Page 148 - Proceedings of the 2018 ITU Kaleidoscope

ETHICAL FRAMEWORK FOR MACHINE LEARNING

2
1
Charru Malhotra , Vinod Kotwal , and Surabhi Dalal 3
1 Indian Institute of Public Administration, India
2 Department of Telecommunications, India

3 India Centre for Migration, India

ABSTRACT considered a subset of Artificial Intelligence (AI) where
algorithms directed by complex neural networks teach
Artificial Intelligence (AI) with its core subset of Machine computers to think like humans while processing ―big data‖
Learning (ML) is rapidly transforming life experiences as and calculations with high precision, speed and supposed
humans begin to grow more dependent on these ‘smart lack of bias [3].
machines’ for their needs – ranging from routine mundane
chores to critical personal decisions. However, these Amongst many current applications, ML is being widely
transformative technologies are at the same time proving used to predict weather conditions, medical diagnosis,
unpredictable too as has been reported worldwide in outcome of elections, facial recognition, criminal justice
certain cases. Therefore, several studies/reports, such as system, make predictions about credit worthiness, examine
COMEST report on Robotics ethics (UNESCO, 2017) point customer churn, automated traffic signals, targeted
to an obvious need for inculcating more ethical behavior in advertising etc.
machines. The present study aims to look at the role and
interplay of ML (the hard sciences) and Ethics (the soft The speed of growth of ML and its pervasiveness in our
sciences) to resolve such predicaments that are daily lives is being fueled by exponential growth in the ‗big
inadvertently manifested by machines not constrained or data‘. However, bad data impacts the quality demands of
controlled by human expectations. Based on focused ML in two ways, first on the historical data used to train the
review of literature of both domains-ML and Ethics, the predictive model and secondly the new data used by that
proposed paper attempts to first build on the need for model to make future decisions. Thus, amongst other
introduction of an ethical algorithm in the domain of things, data must be right data (unbiased data) [25]. The
machine learning and then endeavors to provide a ML algorithms running on the data are benefiting
conceptual framework to resolve the ethical dilemmas. humanity, at large, but recent research is also uncovering
many instances of biases in ML algorithms [3].Increasing
Keywords – Ethics, Artificial intelligence/machine application of and thereby dependence on ML is thus
learning, design approach, spiritual quotient, raising associated concerns including legal & social issues
emotional quotient and ethical biases. The next question which then arises is
that how do we eliminate/reduce biases? This links it to the
1. INTRODUCTION question of ethics and ethical behaviour. Ethics affect every
decision of our lives and they are one of the differentiating
The field of study that deals with the development of principles of how humans react in a given situation as
computer algorithms for transforming data into intelligent opposed to machines that rely on machine learning and
action is known as machine learning (ML) [1]. The increase algorithms. Human beings possess conflicting moral
in ML is driven by simultaneous evolution of opinions therefore the judgments are subjective. Ethical
computational power and statistical methods to handle decision-making is essentially situational and the context
exponential collation and manipulation of data. Whenever defines what may be accepted as ethical or not. Every
an algorithm transforms itself into new actionable culture prescribes a certain ‗code of ethics‘ which govern
intelligence based on data, machine learning takes place. the group of people affiliated to that particular social or
ML learns from experience and improves its performance cultural group. In this backdrop, can we create machines
as it learns [1].Basic learning process has three components that follow universally accepted ethical
viz. data input, abstraction and generalization; these are principles/guidelines/framework?
equally applicable to ML as well as humans though in the
latter they take place subconsciously [2]. Thus, ML is best Etymologically, the term ―ethics‖ corresponds to the Greek
word ―ethos‖ which means character, habit, custom, way of

978-92-61-26921-0/CFP1868P-ART @ 2018 ITU – 132 – Kaleidoscope