Page 151 - Proceedings of the 2018 ITU Kaleidoscope

Machine learning for a 5G future

prevent bias and also emphasize the need for empathy, given field, the push for greater dependence on ML and AI
education, creativity and focus on judgment and creativity. is increasing with Big Data. However, the central question
Similarly, Pichai (2018), has outlined the seven principles at the heart of this discussion should be that are these
that will guide Google in its work forward in this area [15]. machines really trustworthy and free of human prejudice
The seven principles are: (i) be socially beneficial; (ii) only because they provide results based on numerical
avoid creating or reinforcing unfair bias; (iii) be built and calculations. The algorithms or codes which are fed into the
tested for safety; (iv) be accountable to people; (v) machines themselves are flawed, as they are encoded
incorporated privacy design principles; (vi) uphold high opinions of human beings who could be involved with the
standards of scientific excellence; and (vii) be made machine at any given stage in its production and training;
available for uses that accord with these principles. Of which involves: gathering data, cleaning data, choosing
particular importance is the principle that lays down ―avoid algorithms, testing algorithms, selecting models, testing
creating or reinforcing unfair bias‖. It is acknowledged in models, refining models and finally reaching at the
the post that ―AI algorithms and datasets can reflect, operational model [18].
reinforce, or reduce unfair biases. We recognize that
distinguishing fair from unfair biases is not always simple, The flawed and biased opinions and prejudices of the
and differs across cultures and societies. We will seek to human designers find ways to influence the machine
avoid unjust impacts on people, particularly those related to learning. In such a case it is thus highly doubtful if one
sensitive characteristics such as race, ethnicity, gender, may bestow complete trust in a machine‘s decisions
nationality, income, sexual orientation, ability, and political because these biased ideas only turn a machine into a
or religious belief.‖ [15] weapon for silently inflicting harm.To overcome these
issues a framework is being proposed to study the ethical
Given the increasing autonomy of AI and ML it has dimensions in Machine Learning. The two axes i.e. the x-
become an arduous task to pin down as to who should be axis and y-axis represents Emotional Quotient (EQ) and
held responsible for bearing the ethical responsibility for Spiritual Quotient (SQ) respectively. IQ remains the only
the behavior of the machine. Time has come to discuss the constant. The four quadrants represent the four areas of
ethics of the institutions and people behind the machines. learning (Figure 2).

The approach is suggestive and works on the premise that
the machine already possesses Intelligence Quotient (IQ).
Presence of IQ thus fulfils the condition of rational clarity
in decision-making. Moreover, the machine has access to
data both archaic and new (through IOT and Big Data). On
the other hand, EQ is used to express ‗emotional
intelligence‘ in the same way as IQ is used to express
‗intelligence‘. Mayer and Salovey (1990) offered the first
formulation of a concept they called ‗emotional
intelligence‘ [35]. It refers to the ability to process
emotional information accurately and efficiently. Goleman
(2013) presented a model of E.I. that essentially comprises
of five areas: self-awareness, self-regulation, empathy,
social skills and motivation [36]. Similarly the term
Figure 1 - Moral agency increases as autonomy and ‘Spiritual intelligence‘ is a term that has been used by
ethical sensitivity increase. [13] some philosophers, psychologists, and developmental
theorists to indicate spiritual parallels with IQ (Intelligence
5. ‘ETHICAL FRAMEWORK’ : PROPOSED Quotient) and EQ (Emotional Quotient). Zohar (1997)
NEW FRAMEWORK TO MANAGE ETHICAL coined the term "spiritual intelligence" outlining twelve
ISSUES IN ML underlying principles of self-awareness, spontaneity,
empathy, holism, compassion, being vision and value led,
This inherently interdisciplinary field of machine ethics celebration of diversity, humility etc.[37].
actually lies at the intersection of domains of philosophy,
cognitive science, psychology, computer science and The proposed conceptual framework is based on the
robotics [14]. Today the machines have surpassed human premise that the person designing/ developing ML
intelligence and found groundbreaking solutions. For (henceforth, referred to as ‗designer‘, for ease) would
instance, Daniel Lobo and Michael Levin at Tufts always have IQ whereas the other two quotients viz. EQ
University devised an evolutionary algorithm which and SQ would oscillate from ‗Low‘ to ‗High‘ values in her
successfully unraveled mysteries of the planarian worm‘s personality type / design approach. This would lead to four
regenerative biology [16]. Touted as accurate predictors personality types in a ML designer depending upon the
and efficient tools that increase accuracy when used in any variance of these two quotients in two stages ( Low and

– 135 –