PREFACE:

We are all bound to make decisions and our decisions, whether explicit or implicit, form our
existence and the world. This fact puts significant importance on the process of decision
making. So far, this phenomenon has been the cornerstone of the existentialist movement in
the west and many famous novelists and philosophers including Fyodor Dostoyevsky, Soren
Kierkegaard, Friedrich Nietzsche, Martin Heidegger, Jean Paul Sartre, etc. have written
passionately about it. As a result, the process of decision-making deserves reflection as it is the
main process which determines who we are and what world we live in.
A brief investigation in the history of mankind shows that man, in every period, had an
uncontrollable sense of curiosity about different physical and human phenomena. This fact
leads the people to use all they could to understand what they were involved with. we modern
humans are of course no exception. In every period, people had a special methodology for this
purpose. Now and in this era, collective wisdom and reflection in the philosophy of science
sought to put this effort on a robust foundation – Modern Science.
Modern science criticizes former theories as being unreliable and invalid each trying to analyze
a limited amount of data with an unreliable method and a limited computational power. Now
with the authority of modern science and the development of computers, a new field has born
which can investigate decision-making more broadly, precisely and reliably – computational
modelling. Decision-making Modeling, in essence can be categorized as a metacognition which
its main purpose is reflecting on an abstract phenomenon like decision by saying what it is and
how it is done. So, in this article, we’d like to delve deep into the sea of decisions with the
purpose of finding a unifying computational model of decision making.
Introduction1:
Human judgements are faced with a variety of limitations including limited time and available
data. In a world full of data and uncertainty, it seems difficult to interpret new information
based on prior information accurately and make optimal decisions. surprisingly, neuroscientific
evidence shows that our brains is capable of making adaptive decisions in an uncertain world.
What helps us to solve problems quickly and efficiently are heuristics. This phenomenon has
fascinated many researchers since Tversky and Kahneman’s influential work. As a result, the list
of heuristics found by scientists continuously grew larger. This fact emphasized the need to
categorize and describe heuristics more efficiently. There has been many attempts to do this.
These efforts may be considered as descriptive theories or models. So, there is a need to
incorporate all of them into a unifying model to have a comprehensive model of decision
making and this is the main goal of the author in this article which proposes a modular
computational model to do that. The behavior of this model is determined by tuning a set of
parameters. The value of these parameters makes the model able to capture some of the
heuristics that previously, are thought to have been used by humans in their process of
decision-making.
In contrast to descriptive models, a computational model is more precise and based on
procedures which can better model human intelligence and be used in artificial intelligence
applications as the main processor of machines, robots, computer programs, etc.
In addition to representation of decisions, a good model may try to represent how a decision is
made among different alternatives. This may be regarded as an underlying strategy of decision-
making which may have some special feature (for example, being fixed or learned).
Nevertheless, the author focuses mainly on compensatory vs non-compensatory mechanisms
as well as aggregation mechanisms which can be achieved by iterative decision mechanisms
and instantiated in voting methods from the field of social science.
although the mentioned model tries to describe heuristics in general, its parameters must be
tuned for each application in turn. Every set of parameters affects the output of the model, and
this fact is shown by using the traveling salesperson as an example.
Eventually the author states that adjusting any parameter in any step can be considered as
using a specific heuristic and viewed as another decision task, therefore, calculated with the
very same mechanism of the model.
Heuristics can be considered as strategies for quick decision-making. In the next section, two
descriptive approaches to model heuristics from the literature are formalized to be able to be
incorporated as a decision procedure in a computational model.

1
1- All sections are paraphrased or elucidated from the main article (Kirsch, A. (2019). A unifying computational
model of decision making. Cognitive processing, 20(2), 243-259. )
Formalization of heuristics:
when a set of alternatives are available (a i ϵ A), making a decision means choosing an a* in the
set. The choice is made by a process of quality assessment of each alternative through a
number of cues ci ϵ C. Cues can be formalized by either a q value function mapping alternatives
to real values or a q ranking function. As an example, Shah and Oppenheimer (2008) divide the
decision process into five tasks:
0. find a vector of alternatives a.
1. Identify all cues that is all relevant information which is equivalent of calculating vector c.
2. Recall and store cue values that is the values for all pieces of information. This may be done
by calculating matrix M:
c 1 (a 1) … c1 ( an )
⋮ … ⋮
c m ( a1) … cm (an )

3. Assess the weights of each cue. This is done by calculating a vector of weights W.
4. Integrate information for all alternatives—the weighted cue values must be summed to yield
an overall value or utility for the alternative. This is done by calculating a vector of utilities as
the multiplication of decision matrix and weight vector:u=M T .W

5. All alternatives must be compared, and then the alternative with the highest value should be
selected. This is the mathematical equivalent of choosing the alternative with the highest utility
value: a* = maxu a.
This algorithm used by Shah and Oppenheimer (2008) is called the weighted additive rule. By
using it to formalize heuristics, they suggest ways that can reduce cognitive effort such as
examining only a subset of available cues or omitting the weights.
Hertwig et al. (1999) used another way of reducing the cognitive effort by introducing the
Quick-Est heuristic approach which focusses on reducing the number of iterations to reach a
decision:
1. a vector of alternatives, a, is identified (retrieved from memory or the environment)
2. cues are ranked according to how fast they can eliminate as many alternatives as they can.
Thus, the vector c is calculated.
3. step 3 includes a way in which an iterative approach is used to determine the best alternative
sooner by using the vector c in the previous step and reducing the number of alternatives step
by step:
Repeat, starting with i ← 1 and ã ← a:
(a) Apply cue ci to alternatives in ã, resulting in a set of alternatives with positive evaluation ã+
and alternatives with negative evaluation ã-.
(b) alternatives with negative evaluations are dismissed and because cues are assumed to be
Boolean functions, if |ã+|=1, then ã+ is the best alternative, otherwise, the algorithm repeats
itself only with the values of ã+ as the alternatives:
repeat with i ← i + 1 and ã ← ã+

The integrated model:

In this section, the author combines the above-mentioned algorithm to find a unified model of
decision-making – DECIDE function (figure 2). This function has three arguments a, c and p
standing for alternatives, cues and parameters. The value of the arguments is calculated first by
using get () functions. The get () functions retrieves these values from environments, memory
or from the knowledge in the last iteration. In the next step matrix M is calculated by FILL-
DECISION-MATRIX which calls all values of c for all values of a. next, the alternatives are sorted
and ranked so that the highest ranked alternative is the first. Next, the best alternative is stored
in a*. next, function ACCEPTABLE judges whether the best alternative is good enough to be the
final decision or not. an alternative may be judged unacceptable when the next-best alternative
in the ranking has an equal or very close utility score. Or it may be rejected as being rated too
low on certain cue values. Eventually, If the alternative is rejected, the function DECIDE is called
again.

Fig 1- general model of heuristic decision making

Discussion and conclusion:
This article summarizes the important points in an article which tries to not only unify other
descriptive models of heuristics but also previous algorithms that try to formalize heuristics.
The author believes that the model provides a better framework to explore, model and
simulate the process of decision-making. To support this belief and applicability of the new
model, a case study of computational problem solving (the traveling salesperson) is presented
in the main article which we omitted in this article for the sake of briefness.
Reference:
Kirsch, A. (2019). A unifying computational model of decision making. Cognitive processing, 20(2), 243-
259.