Page 150 - Proceedings of the 2018 ITU Kaleidoscope

2018 ITU Kaleidoscope Academic Conference

ascertain what the machine is learning and from where. By classified learning‘s from the existing body of work and
the year 2020, it is estimated that the total number of based on this attempted to propose a unique conceptual
Internet-connected devices being used will be between 25 framework that is expected to lend strong ‗ethical flavours‘
and 50 billion. As the numbers grow and technologies to the future attempts of the researchers, designers and
become more mature, the volume of data published will policy makers associated with the domain of ML.
increase [33].However, this vast quantity of data that is
available for ML should be correct, properly labeled, and 4. EXISTENT FRAMEWORK TO MANAGE
right data [25]. There is however, a distinction between ETHICAL ISSUES
right data and data is right standards. Data is right
standards may not be met as it might not be right data but Efforts are being made to address the concerns arising from
also because there may be human errors during data the use of AI and ML by proposing
collection; data creators may not know the purpose for guidelines/principles/frameworks. For instance, Isaac
what it will be used; poorly calibrated measurement tools Asimov (1942) in his short story ―Runaround‖ introduced
etc. the ―three laws of robotics‖ to which fourth, or zeroth law
was added by him later on in the year 1985 [29]. He
The learning based on the quality of data and algorithms envisioned a world where human-like robots would act
will inevitably lead to the machines learning both good as like servants and would need a set of programming rules to
well as bad practices. Machine ethics is not merely science prevent them from causing harm[30].These laws
fiction; it is a topic that requires serious consideration, enunciated in a work of fiction are still mentioned as
given the rapid emergence of increasingly complex template for guiding development of robots [30] while at
autonomous software agents and robots. Machine ethics is the same time being questioned for their relevance [31].
an emerging field that seeks to implement moral decision- Moving on, the World Commission on the Ethics of
making faculties in computers and robots [9]. The Scientific Knowledge and Technology (COMEST) of
increasing cases reported worldwide of machines harming United Nations Educational, Scientific and Cultural
humans or turning incapable of being assistance to humans Organization (UNESCO) proposes a technology-based
shows how ethical concerns in the use of machine learning ethical framework on robotics ethics based on the
is an urgency that needs to be addressed quickly. The distinction between deterministic and cognitive robots
technological advancement should not take place in (October, 2017). These relevant ethical principles and
isolation rather it should come up with ways to mitigate values include: (i) human dignity; (ii) value of autonomy;
accidental or intentional harms caused by the Artificially (iii) value of privacy; (iv) ―do not harm‖ principle; (v)
Intelligent machines relying on Machine Learning [10]. principle of responsibility; (vi) value of beneficence; and
(vii) value of justice. The principle of human responsibility
In information societies, operations, decisions and choices is the common thread that joins the different values that are
previously left to humans are increasingly delegated to enunciated in the report [12].
algorithms, which may advise, if not decide, about how
data should be interpreted and what actions should be taken Wallach and Allen (2009) suggest two basic approaches of
as a result [11]. implementing machine morality: top-down and bottom-up
— as well as a hybrid approach [13] . The top-down
3.1 Methodology approach (Figure 1) is concerned with borrowing moral
frameworks from philosophers including Kant (1724-
The methodology adopted for this study is descriptive, 1804),Hegel (1770-1831), Hume (1711-1776) and concepts
exploratory, and analytical in nature based on review of like utilitarianism et al ( referred in the previous section)
literature. Secondary sources, related to the domain of both and the system relies on these set of moral guidelines to
ML and ethics, have been referred to glean the existing make decisions in future. In the bottom-up approach the
philosophies and frameworks for which scholarly machine learns through manipulation just like a child learns
publications, related articles, research reports, and books morality while growing up similarly the machine through
have been examined. Since ML is an emerging domain, evolutionary algorithms learns to make ethically acceptable
therefore references to social media have been made decisions. Wallach however also points out to the
wherever appropriate. However, such public domain drawbacks of each of these approaches [14].
references might not have the desired academic rigor but
such references have provided authors a larger Recognizing the fact that powerful technologies like AI
understanding of emerging aspects of ML. These generic raise questions about its use, industry leaders have been
references have also helped the authors to debate on varied proposing set of principles/guidelines to guide this area.
interpretations of emerging trends and have further aided in Nadella, Microsoft CEO outlined ten essential rules to
creating a bibliography to locate other important secondary approach AI (2016). The first six rules on the list discuss
sources. After collating all these aspects of the nascent what an ideal AI should contain or do. The remaining four
domain, including the existing frameworks elaborating the rules stress on few important attributes that must be
interplay of ML and ethics. The authors have further incorporated [34]. These rules clearly state that AI must

– 134 –