Modelling of Self Control through Precommitment Behaviour in a Functionally
Decomposed Connectionist Architecture
CASE FOR SUPPORT, PART 2: Description of Proposed Research and its Context
The aim of the proposed research is to investigate how evolution has resulted in self control such that people must use
precommitment behaviour to control their future actions. Why is it that some people have organised, successful lives
while others ruin their lives, or in the extreme, kill themselves with overeating, lack of exercise, smoking, or drug abuse,
which are generally deemed to be irrational behaviour? The proposed research aims to investigate these questions by
gaining a greater understanding of self control through simulation of the brain as a functionally decomposed parallel
system. Unlike a serial system, a functionally decomposed system can have internal conflicts, as different processes
exhibit different behaviours. Self control through precommitment is an example of this internal conflict. In reality, self
control manifests itself by how people limit themselves by choosing a later reward over an immediate reward.
Precommitment is a mechanism for doing this, by making a choice now that will make it impossible to change our minds
later or, if we do, the change is costly. Although there has been much research in psychology and economics in the area
of self control, to the best of our knowledge this is the first time that it is being proposed to simulate self control through
precommitment behaviour in a computational model that is both biologically and psychologically relevant, in a
competitive interaction. The research will aim to answer the questions: is this behaviour learned as part of socialisation
and survival, or are we born with this capacity? This research will facilitate the advancement of academic knowledge
from cognitive and computational neuroscience as well as related areas, most notably psychology and economics. It will
therefore contribute to bridging the gap between the modelling community and the experimentalists.
1. Background
The project will test the theories proposed in psychology and economics on self control behaviour. Self control can be
broadly defined as is choosing a large delayed reward over a small immediate reward (Rachlin, 1995). Figure 1a
illustrates a choice between a smaller-sooner (SS) reward, available at time t2 and a larger-later (LL) reward, available at
t3. The thin lines subtended from points SS and LL are temporal discount functions indicating the effect of increasing the
delay of the reward or delay of gratification (Mischel et al. 1989). The gradient of the lines represents the fact that the
closer you get to a reward the faster its current value increases (Rachlin, personal communication, 2003). The crossing of
the discount functions indicates a reversal of preferences as in hyperbolic discounting (Green et al., 1994). At time t 1 for
instance, the value of the larger-later (LL) reward exceeds the smaller-sooner reward (SS). However at t2 when SS
would be immediately available, the value of SS exceeds LL value.
LL
Reward
SS
Choice X
value (v)
LL
SS Reward
Choice Y
LL
t1
t2
t3
t1
time (t) 
t2
t3
time (t) 
Figure 1a. Illustration of choice of larger-later over
smaller-sooner reward (adapted from Rachlin, 1995)
Figure 1b. Illustration of Precommitment
behaviour (adapted from Rachlin, 1995)
To give an example where self control behaviour is exercised, with reference to Figure 1a let LL represent obtaining
good grades and SS going to the pub. Let t 1 indicate the start of an academic year. At this time for most students the
value of getting good grades exceeds that of going to the pub. When invited to the pub at t 2 however, the value of SS is
higher than their long term goal of getting good grades (LL). If the student exercises self control s/he will choose study
(LL) over the pub (SS). Research to date suggests that we recognise that we have self control problems and try to solve
them by precommitment behaviour (Rachlin, 2000; Ariely, 2002). Precommitment is defined as making a choice with the
specific aim of denying oneself future choices (Rachlin, 1995). Examples are:
 putting an alarm clock away from your bed, to force you to get up to turn it off.
 saving part of your monthly pay cheque into an investment fund, to prevent you from spending it.
 keeping chocolate biscuits out of your house to prevent late-night binges.
 disconnecting the Internet connection to your computer when you have a deadline.
1
Precommitment is illustrated in Figure 1b, which is based on experiments by Rachlin and Green (1972) on commitment
behaviour by pigeons. At Choice X at time t2 choosing between a small-immediate reward, SS, and a large-delayed
reward, LL, the preference was for the smaller-immediate reward SS. However, at a prior point in time t 1, shown as
Choice Y, an alternative was preferred (the lower arm) that restricted choice to LL only at t 2, i.e. at t1 they committed to
LL. Precommitment behaviour can be viewed as an example of the internal conflicts that arise in our brain (Nesse,
2001). This could be representative of the competition between the "higher" centre of the brain (i.e. rational thought) and
the low-level centre (i.e. instinctive behaviour) (Jacobs, 1999). If we were truly rational our preferences would not
change over time: if it is in your interest to get up when the alarm clock goes off, you should not want to go back to sleep
when it wakes you (Samuelson and Swinkels, 2002). Is precommitment behaviour learned as part of socialisation or are
we born with this capacity? Why do people engage in precommitment behaviour, when there is always a cost? For
example alcoholics exercising precommitment by taking the drug antabuse, which causes severe pain after drinking. This
research aims to provide explanations for both how and why this apparently inconsistent behaviour occurs.
2. Programme and Methodology
2.1 Overall Aims and Objectives
Self control behaviour is a universal trait, which touches our lives at all levels. Although there has been much research in
psychology and some biological evidence on the neuroanatomy of self control (Matsui et al., 2002), to the best of our
knowledge there is no research to date that integrates these findings into an analysis of how the apparently inconsistent
behaviour of self control through precommitment came about. This project is the first attempt to simulate self control
through precommitment behaviour in a biologically and psychologically relevant computational model with learning.
The overall aim of the proposed research is to find an explanation for self control through precommitment. We will do
this by simulating this behaviour in a computational model of the neural cognitive system. The model must encompass a
cognitive architecture that provides a general explanation of self control. It also needs to cater for anomalies. For
example, it needs to deal with discounting, that is, the reduction in value of a reward due to delay; and reversal of
preferences as shown in Figure 1a. The model should incorporate a continuous learning process by interacting with the
environment through its sensors, i.e. it will have minimal pre-programmed knowledge. It will also be an evolutionary
system. The aim is to develop a model that is simple, but also sufficiently detailed to capture essential characteristics of
the real neural system, so as to be useful for understanding how the human brain functions. The resulting model will be
subject to simulation of evolution with learning in order to find an explanation for self control through precommitment.
Possible explanations that will be explored, which are neither mutually exclusive nor exhaustive, are:
1. It results from one of the following: (a) an animal evolving optimal low-level, (i.e. instinctive) behaviours in
response to certain cues in an ancestral environment; (b) the animal being moved to a novel environment where
these low-level behaviours are inappropriate to its higher goals; (c) the animal learning cognitively in the
"higher" part of the brain that the low-level behaviours are inappropriate; (d) the animal trying to devise a way
in the higher centre of the brain to bypass the low-level behaviours (Sozou, 2003). As such, standard
psychological self-control problems can be understood as part of a spectrum of phenomena involving
overcoming behaviours which cognition can directly control only partially or not at all. Nesse (2001) alluded to
this behaviour, in that the frontal lobes, the most recent part of the human brain in evolution terms, are essential
to the inhibition of short term goals in order to fulfil long term objectives.
2. It results from a best evolutionary compromise to environmental complexity and variability. It is not feasible for
evolution to program the brain with a direct hard-wired response to every situation it could meet. Instead, there
is goal-directed behaviour and a capacity for learning. However, these goals cannot perfectly correspond to
fitness. Hence, natural selection has allowed low-level behaviours to effectively take control when cues are
strong enough to reliably reflect fitness consequences. This gives rise to the multiple personalities theory; for
example, the person that wakes up in the morning is different from the person who went to sleep the previous
night. This theory was suggested by Trivers (2000) in the evolution of self-deception.
3. It is a side effect of the necessarily functionally decomposed structure of the nervous system and brain (Jacobs,
1999). Natural selection has not removed this apparent inconsistent behaviour because the imposed cost is low
or the constraint to be overcome is too small.
4. It is a direct consequence of a conflict of interest between altruistic behaviour associated with the higher centre
of the brain and behaviour such as emotion and aggression associated with the limbic system (suggested by
Trivers (2000) to be a genetic conflict).
5. It is a side effect of the evolution of commitment mechanisms for game-theoretical situations where
commitment is useful, e.g. anger and self deception (Nesse, 2001).
The individual measurable objectives of the proposed research are as follows:
1. To bring together research in psychology, neuroscience, economics and computer science, and to validate this
research by implementation of a neural network system that is both biologically and psychologically relevant.
2. To investigate functional decomposition in the brain with particular reference to internal conflict related to self
control through precommitment behaviour, with the aim of building a neural model of these decomposed
functions. The model will provide a cognitive architecture of the human brain with learning. It should not
exclude the other scales of brain modelling, i.e. networks of neurons and single neurons. The model should be
able to provide a general explanation of self control behaviour through precommitment, but must also deal with
anomalies.
2
3.
To test the explanations 1 to 5 above by simulation of a functional decomposed neural system undergoing
evolutionary adaptation with robust learning. In order to do this, the conditions and parameter values required
for each theory will be established empirically.
2.2 Detailed Methodology
A brief description of the overall methodology to be used
The research will explore the theory that the human brain is a modular system, i.e. it is composed of different modules
each specialising in different tasks with each module skilled in all areas but dominant in one (Jacobs, 1999). A
computational model of the neural cognitive system of self control through precommitment behaviour will be developed
inspired by the structure and function of the human brain. From the viewpoint of modern cognitive neuroscience, self
control as an internal process can be represented as in Figure 2 (Rachlin, 2000).
Environment
Agent
1
Higher Brain
(cognition)
State
Arrow 1 Information
Arrow 2 Behaviour
Arrow 3 Stimulus
2
Action
Lower Brain
(motivation)
3
Stimulus
Figure 2. Self control as an internal process within a Reinforcement Learning (RL) framework
(based upon Rachlin, 2000)
Arrow 1 represents of information coming into the cognitive system located in the higher centre of the brain, which
represents the frontal lobes associated with rational behaviour such as planning and control. This information combines
with memory located elsewhere in the brain (possibly the hippocampus) to form ideas about the world. These ideas
combine with messages coming from the lower brain, representing the limbic system which is associated with emotion
and action selection (Trivers, 2000; O’Reilly and Munakata, 2000; Rachlin, 2000). This forms purpose, which travels
back down countermanding or augmenting stimuli entering the lower brain (arrow 3). This finally results in behaviour
(arrow 2). In this simple model, the modules can be represented as a network architecture of two interacting networks of
neurons, to reflect the higher and lower centres of the brain. This follows on from the ideas proposed in the Section 1
above that precommitment behaviour is an example of some internal conflict between the lower and the higher centres of
the brain. Learning from interaction with our environment is a fundamental idea underlying most theories of learning and
for this reason a neural network with reinforcement learning is to be used. Different forms of weight update rules will be
introduced for each module, with various number of neurons with the aim of determining what dependencies there are
and what internal conflicts exist in a functionally decomposed neural system. The resulting model will go some way in
investigating the possible explanations as listed in Section 2.1, for this apparently irrational behaviour, i.e. if individuals
were fully rational, precommitment would be unnecessary as any later temptation that would jeopardise their true
preference would be rejected (Nesse, 2001). In order to simplify the model there may be a requirement to differentiate
between personal precommitment and precommitment as viewed within game theory (Nesse, 2001), i.e. in games one
can safely make the assumption that a player is playing to win, as opposed to personal precommitment where one has
moral and other dilemmas. The scope of the project will focus on the latter initially, with a view to using this as a basis
for future modelling of personal precommitment.
The resulting Artificial Neural System will be verified by examining how people play games, in particular games which
have a real world application. For example, pollution can be regarded as a game of Prisoner’s Dilemma (Hamburger,
1979); war can be regarded as another dilemma game called Chicken (Binmore, 1992), where two players compete for a
piece of territory. If one chickens out he looses, if both chicken out the situation remains the same and if neither chickens
out the consequences are unpleasant for both players. The effect of reward and punishment on the two neural networks’
behaviour will be observed in various game-theoretical situations. The neural system will play against other neural
systems and also against human players. The results of these tournaments will be compared to theoretically optimal
strategies. A strategy or policy is defined as the decision making function which specifies what action to take in any
situation. In psychology this would be a set of stimuli-responses. An optimal strategy is defined in terms of Nash
Equilibrium where each player’s strategy choice is a best reply to the strategy choice of other players (Rubenstein, 1982).
3
In the final simulation of this functionally decomposed neural network system which is able to undergo evolutionary
adaptation with learning, the parameters defining the network will be subject to simulated genetic evolution using
genetic algorithms (Holland, 1992). The resulting networks will be subject to learning thereby exploring evolution with
robust learning within a modular neural system.
Methodology for neural modelling of a computational functionally decomposed cognitive system
The model must explain in computational terms how the brain generates the apparent inconsistent behaviour of self
control through precommitment based on the known neurophysiology of the brain. It will do this by building a network
architecture of two networks exhibiting different behaviours to represent the higher versus lower cognitive functions as
described in Figure 2. Current research indicates that the higher cognitive functions are not based on the action of
individual neurons in a limited area but are based on the outcome of integrated action of the brain as a whole (O’Reilly
and Munakata, 2000). For this reason, a holistic approach to the brain as a functionally decomposed system will be
adopted. Whilst the model will be a cognitive architecture of higher cognitive functions, it should not exclude research
into brain modelling at the neuron level and the morphology of individual neurons or networks of neurons responsible
for specific functions (Fodor, 1983; Jacobs, 1999). The model will explore the development of the brain as a functionally
decomposed system with a bootstrapping strategy, i.e. building blocks are constructed from simpler tasks first as a
framework for more difficult tasks later (Cohen et al., 2002). Particular attention will be given to neural competition
between modules, without excluding inhibitory competition at the neuron level (Jacobs, 1999). The model will also take
into consideration the complexity of the environment as well as behaviour. The variables that define the network will be
parameterised to enable control of the model. At a minimum these will include the form of learning, the learning rate and
the number of neurons in the each module. We will validate the model by comparing with the behaviour of more detailed
models (Cohen et al., 1990; Taylor, 2002).
Methodology for building the neural model and testing with game-theoretical scenarios
A feasibility study has already been conducted where we constructed a computer simulation of a neural network playing a game
with real world consequences using natural forms of learning, e.g. Reinforcement Learning (Banfield and Christodoulou, 2003;
2004). The Reinforcement Learning framework as described by Barto and Sutton (2002) was implemented successfully in
games where the players' payoffs are neither totally positively correlated nor totally negatively correlated (general-sum games).
The results showed that Reinforcement Learning worked successfully when the network competed against an artificial opponent
whose responses were generated randomly with uniform probability (Banfield and Christodoulou, 2003) and when two networks
learned simultaneously in a shared environment (Banfield and Christodoulou, 2004). The current study will build on the results
of Banfield and Christodoulou (2003; 2004) to implement two artificial neural networks representing the higher and lower
centres of the brain, which will compete against each other in general-sum games that model a real-world situation. In the first set
of experiments a simple general-sum game will be used. Rubinstein’s Bargaining game (1982) is such a game and therefore is
appropriate. Rubinstein’s Bargaining game involves two players and a resource, or pot, such as money. The two players seek to
agree how to divide the pot. The pot decreases at each turn of the game, by a fixed amount, hence it pays both players to reach an
agreement sooner rather than later. At the beginning of a turn, one player makes an offer to the other player, which is the fraction
of the pot that player is willing to give to the other player. The other player can either accept or reject. If he rejects the offer, he
then makes a counter-offer and the game continues for another turn. The game terminates when either nothing remains in the pot
or one of the players accepts an offer. Each player seeks to gain as much of the pot as he can. The concept of time is only
relevant for the duration of one game, which is found to last at most five turns (Banfield and Christodoulou, 2003). Thus there is
no need to retain details of previous games and no need to employ temporal neural networks such as Time Delay or Recurrent
neural networks.
The Neural Network will be implemented as a simple feed forward multi-layer perceptron network using a variety of
weight update rules. To verify the results of the Rubinstein’s Bargaining Game we will test if a similar neural
architecture would give equivalent results with a different game. For this purpose, it is intended to repeat the
experiments with a simulation of the Prisoner’s Dilemma game. Research on self control suggests that there is a
relationship between cooperation and self control (Brown and Rachlin, 1999). Human cooperation has been modelled as
a game of Prisoner’s Dilemma (Axelrod and Hamilton, 1981) and therefore this game is appropriate for this stage of the
research. Brown and Rachlin (1999) played a variation of the Prisoner’s Dilemma game whereby choosing a higher
current reward conflicted with behaviour that maximised the overall reward. The results of our simulation will be
compared with the empirical results of Brown and Rachlin (1999), which showed a close analogy between self control
and social cooperation. We will implement precommitment by adding or changing a strategy to include some signalling
information, which will tell the network when precommitment is activated. In this first case we still have one network
against an artificial opponent whose responses are generated randomly with predefined strategies, which are the optimal
strategies for that game. For the second stage of the verification, we will assume that the network behaves as expected
and then we will play the network against “itself”, i.e. another network with same strategies or different strategies
reflecting the internal conflict of different parts of the brain wanting different behaviour, i.e. higher versus lower parts of
the brain (refer to Figure 2.)
Methodology for evolution of the neural cognitive system
The final step is to create alternative strategies through evolutionary methods with a view to enhancing the network and
explore the possible explanations given in Section 2.1. We will consider two approaches being considered for the
building of the evolutionary system. The first is to introduce evolutionary ideas in a game theory context such as in
Axelrod’s evolution of co-operation (Axelrod and Hamilton, 1981), where different strategies were represented as a
4
string of chromosomes for the Iterated Prisoner Dilemma game. These were played against the same opponent for 200
games. Each strategy was ranked by the total payoff accumulated (not by the number of opponents defeated). The most
successful strategies were subject to genetic evolution using genetic algorithm (GA) techniques (Holland, 1992). The
aim was to see which strategy emerged as the Evolutionary Stable Strategy (ESS). This approach will be extended to
include with the string of chromosomes, the network topology and learning parameters, as described in the research by
Bullinaria (2004). The second approach will use cascade-correlation neural networks in models of development as
described by Shultz and Mareschal (1996). Shultz and Mareschal (1996) start with a simple network containing no
hidden units, and then grow the network to some acceptable level using the cascade-correlation algorithm of Fahlman
and Lebiere (1990).
The simulation of evolution on the two Artificial Neural Network (ANN) model from our research will be tested using
games. The simulation will focus on the functional decomposition of the brain (see Section 2.1) and attempt to explain
such behaviour as a by-product of some internal mechanism. The simulation will examine the role of reinforcement and
reinforcement history to explain variances in an individual’s behaviour. Finally, the simulation will explore the theory
that such behaviour may be adaptive.
In summary, the model will go some way to answering the question whether self control behaviour through
precommitment is (i) a biological by-product, (ii) an internal conflict or (iii) an adaptation to enhance the survival of the
species. The behaviours, which evolve, will be evaluated in different environments, and in different game theoretical
situations. The extent to which:

inconsistent behaviours occur when the neural network is tested in conditions similar to those in which it
evolved will go some way to supporting the theory that self control through precommitment is a biological byproduct;

inconsistent behaviours occur when the neural network is tested in conditions different to those in which it
evolved will support the theory that self control through precommitment is a manifestation of an internal
conflict (higher v. lower brain centres);

commitment behaviour evolves when the network is subjected to evolutionary change in game-theoretical
interaction will support or dismiss that self control through precommitment is adaptive.
2.3 Programme of Work and Management
Based on the Overall Aims and Objectives described in Section 2.1, a Diagrammatic Work Plan of the research
indicating the length and interactions of all tasks is attached. The task list is as follows:
a. Identification of possible techniques (ANN with RL, GAs) and testing of these techniques through
game play;
b. Building of a simplified Neural Model from a top down perspective, looking at which regions of the
brain do what, i.e. structure-function;
c. Verification/Validation of the model in terms of structure;
d. Testing of the model in game theoretical situations;
e. Analysis of results/refining of model;
f. Building of an evolutionary system using relevant computational techniques;
g. Verification of the system through simulation of self control through precommitment (the model must
encompass a cognitive architecture that provides a general explanation of self control);
h. Simulation of the behaviour for anomalies (e.g. the reduction in value of a reward due to delay and the
reversal of preferences as shown in Figure 1a).
Deliverables will take the form of refereed articles to be presented at relevant learned society journals and prestigious
international conferences. The research will benefit from close collaboration with the Psychology Department of
Birkbeck College through Dr. Denis Mareschal and Dr. Peter Sozou from the CoMPLEX (Centre for Mathematics and
Physics in the Life Sciences and Experimental Biology), University College London, with whom we will have monthly
meetings for updating them on the research progress and for receiving their feedback from their point of view (letters of
support are included from both). We will also seek new relevant collaborations. The principal investigator (who will
devote eight hours per week to this project) will be responsible for the efficient, productive, technical and administrative
management of the project. He will have regular meetings with the researcher to discuss and resolve technical issues; to
define the direction of the project and identify solutions if required; to control and review technical work carried out in
the tasks; to suggest technical modifications according to the aims of the project; to verify the correct implementation of
the project and to monitor the project progress and evaluate the deliverables. All these will be facilitated by progress
reports at regular intervals, which the researcher will be asked to prepare. The researcher will also attend short courses
in relevant research areas, like computational and cognitive neuroscience.
3. Relevance to beneficiaries
A detailed theory for the evolution of self control promises to provide a foundation for a science of self control which
will eventually be able to predict both the circumstances expected to induce greater self control and the forms of self
control induced. The theory will enable problems of self control to be identified and rectified, and thus it will have a
significant impact in the area of healthcare with direct beneficiaries being the general public. This will facilitate the
advancement of academic knowledge in related disciplines. The main field of study is computational neuroscience;
however the research touches on a wide spectrum of subjects, most notably cognitive science and game theory. The
5
extent to which reinforcement learning can be realised in games that model real life situations will interest game theorists
and economists. The results from the later stages of the project will contribute to bridging the gap between the modelling
community and experimentalists.
4. Justification of Resources
Funding is requested for a research assistant for a 36-month period. This will be Gaye Banfield who will carry out the
tasks detailed in Section 2.3 and the Diagrammatic Work Plan (attached) under the supervision of Dr Christodoulou. A
laptop is requested along with the software MATLAB plus toolboxes for Neural Networks. The developed system will
also be tested on departmental server resources for which 10% of a technical support person is requested. Clerical
support of 5% of a person is also requested for assisting with the administration of the project. Due to the crossdisciplinary nature of the research it is important to present results to both the psychology and computer science
communities and we request support for one national conference per year (Neural Computation and Psychology
Workshop; two people to attend) and two international ones (Cognitive Science conference; one person to attend).
Support is also requested for Gaye Banfield to attend an advanced course in Computational Neuroscience for learning
the state of the art in the subject.
References
Ariely D., 2002, Procrastination, Deadlines, and Performance: Self-Control by Precommitment, MIT Press, MA.
Axelrod R. and Hamilton W.D., 1981, The Evolution of Cooperation, Science 211, 1390-1396.
Banfield G. and Christodoulou C., 2003, On Reinforcement Learning in two player “real-world” games, In Proc. ICCS
ASCS Int. Conf. on Cognitive Science, 22.
Banfield G. and Christodoulou C., 2004, Multiagent Reinforcement Learning in General Sum Games, Submitted to the
IEEE Transactions on Evolutionary Computation.
Barto A. G. and Sutton R. S., 2002, Reinforcement Learning: An Introduction, MIT Press, Cambridge, MA.
Binmore K., 1992, Fun and Games A text on Game Theory, D. C. Heath and Co., Lexington, MA.
Brown J. and Rachlin H., 1999, Self-control and Social Cooperation, Behavioral Processes 47, 65-72.
Bullinaria, J. A. 2004, On the Evolution of Irrational Behaviour. In H. Bowman and C. Labiouse, editors, Proceedings of
the 8th Neural Computation and Psychology Workshop, Connectionist Models of Cognition and Perception II, vol.
15 of Progress in Neural Processing, Singapore, April 2004. World Scientific.
Cohen, J. D., Dunbar, K. and McClelland, J., 1990, On the Control of Automatic Processes: A Parallel Distributed
Processing Account of the Stroop Effect, Psychol. Rev. 97, 332-361.
Cohen L. B., Chaput H. H., and Cashon C. H., 2002, A Constructivist Model of Infant Cognition, Cognitive
Development 100, 1-21.
Fahlman, S., and Lebiere, C., 1990, The Cascade-Correlation learning architecture, Technical Report CMU-CS-90-100,
Comp. Sci. Dept., Carnegie Mellon University, Pittsburgh, PA.
Fodor, J. A., 1983, The Modularity of Mind, MIT Press, Cambridge, MA.
Green, L., Fry A. F., and Myerson J., 1994, Discounting of Delayed Rewards: A Life Span Comparison, Psychological
Science 5, 33-36.
Hamburger H., Games as Models of Social Phenomena, W.H. Freeman and Co, San Fran. US, 1979.
Holland J.H., 1992, Genetic Algorithms, Scientific America 267, 66-72.
Jacobs R. A., 1999, Computational Studies of the Development of Functionally Specialized Neural Modules, Trends in
Cognitive Science 3, 31-38.
Matsui M., Yoneyama E., Sumiyoshi T., Noguchi K., Nohara S., Suzuki M., Kawasaki Y., Seto H. and Kurachi M.,
2002, Lack of Self-Control as Assessed by a Personality Inventory is Related to Reduced Volume of Supplementary
Motor Area, Psychiatry Research Neuroimaging 116, 53-61.
Mischel, W., Shoda, Y., and Rodriguez, M., 1989, Delay of Gratification in Children, Science 244, 933-938.
Nesse R. M., 2001, Natural Selection and the Capacity for Subjective Commitment. In R. M. Nesse (Ed.), Evolution and
the Capacity for Commitment, (pages 1-44), New York, Russell Sage.
O’Reilly R. and Munakata Y., 2000, Computational Explorations in Cognitive Neuroscience, MIT Press, MA.
Rachlin H., 1995, Self-Control: Beyond commitment, Behavioural and Brain Sciences 18, 109-159.
Rachlin H., 2000, The Science of Self-Control, Harvard University Press, Cambridge, MA.
Rachlin, H. and Green, L. 1972, Commitment, Choice and Self-control, Journal of the Experimental Analysis of
Behavior 17,15-22.
Rubinstein A., 1982, Perfect Equilibrium in a Bargaining Model, Econometrica 50/1, 99-109.
Samuelson L. & Swinkels J. M., 2002, Information and the Evolution of the Utility Function, Journal of Economic
Literature, Working Papers 6, Wisconsin Madison - Social Systems.
Shultz, T.R. and D. Mareschal, D., 1996, Generative Connectionist Networks and Constructivist Cognitive
Development, Cognitive Development 11, 571-603.
Sozou P. D., 2003, The Evolutionary Context of Self-Control Problems, Presented at the Workshop on the Evolutionary
Biology of Learning, Fribourg, Switzerland, 21-22 February 2003.
Taylor J.G., 2002, Paying Attention to Consciousness, Trends in Cognitive Science 6, 206-210.
Trivers, R., 2000, The Elements of a Scientific Theory of Self-Deception, Annals of the New York Academy of Sciences
907,114-131.
6
Modelling of Self Control through Precommitment Behaviour in a Functionally
Decomposed Connectionist Architecture
Diagrammatic Work Plan
Task Description
Identification of
possible techniques
(ANN with RL, GAs)
and testing of these
techniques through
game play
Building of a
simplified Neural
Model from a top
down perspective,
looking at which
regions of the brain
do what, i.e.
structure-function
Verification/
Validation of the
model in terms of
structure
Testing of the model
in game theoretical
situations
Analysis of
results/refining of
model
Building of an
evolutionary system
using relevant
computational
techniques
Verification of the
system through
simulation of self
control through
precommitment
Simulation of the
behaviour for
anomalies
4
8
12
16
20
24
28
32
36
4
8
12
16
20
24
28
32
36
Final Report
Months
7