb) of the paper.
A Cognitive Model of Agents in a Commons Dilemma
Nerb, J., Spada H., & Ernst A. M.
Department of Psychology,
University of Freiburg
D-79085 Freiburg, Germany
{nerb,spada,ernst}@psychologie.uni-freiburg.de
Abstract:
KIS (knowledge and intentions in
social dilemmas) is a process model of a
cognitive-motivational theory of acting in a three person commons
dilemma. The model provides an experimental tool to study how
ecologically harmful actions evolve in commons problems by having
differently parameterized variants of KIS interact with each other
and with human subjects. KIS models the application and acquisition
of ecological, social, and practical knowledge using a motive-driven
decision procedure. To test this model, 42 subjects played a
commons dilemma game in an unselfish or greedy social environment.
Both environments were realized by pairs of appropriately configured
KIS variants. Subjects did not recognize these co-players as being
artificial and judged their motives accurately. Subjects' behavior
in the unselfish environment was well predicted, however, in the
greedy environment subjects based their decisions more on the state
of the resource than was expected. To further test the model, we
constructed a KIS variant for each subject with respect to the
assessed individual motive structure and knowledge. These
variants played the game in the same environments. Their actions were
compared to the subjects' on both an aggregate and individual level.
We obtained good fits in the unselfish environment. Systematic
deviations in the greedy environment revealed that under this
condition behavior was more determined by ecological aspects than by
social comparison.
KIS (knowledge and intentions in social dilemmas; Ernst & Spada,
1993; Ernst, 1994) is a model that integrates assumptions about
problem solving, motivation, social cognition, and learning into a
cognitive-motivational model of an agent involved in a commons
dilemma. A commons (or resource) dilemma is a specific type of a
social dilemma (Dawes, 1980). It refers to a situation in which a
group shares a common resource (e.g., fish, water, forest, or clean
air) from which the individual member can harvest. If too many members
take too much from the common source, it is exhausted. Thus, the group
interest requires moderate harvests, but personal interests may induce
the individual members to harvest excessively (Van Lange, Liebrand,
Messick, & Wilke, 1992). The classical example of a commons dilemma
is where herdsmen involuntarily ruin a shared pasturage by adding
animals to their individual herds (Hardin, 1968). Two traps are
characteristic of such a dilemma situation. The first is a social
trap: Gain of an action for one, losses to all. The second is a
temporal trap: Gain of the action now, losses later. As a general
rule, the immediate gain for individuals exceeds their share in the
damage, which affects everybody in the community on a long term basis.
The group as a whole would be better off if everybody restricted
his/her resource use.
Commons dilemmas are typical for a variety of critical environmental
problems--such as resource depletion or pollution. Agents may be
individual persons, organizations, or countries as the quotation
illustrates:
The main reason for over-fishing is one familiar to economists. If
all fishermen restrained themselves, each would benefit from
enlarged stocks in the future. But no individual boat or fleet sees
any gain from holding back unless it knows competitors are doing the
same. (The Economist, 18 March 1995, p. 74)
The benefits of having a computational process model of a social agent
involved in such a dilemma are twofold. (a) Artificial social
environments consisting of these agents can be created enabling the
controlled variation of social conditions in experimental work.
(b) These artificial social environments can also be used to test the
theoretical assumptions underlying the model, i.e., the model is able
to assist its own evaluation in a non-trivial way, as will be shown
later.
Much work in the area of social dilemmas is based on a very restricted
minimalistic action scenario. A minimalistic scenario is
defined by two (N) parties, two mutually exclusive actions (defect,
cooperate), and four ( 
) possible outcomes (pay-offs) linked to
these actions. Usually, outcomes are held constant in iterated games.
Subjects' decisions (defect or cooperate) are often analyzed in the
framework of classical game theory (Dawes, 1980). Another venue of
research uses Monte-Carlo simulations (Axelrod, 1984; Danielson, 1992;
Glance & Huberman, 1994; Macy, 1995; Messick & Liebrand, 1995),
where two or more artificial players employ simple decision strategies
(e.g., tit-for-tat, win-stay-lose-change, win-cooperate-lose-defect)
or statistical decision rules in iterated games.
In laying down cognitive prerequisites for adversarial
problem solving and social problem solving in general, Thagard (1992)
complains about the neglect of social inferences (e.g.,
understanding intentions and traits, anticipating the action of the
opponent) in these approaches leading to two severe deficiencies: (a)
the theoretical narrowness does not provide a sufficient
explanation for behavior, nor (b) does the minimalistic
scenario realistically account for the broad variety of possible human
behavior in social conflict situations. These shortcomings enormously
restrict the validity of generalizations based on minimalistic action
scenarios.
In the following sections, we introduce a more complex and realistic
instantiation of a specific social dilemma (a commons dilemma), the
Fishing Conflict Game, and present a cognitive process model of the
behavior in such a conflict situation, which allows to predict and
explain behavior.
One instance of a commons dilemma is the Fishing Conflict Game,
the stock of fish being the resource, the harvesting the conflict
partners' actions (Spada, Opwis, Donnen, Schwiersch, & Ernst, 1987).
In the setting of this game, three players act as fishermen at a
simulated lake. They are instructed to pursue the goal of achieving a
maximum gain from fishing. A game consists of two phases, but the
number of rounds is unknown to the players beforehand. The first
phase of the game starts with the simulated resource (i.e., the fish
population) in the range of optimum propagation, whereas in the second
phase the simulated resource is in the range of suboptimum
propagation. The players are not allowed to communicate with each
other, but throughout the game each player is informed about the
other players' harvesting decisions. The fish propagation
rate is a non-linear function, which is not imparted to the players.
Overharvesting reduces the fish population, thus its propagation, and
subsequently, the possible long term gain. In the extreme, it may lead
to the extinction of the resource. From a psychological point of view,
the setting that is simulated by the Fishing Conflict Game, can be
characterized by: (a) multiple agents; (b) a dynamically changing
environment; (c) inter-individual conflicts: The participants compete
over the resource; (d) intra-individual conflicts: Given a particular
state of knowledge and motives, cognitive-motivational conflicts are
likely to result (e.g., maximizing one's gain or
protecting the resource?); (e) incomplete knowledge about
the robustness of the resource; (f) incomplete knowledge about the
behavior of the other participants; and (g) knowledge acquisition
through participation in the game.
The empirical results point to the importance of several factors that
influence behavior in a resource dilemma situation (e.g., Dawes, 1980;
Van Lange et al., 1992). (a) A ubiquitous finding is that behavior is
tied to the individual motives of the participants. (b) Domain
knowledge and experience with the problem are generally considered
important determinants of the behavior. Domain knowledge in the
present case means ecological knowledge. (c) Findings concerning
interpersonal trust highlight the role of social knowledge in forming
an estimate of the other participants' intentions and predicting their
future behavior. At our laboratory, a series of five experiments was
conducted with the Fishing Conflict Game (Spada et al., 1987). The
findings can be summarized as follows: Participants of groups with
high individual gains can be characterized by less destructive
motives. They show a better ecological and social knowledge. A
tit-for-tat strategy (i.e., repay overharvesting by overharvesting)
of an instructed participant is misunderstood as unpredictable and
exploiting. On the other hand, a confederate of the experimenter
using a resource adapted equal share strategy brings about positive
effects: cooperation among co-players and a stable resource at the
state of a maximum sustainable yield.
KIS
(Ernst & Spada, 1993; Ernst, 1994) models the
behavior of an agent participating in the Fishing Conflict Game. In
the model, we spell out what types of knowledge and which
motives come into play in determining people's behavior in this
particular multi-agent situation, in which an agent has to balance
multiple ecological, economic, and social short-term and long-term
goals. Furthermore, following the desiderata outlined
above, we have to be specific about how people understand, anticipate
and adapt to the behavior of others in these situations.
In the KIS model, motives define a set of individual preferences,
that are relatively constant over time. This interpretation of
motives is in line with the conceptualization of
social values orientation (cf. Liebrand, 1984), according to
which subjects' preferred social orientations remain invariant over
time.
In the KIS model, three motives are postulated that correspond with
psychological findings (e.g., Van Lange, et al., 1992), economic
theories (e.g., Etzioni, 1988), and philosophical considerations
(e.g., Danielson, 1992): (a) greed, a utilitaristic 'get as much
as possible', (b) resource orientation, the interest or moral
obligation to stabilize the resource at the state of maximum
sustainable yield, and (c) social fairness, which aims to
minimize the differences between the own gain and those of the other
players. In the KIS model, the strengths of the motives--graded as
strong, moderate, and weak--make up the so called motive
structure of a simulated player; different motive structures lead
to different types of players. As three motives are postulated, KIS
allows to construct six different patterns of motive structures. The
motive structure influences the decisions of a player and determines
the evaluation of a situation as being more or less desirable.
In a commons dilemma three sources of knowledge play a major role:
ecological knowledge, social knowledge, and practical knowledge (Spada
et al., 1987). Ecological knowledge refers to detecting the
regularities about the resource propagation. It improves with
experience and is implemented by a simple learning mechanism. Social
knowledge permits the ascriptions of others' intentions, motives, and trustworthiness and allows to predict their future
actions. Practical knowledge (Ohlsson, 1995) is represented by action
schemas and a decision procedure.
An action schema in the KIS model reflects the knowledge of how to
generate behavior necessary to achieve a short-term goal.
Furthermore, an action schema in the KIS model represents the
experiences about its past and/or hypothesized future success with
regard to the different motives in the form of motive-specific schema
strength parameters. Thus, an action schema comprises a (short-term)
goal, a method how to achieve that goal, and an evaluation how this
short-term goal might satisfy the different motives (see
Table 1).
Table 1: Action schemas within the KIS model.
Note. 
and 
are greater 1. Schema strength parameters are
motive-specific and change through learning; higher values mean the
action schema is evaluated as more useful to the motive. '*'
indicates that
this value has to be estimated by the model based on
its social and ecological knowledge. All methods are state
dependent and are specified for a game with three players.
Four action schemas are implemented in the KIS model: (a) The
equal share action schema yields a catch quota which is
as close as possible to the predicted quotas of the other
players. Predicting others' catch quotas is accomplished by bringing
social knowledge to bear. (b) Similarly, the relative gain
maximization action schema is social in nature and uses predictions
of the others' quotas as well, but the result will be a quota
exceeding the others' quotas. (c) Integrating both fairness and
ecological concern, the resource adapted equal share action
schema uses the equality principle, but takes into account the
(estimated) optimal resource propagation at the same time. (d) In
contrast, the ecological-social overharvesting action schema
generates a catch quota markedly above the latter one (see again
Table 1).
Based on a player's knowledge about the current state of the resource
and/or the catch quotas to be expected by the other players, the
selected and instantiated action schema specifies the player's catch
quota. Thus, there are three steps taken for goal-directed action:
(a) Selecting a schema (build a short-term goal), (b) instantiating it
(adapt to the situation), and (c) executing its procedural
part (act). The selection procedure is described next.
A decision in the KIS model consists of selecting the (subjectively)
best action schema in a given situation. The decision procedure
integrates the action schema parameters and the motive structure into
one single value by simply summing up the products of the motive-specific
action schema parameters multiplied by the strength of the
corresponding motives. For instance, a KIS agent with high resource
and moderate fairness orientation using the schema strength parameters
presented in Table 1 would decide for the
resource adapted equal share action schema.
Implementing the two social action schemas (equal share, relative
gain) and predicting future developments of the resource requires
anticipating the actions of the fellow players. Prediction implies
understanding. But for an observer, the overt behavior in the KIS
framework is ambiguous, because it is determined by an action schema
and by ecological and/or social knowledge (see method part of
Table 1). Understanding an actor thus essentially
consists of disambiguating possible explanations for the observed
actions. Here, social knowledge comes into play. Social knowledge
allows an observer to attribute short-term goals and motives to other
people's actions.
According to the predominating view in social
psychology, an observer acts in this situation like a naive scientist
and applies a primitive theory of mind to understand other
people's actions. This approach, often called 'theory-theory', is at
odds with 'simulation theory' which states that we understand others'
actions in the absence of any theory of mind, by using the resource of
our own minds to simulate the beliefs and intentions of others
(Goldman, 1993).
Recently, Barnes and Thagard (in press) have argued that these
alternatives are not mutually exclusive. In presenting a computational
account for empathy they concluded that empathy always involves
simulation, but may simultaneously include theory application. They
present a computational model in which the observer tries to construe
correspondences between prior own experiences and the action of the
other person by analogical mapping. Close
analogies appear to involve little or no theoretical work.
Long-distance analogies inevitably require the application of a
theory to compensate for disparate situations and goals. A similar
view is taken in the KIS model.
Because in the Fishing Conflict Game all players are in an identical
ecological situation, the KIS model interprets the catch quota of
another player by applying all its own action schemas, which get
instantiated by the model's own ecological knowledge. The model then
compares the outputs with the observed catch quota. The action schema
whose result comes closest to the observed behavior is attributed as
being the short-term goal of that person. Note that as
a consequence of this process, poor ecological knowledge will lead to
substantial mis-interpretation of the observed behavior.
In a similar vein, the guiding motive of another player is determined.
The situation and the already ascribed action schema are fed into the
model and this action schema then is evaluated for each motive. Again,
this simulation is done using the own subjective knowledge (in this
case, the own action schema strength parameters). The motive that best
explains the decision to use this particular action schema is considered
the momentarily guiding motive of the fellow player.
Since motives are assumed to be relatively constant over time, the
overall motive ascription is conservative. Applying its naive theory
about motives and short-term goals, the model ascribes that motive as
stable and trait-like for a fellow player which it had determined most
often as his or her momentarily guiding motive during the whole game.
The KIS model behaves adaptively through instantiating action schemas
to each given situation. In addition, knowledge acquisition
mechanisms are provided. In KIS, all learning of practical knowledge
results in decreasing or increasing schema strength parameters.
This mechanism is based on the evaluation of the consequences of one's
own actions. Learning by doing is considered to occur with every
action taken. It only affects the dominant action schema, i.e., the
one that was chosen for action.
Learning by mental simulation is triggered by impasses
(Rosenbloom, Laird, Newell, & McCarl, 1991; VanLehn, 1988). In KIS,
an impasse results from the detection of an inconsistency between
the expected and the observed behavior of a co-player, or when a tie
between the evaluation of action schemas exists. The outcomes of
mental evaluations of possible future states are then integrated into the
strength parameters of the action schemas, thus possibly leading to
new rankings. Possible future states are simulated using own
ecological knowledge and by anticipating actions of the co-players.
Mental simulation is restricted to a look-ahead depth of two game rounds.
Figure 1 shows as an illustrative example the
development of the resource in a Fishing Conflict Game with three
artificial players. A resource oriented artificial player with very
poor ecological knowledge interacts with a greedy and a fairness
oriented artificial player with good ecological knowledge. While the
greedy player decides on high catch quotas for several rounds--based
on the action schemas relative gain and overharvesting--and then
reduces its catch quotas drastically due to a negative evaluation of
these action schemas in the light of a rapidly diminishing resource,
the resource oriented player takes too high harvests according to its
poor ecological knowledge. With the negative experience of the third
round, this player decides to select the equal share action schema.
Thus, the poor ecological knowledge is overridden by a
socially oriented action schema, namely the equal share schema.
This leads to a recovery of the simulated resource.
Figure 1: Development of the resource (a;left) and individual catch quotas
of three artificial players (b;right). MSY denotes the range of maximum
sustainable yield. A resource oriented player with poor ecological
knowledge ( 
) acts in a greedy social
environment ( 
).
The aim of our study imposes three empirical demands on the model: (a)
The different artificial social environments defined by KIS models
should be useful as experimental settings. One point is that
artificial agents should be believable. (b) The effects of different
social environments should be in line with previous findings in this
domain. (c) Human and artificial subjects confronted with the same
social environments should behave in the same way.
Two experimental environments were built differing only in the motive
structure of one of their two artificial players. In the
unselfish environment, this player had a dominant
resource orientation accompanied by a moderate fairness orientation.
In the greedy environment, this player had a high greed
orientation and a moderate resource orientation.
The other artificial player had an identical motive structure in both
environments being highly fair and moderately greedy.
Both artificial players had good ecological knowledge.
The third player was either the human or a matched artificial subject,
i.e., a modeled KIS counterpart. For each subject, we assessed in
each round the estimation of the optimum total catch quota, the actual
catch quota, and the ascription of short-term goals of the co-players.
Forty-two students took part in the experiments in single sessions,
distributed equally between both experimental environments.
Participants were told that their co-players were seated in adjacent
rooms with networked computers.
Did human subjects consider the artificial environments realistic? In
a short, open debriefing session after the experiment, no subject
conjectured that the co-players were artificial.
Moreover, in a questionnaire following the game the frequency of
correct estimations of the co-players' dominant motives was quite high
with 67% and 69% correct estimations for the resource oriented and
gain oriented player, respectively. Thus, the model succeeded in a
kind of Social Turing Test as proposed by Carley and Newell
(1994).
In line with previous studies (Spada et al., 1987; Van Lange et
al., 1992) we hypothesized that the greedy environment would lead
subjects to harvest too much in reaction to the catches of the greedy
artificial player. As a consequence, bad overall ecological
performance should result. The unselfish environment in turn was
expected to lead to a good resource management, i.e., ecologically
adapted catches and a good overall yield (joint gain) from the
resource.
(a)
(b)
Figure 2: Catch quotas (a) and tons of fish
(b) in both environments. Rounds are pooled into groups of
three starting with the third round. Data represent averaged
values (N=42).
According to predictions and reflecting the very dilemma structure of
the game, the joint gain of all three players was higher in the
unselfish environment (M = 432.0) than in the greedy
environment (M = 302.9),t(40) = 8.38,p <
.0001. For further statistical analysis, catch quotas and the amount
of fish in the resource were pooled into groups of three starting with
the third round. Using these pooled variables, ANOVAs containing the
between-subject variable of social environment (unselfish or greedy)
and the within-subject variable of rounds (3-5, 6-8, 9-11, or 12-14)
were computed. For both variables (catch quotas and amount of fish in
resource), the ANOVAs revealed significant (all ps < .05) main
effects for environment (Fs(1,40) = 4.38 and 70.43) and for
rounds (Fs(1,120) = 3.06 and 11.18), and a significant interaction
between environment and rounds (Fs(1,120) = 11.23 and 13.31).
Figure 2 shows the respective means.
Whereas in the unselfish environment the resource remained stable
during the first phase and recovered rapidly in the second phase, the
resource was twice exploited nearly to extinction in the greedy
environment (Figure 2b). As predicted, subjects in the
unselfish environment behaved sensitively to the size of the resource
throughout the entire game--taking more when the fish population was
high and vice versa--and exhibited positive transfer of their
acquired ecological knowledge from the first phase to the second
phase. Subjects in the greedy environment showed a notable deviation
from this principle. At the beginning of the second phase, which
started for both environments with a new fish population of 70 tons,
they exhibited, on the average, catch quotas far above the
corresponding quotas in the unselfish environment. But beside this
exception and contrary to predictions, subjects in the greedy
environment showed relatively low catch quotas, when the resource was
on the way to extinction. Because subjects should have acquired
almost identical ecological knowledge in the fist phase of the game,
we interpret the high quotas in the greedy environment at the
beginning of the second phase as retaliation in conjunction with an
attempt to compensate a meagre yield during the first phase in which
the greed oriented simulated player had overexploited the resource
dramatically.
On the basis of their behavior in the second round of the game, we
categorized our subjects as more resource, fairness, or greed
oriented. Then we built the motive structure of the KIS model
according to our diagnosis to match each subject by a twin.
Similarly, we adjusted the ecological knowledge of the artificial
subjects. The aim of this design was to compare the behavior of human
and artificial subjects from Round 3 onward.
On an aggregated level, we replicated the effects concerning the fish
population for both environments (Figure 3b). Although
the environments (i.e., the experimental manipulations) themselves
contribute to this finding, it is nevertheless important for
an unbiased comparison of catch quotas.
(a)
(b)
Figure 3: Subjects (solid lines) and their artificial counterparts
(dashed lines): Catch quotas (a) and tons of fish (b) in
both environments. Rounds are pooled into groups of three starting
with the third round. Data represent averaged values (N=42).
For the mean catch quotas of the artificial subjects, we obtained a
good fit for the unselfish environment. In the greedy environment,
the artificial subjects showed much higher catch quotas in the first
phase compared to our subjects. In fact this is in line with earlier
empirical findings (Spada et al., 1987). But we also got different
trends in both phases (Figure 3a). Contrary to human
subjects in the greedy environment, the artificial subjects took in
reaction to the overharvesting of the other players even more although
the fish population was decreasing. On an individual level, we could
model about one third of our human subjects exactly, mainly due to
good fits between pairs within the unselfish environment. For the
greedy environment, this analysis revealed major disparities. Most
notably, the modeled subjects harvested more than human subjects in the
first phase and consequently did not show compensation or retaliation
behavior at the beginning of the second phase.
These mismatches suggest an important shortcoming of the KIS model.
KIS assumes the existence of stable motives that guide behavior. In
the Fishing Conflict Game, motives themselves can be affected by
situational factors (e.g., when the state of the resource diminishes
dramatically) and by emotional reactions (retaliation).
A major contribution of this research consists in using a process
model for defining plausible, realistic, and believable agents in
artificial social environments. The validity of the model was tested
in a study that was designed by means of these environments. Our
prior finding, that a resource adapted equal share strategy has
positive influence on the co-players by producing predictable and
ecologically sensible behavior, was confirmed and could be explained
by the model. In a greedy social environment, the model did not
account appropriately for the influence of situational and emotional
aspects. But despite this shortcoming, the model allows to create
reactive and sufficiently realistic social learning environments based
on a broad range of different artificial agents. An advantage of such
a procedure is that one can define standardized sequences of learning
opportunities, for which the effects can be deduced on the basis of
the model.
This research was supported by grant no. Sp 251/5-x from the German
National Research Foundation (DFG) to the second author. We would like
to thank Michael Scheuermann and Hansjoerg Neth for assistance, and
Volker Franz, Frank Ritter, and Paul Thagard for comments on an
earlier draft.
- Axelrod, R. (1984). The evolution of cooperation. New
York: Basic Books.
- Barnes, A., & Thagard, P. (in press). Empathy and analogy,
Dialogue: Canadian Philosophical Review.
- Carley, K. M. & Newell, A. (1994). The nature of the social
agent. Journal of Mathematical Sociology, 19, 221-262.
- Danielson, P. (1992). Artificial morality. London:
Routledge.
- Dawes, R. M. (1980). Social dilemmas. Annual Review of
Psychology, 31, 169-193.
- Ernst, A. M. (1994). Soziales Wissen als Grundlage des Handelns in
Konfliktsituationen (Social knowledge as a basis for acting in
conflict situations). Frankfurt/M.: Peter Lang.
- Ernst, A. M., & Spada, H. (1993). Modeling agents in a resource
dilemma: A computerized social learning environment. In D. Towne, T.
de Jong, & H. Spada (Eds.), Simulation-Based Experiential
Learning (pp. 105-120). Berlin: Springer.
- Etzioni, A. (1988). The moral dimension: Toward a new
economy. New York: The free Press.
- Glance, N. S. & Huberman, B. A. (1994). Social dilemmas and
fluid organizations. In K. M. Carley & M. J. Prietula (Eds.),
Computational organization theory (pp. 217-239). Hillsdale, NJ: Lawrence Erlbaum Associates.
- Goldman, A. (1993). Philosophical applications of
cognitive science. Boulder: Westview Press.
- Hardin, G. R. (1968). The tragedy of the commons. Science,
162, 1243-1248.
- Liebrand, W. B. G. (1984). The effect of social motives,
communication and group size on behavior in an N-person multi-stage
mixed motive game. European Journal of Social Psychology,
14, 239-264.
- Macy, M. (1995). PAVLOV and the evolution of cooperation: An
experimental test. Social Psychology Quarterly, 58, 74-87.
- Messick, D. M., & Liebrand, W. B. G. (1995). Individual
heuristics and the dynamics of cooperation in large
groups. Psychological Review, 102, 131-145.
- Ohlsson, S. (1996). Learning from performance errors.
Psychological Review, 103, 241-262.
- Rosenbloom, P. S., Laird, J. E., Newell, A., & McCarl, R.
(1991). A preliminary analysis of the Soar architecture as a basis
for general intelligence. Artificial Intelligence, 47,
289-325.
- Spada, H., Opwis, K., Donnen, J., Schwiersch, M., & Ernst,
A. M.(1987). Ecological knowledge: Acquisition and use in problem
solving and decision making. International Journal of
Educational Research, 11, 665-685.
- Thagard, P. (1992). Adversarial problem solving: Modeling an
opponent using explanatory coherence. Cognitive Science, 16,
123-149.
- Van Lange, P. A. M., Liebrand, W. B. G., Messick, D. M., &
Wilke, H. A. M. (1992). Social dilemmas: The state of the art:
Introduction and literature review. In W. B. G. Liebrand, D. M.
Messick, & H. A. M. Wilke (Eds.), Social dilemmas. Theoretical
issues and research findings. (pp. 3-28) Oxford: Pergamon
Press.
- VanLehn, K. (1988). Toward a theory of impasse-driven learning.
In H. Mandl, & A. Lesgold (Eds.), Learning issues for
intelligent tutoring systems. (pp.19-41). New York: Springer.
- KIS
- KIS was implemented using KEE and Common LISP
on a UNIX workstation.
Back to text
A Cognitive Model of Agents in a Commons Dilemma
Proceedings
of the Nineteenth Annual Conference of the Cognitive Science
Society (1997). pp. 560-565.
Mahwah, NJ: Lawrence Erlbaum Associates.
Click here to get a postscript-version (142kb) of the paper.
Back to table of contents
This document was generated using the LaTeX2HTML translator Version 96.1 (Feb 5, 1996) Copyright © 1993, 1994, 1995, 1996, Nikos Drakos, Computer Based Learning Unit, University of Leeds.
Josef Nerb
Tue Apr 22 17:11:54 MET DST 1997
