Anonymous Code: 1589798

Number of Words: 2000

Application of Algorithm in Decision Making

(Does Algorithm Mollify or Magnify Bias in Decision Making?)

Introduction
In the era of digitalization many tasks that were performed by human are now handing
over to algorithms .In every industry from healthcare, education, energy, business solution to
agriculture we can observe the existence of algorithm. We have become so reliant on algorithm
that we are giving algorithms our decision making platform without any hesitation. For example
many companies are now using algorithm to shortlist CV’s of job applicants, algorithm is now
deciding whom to call for interview and whom to not.

According to Hill, “Algorithm is an abstract, effective mathematical construct that accomplish a

given task under given instruction.” An algorithm is a procedure that solves problem. According
to European Parliamentary Research Service, Algorithm is consists of a set of instruction that take
some input data, process the data and produce data as output. In other words algorithm is a series
of instruction that solve a specific problem. Generally an algorithm is coded by a programmer or
generated automatically. Algorithm prioritize, classify, associate and filter data for decision
making (Diakopoulos, 2016). Effectiveness of Algorithm depends on data. And the most important
one is training data. Algorithm use training data to generate prediction scores and then select
anyone who has score more than the threshold score (Runshan Fu & Srinivasan).

The use of algorithm in decision making is increasing terrifically because human decision makers
are coming up with wrong decisions, they take long time to make decisions, they have less capacity
to handle big data, their decision making capacity is triggered by internal factor like anger, love,
fear, joy as well as external factors like time of the day, noise and so on. This is causing in
consistency in their decision making capacity. On the other hand, algorithms are well defined, they
mostly make accurate decision within a limited time, can handle big data properly and are not
influenced by emotion and noise. Algorithms are consistent in their decision making capacity.
However, algorithm is introduced with a goal to overcome above mentioned human limitations in
decision making. But the burning question is: Does Algorithm Mollify or Magnify Bias in
Decision Making?

1
 Anonymous Code: 1589798
Number of Words: 2000
Algorithm and Fairness

First of all let’s scrutinize the evidence that has proved algorithm as unbiased or fair
in terms of decision making. Decision making algorithm is now widely and successfully used in
several areas like e-commerce, health, justice, banking and insurance. They are beneficial for both
individual and organization. It has been proven that the use of algorithm in decision making are
bringing big changes. It is reducing error and saving our time. Algorithm has excellent predictive
power. For instance an automated underwriter algorithm in mortgage lending industry predicts
default more accurately than a manual underwriter do which increase underserved applicants
approval rate than ever (Miller, 2018).Google’s smart compose algorithm can predict the word
sequence ,even before we write it predicts and suggests what could be the next word. One evidence
proved that algorithm can select best board members without any gender discrimination (Miller,
2018).

Moreover, algorithm are much fair in decision making than human being because they always
follow same process and emotional factors like anger and love have no influence on them(Lee,
2018). Fairness in decision making is treating everyone equally and considering all while making
decision. Algorithm now-a-days not only is used in financial decisions but also in shopping
decisions.80% decisions and 30% decisions made on Netflix and Amazon simultaneously are
result of algorithm (Dalio, 2019).Use of algorithm in bailing decision making reduced jailing by
41.9% without increase in crime (Dalio, 2019). According to Cowgill (2017), algorithmic hiring
decision making is less biased than human decision making.

Algorithm can make fair and unbiased decision as it is not influenced by greed, prejudice, fatigue
or hunger whereas the human being is. We human being might do hurry in decision making, we
follow system 1 – our gut feelings and intuition while making decision which result in biased
outcome (Bazerman & Moore, 2012) . On the other hand algorithm follow system 2- it is objective
and deals with analytics. So, the decision based on algorithm becomes bias free. We human being
suffers from confirmation bias- we search feedback that support our decision and avoid the one
that does not support our’ decision. But algorithm is free from confirmation bias which help it to
generate bias free outcome. That means algorithmic decision making reduce bias than the human
oriented decision making.

2
 Anonymous Code: 1589798
Number of Words: 2000
Algorithm and Biasness
Now we will dig into the incidents that show the presence of biasness in algorithmic
decision making. Bias is something that is problematic and should be avoided. Bias is the deviation
from standard (Danks & London, 2017).In other word algorithm bias is unfair treatment of group
including ethnic minority group, religious group and gender by algorithm (Michel Rovatsos &
Koene).According to Professor Louise Amoore of Durham University ‘’Bias is inherent to the
algorithm-that means to operate it needs to give weight to some information over others”. Even
unintentionally algorithm like human can generate biased outcome.

According to House of Common Science and Technology Committee “Biasness in algorithm

occurs due to inappropriate data, inadequate data or generate automatically at any stage of
algorithm lifecycle”. For example programmer of recruitment decision algorithm might delete all
data related ethnicity before giving input but still automatically algorithm might be racist by
getting zip code data as input. Bias may also arise from programmers or designers and from
technical constraints.

Use of algorithm in decision making does not reduce bias rather it produce different types of bias
(Linda J.S. & B, 1999) .The prime source of bias in algorithm is training data. Training data is data
that teach algorithm and helps algorithm to identify patterns. Therefore if training data is biased
then the algorithm will produce bias outcome. Biasness in training data that feed the algorithm is
generally hidden.

Bias can also be produced inherently in the design of an algorithm, for example, if consciously or
unconsciously variables are incorrectly weighted by programmer (Michel Rovatsos & Koene).The
programmer of the algorithm decides which data to include and which data to exclude .That is why
the extent to which an algorithm will be unbiased depends on the extent to which its creator is
unbiased (Mathisen, 2017). Algorithm can take over the bias of previous decision maker. The
prejudice of the author willingly or unwillingly are transferred into the code. This type of bias is
latent. Social bias or individual bias is implanted into the algorithm purposefully by the
programmer for example the ranking criteria (Goldman, 2006). It can also be unintentional in few
situation. Whereas technical bias arise from technological error.

Algorithm can adopt flaws in data and can be hidden in output (Romei & Ruggeri,2014).So, there
is also evidence that algorithm produce bias outcome.

3
 Anonymous Code: 1589798
Number of Words: 2000
Guidelines
Algorithm is widely used in decision making and it has a great impact on our life. As algorithm
has profound impact on us so it is crucial to bring it under certain set of rules. Biasness in algorithm
can be reduced by intervention of human in terms of accountability and creating transparency.

Accountability and Transparency

Accountability is liability of someone for the specific outcome. Algorithmic accountability might
be lie on programmer, owner and operator who will give clarification if the algorithm is biased or
failed to generate intended outcome. Accountability can be facilitated by introducing transparency
in the model. And transparency is how understandable a model is to the stake holders. But is
difficult to introduce transparency in algorithmic decision making. According to Rebecca
Mackinnon from think tank New America, Even to the programmer algorithmic decision making
process quickly become cloudy who don’t understand the logic behind the outcome. However
transparency is important for increasing acceptability of decision made by algorithm. To increase
acceptability each stage of algorithmic decision making should be transparent. To get more
acceptance algorithm should explain the logic behind the decision to its user.

Auditing
Audit is crucial to build trust in algorithm. When any decision harm individuals and group data
controller can go for auditing to know how it occurred and in which condition it might be happened
in future again.

Compliance with the regulations

To reduce the biasness in algorithmic decision making it is necessary to bring the model under the
law. The following European Commissions AI Ethics Guideline besides accountability and
transparency can be adopted to reduce biasness:

 Including fundamental rights, human agency and human oversight.

 Technical robustness and safety
 Privacy and data governance
 Diversity, Non-discrimination
 Societal and Environmental Wellbeing
As algorithmic decision making has high societal consequence so it is crucial that the technology
is used to serve the people and planet.

4
 Anonymous Code: 1589798
Number of Words: 2000
Conclusion
Subsequently, algorithm is taking over the decision making floor from we human being. In this
high digitalized world algorithm is making very critical decisions frequently. In a wide range of
past cases it has been found that human decision makers are becoming puzzled to deal with big
data, they are inconsistent with their decision making capacity, they were in accurate, their
forecasting include lot of errors and they did wrong interpretation. Very simple algorithm
outperforms expert in some situations.

Hence, to alter this scenario humans are now highly relying on algorithms. From previous
literatures and from the discussion in this essay we now know that Algorithms can predict future
scenario with minimal forecasting error, they are free from materialistic factors, physical and
psychological factors. They deliver more effective outcomes. Replacing human with algorithm
result in accurate and biasfree decision.

However, there is also some evidence that algorithm is somewhat bias in making decision. This
biasness occurs due to training data, can comes from the values of the programmer and also from
technological issues. This biasness might be either intentional or unintentional. But the biasness in
algorithm is not as severe as in human. Biasness in algorithm can be reduced by the intervention
of human, by developing accountability, transparency and by adopting ethical code of conduct
designed for the algorithms.

That means algorithm is not completely free from bias but the level at which an algorithm is biased
lower than that of a human. So, we can come up with the answer of previously mentioned question
that- algorithm does not magnify bias in decision making rather it mollify or sometimes
replace the current bias in another form. As it is difficult for both human and algorithm to come
to a decision so, in this digital society the possible best solution is not the human or machine rather
it is human and machine .And by the chemistry of both human and machine better decision can be
made.

5
 Anonymous Code: 1589798
Number of Words: 2000
Appendix
A) References
Books and Journal
Bazerman, M. H. & Moore, D. A., 2012. Judgment In Managerial Decision Making. Eighth ed.:Willey.

B, C., 2017. Automating judgment and decision-making:Theory and evidence from resume screening.
New York: Working paper ,Columbia University.

Danks, D. & London, A. J., 2017. Algorithmic Bias in Autonomous System. Pittsburg,USA, Center for
Ethics and Policy.

Diakopoulos, N., 2016. Accountability in algorithmic Decision Making. Communication of the ACM,
59(2), pp. 56-62.

Lee, M. K., 2018. Understanding perception of algorithmic decisions: Fairness, trust, and emotion
inresponse to algorithmic management. Big data and society, pp. 1-16.

Linda J.S., k. L. & B, M., 1999. Does Automation Bias Decision Making. Int.J. Human Computer Studies,
Volume 51, pp. 991-1006.

Michael R, B. M. & Koene, A., n.d. Landscape Summary:Bias in Algorithmic Decision-Making. Center
for Data etics and Innovation.

Miller, A. P., 2018. Want Less-Biased Decisions? Use Algorithms.. Harvard Business Review, 26 07.

Romei A. & Ruggieri S., 2014. A multidisciplinary survey on discrimination analysis. The Knowledge
Engineering Review, 29(5), pp. 582-638.

RK, H., 2015. What an algorithm is?. Philosophy and Technology, 29(1), pp. 35-59.

Runshan Fu, M. A. P. V. S. & Srinivasan, K., n.d. ''Un"Fair Machine Learning Algorithms.

Service, E. P. R., 2019. Understanding Algorithmic Decision- Making:Opportunities and Challanges.

Web site
Dalio, R., 2019. Bondora. [Online]
Available at: https://www.bondora.com/blog/the-future-of-algorithmic-decision-making/
[Accessed 28 09 2019].

Available at: https://www.publictechnology.net/articles/opinion/algorithms-decision-making-inquiry-

stephanie-mathisen-challenging-mps-investigate
[Accessed 28 09 2019].

6
 Anonymous Code: 1589798
Number of Words: 2000
B) Slide Package

7
Anonymous Code: 1589798
Number of Words: 2000