The article deals with the current issues of classification of the challenges that arise and the principles that should been used, during the development and implementation of artificial intelligence systems. In the subject area of AI ethics, the notion of an AI system ethics index has been introduced. The author made a detailed analysis of ideas and methods in the subject area of ethics of artificial intelligence and proposed a general approach for quantifying the level of ethics of developed systems by classifying the main challenges, evaluating them and introducing compensatory measures. The approach reflects the general idea, which could been detailed by specialists from the respective subject areas. The research is purely theoretical in nature, summarizing existing ideas and principles, for the first time putting forward the idea of a quantitative assessment of the question of the ethics of artificial intelligence.
The intensive development of information technology in the last decade, especially in the field of
artificial intelligence and hardware in the form of neurosynaptic and quantum computers, poses new
challenges to society in its harmonious development in terms of moral and ethical issues, as well as
information security. Numerous AI programs by the world's leading governments published in the past
few years also highlight the urgency of the ethical issues that will arise as these technologies develop.
plan to overcome limitations on AI technology” [
The UK government publishes its “AI Sector Deal” which invests 950M pounds (1.2B USD) to
support research / education, and enhance the UK’s data infrastructure [
Since 2014-15, public, private companies, educational and research institutions have begun to
publish various regulatory documents, materials related to the ethical issues of the development,
implementation and application of artificial intelligence systems. The importance of moral issues in
information systems and AI are also been evidenced by the research on ethical issues highlighted in
separate sections of the systematic AI index reports [
unite these documents, among them: confidentiality, accountability, transparency, and explainability.
The very fact that such documents appear shows that society is beginning to pay serious attention to such a difficult issue as ethics and human rights in the field of artificial intelligence. However, criticism should been noted, that has arisen from experts in ethics and human rights in connection with the possible ambiguous or inaccurate use of existing terms in this field.
2022 Copyright for this paper by its authors.
The abstract nature of the introduced principles does not allow us to speak about their adequacy from the point of view of a correct description of the subject area and, accordingly, about the possibility of their use to control compliance with ethical norms in the field of AI.
Research [
engagement
development and use of AI systems, respectively
process in a clear and understandable manner
development and operation of AI systems
This list is not exhaustive, but shows general trends in AI. Researches show that fairness, interpretability and explainability, transparency are most mentioned across all documents studied.
Research [
In [
In research report [
The research methodology involves the study of the key causes of disagreement between the fields of research in artificial intelligence and ethics, as well as the classification (formalization) of the basic concepts of the field of study. Obviously, all the disagreements between AI specialists and ethicists in its classical sense arise from different interpretations of AI terms in terms of ethics. It is necessary to turn to the definition of ethics and the tasks it addresses in a broad and narrow sense.
Ethics is a philosophical discipline whose subject is morality. Ethics has two main functions –
moraleducational and cognitive-educational, so two areas could been distinguished in ethics - normative
ethics, aimed at teaching about life, and theoretical ethics, studying morality. [
Theoretical ethics is a scientific discipline that examines morality as a special social phenomenon, finds out what it is, how morality differs from other social phenomena. Theoretical ethics studies the origin, historical development, regularities of functioning, social role and other aspects of morality. Its methodological basis is the knowledge, concepts and ideas concerning the scientific knowledge of morality. Normative ethics searches for a principle (or principles) that governs human behavior, guides one's actions, establishes criteria for evaluating the moral good, and a rule that can act as a general principle for all cases.
Applied (practical) ethics studies particular problems and the application of moral ideas and principles articulated in normative ethics to specific situations of moral choice. Applied ethics interacts closely with the social and political sciences and has a number of sections, e.g. business ethics, medical ethics, computer ethics, etc. Obviously, AI ethics could been classified as a section of applied (normative) ethics, which is very close to information (computer) ethics. Let us make a brief excursion into information ethics and consider its principles.
Broadly speaking, computer ("information" or "cyber ethics") ethics investigates the behavior of
people who use information systems, based on which appropriate moral precepts and norms of behavior
are developed [
3. Cyber security, determination of the status of incidents and crimes, that is, predominantly legal problems, as a rule, formalized in the form of national legislative acts on information security.
Principles developed in computer ethics: 1. Privacy – a person's right to autonomy and freedom in private life, the right to be protected from intrusion by authorities and others. 2. Accuracy (accuracy) – compliance with the norms related to the accurate execution of the instructions for the operation of systems and information processing, honest and socially responsible attitude to their duties.
3. Property – inviolability of private property. Adherence to this principle means observance of the right of ownership of information and copyright norms.
4. Accessibility – the right of citizens to information, its accessibility at any time and in any place.
The principles of information (computer) ethics developed are very similar to the ethical principles of AI that we reviewed in the literature review, but they relate only to data processing and issues of security and ownership.
The only issue is that not all these principles have been systematized in the framework of a corresponding standard and each developer tries to take into account all possible cases of AI impact on society, which leads to their duplication, different interpretation of sometimes the same principles and challenges. For the same reason various so to speak analytics are also mixed, for example, direct impact on society and incorrect historical data, etc.
The main ethical challenges could been divided into several main groups (fig.1).
All challenges and problematic issues that arise in one way or another during the development of AI systems can be attributed either to the issues of data and algorithms, or to those that affect the economy and society, or conflict with existing legal norms. Very often the challenges are complex and can simultaneously positively affect the economy as a whole, while also causing social contradictions, such as a couple of productivity and employment issues. In the aspect of data processing, special attention should been paid to the creation of test (training sets) of data, which require strict adherence to the principles: data neutrality, representative data, accuracy, reliability, openness and diversity.
Among the many approaches to classifying ethical principles in AI issues, the author would distinguish four large global groups (fig.2). Figure 2 shows the classification of ethical principles into four main groups, as well as the relationship of these groups to the principles described in papers [911]. Group “Safety and Security” includes all the principles that describe the security and protection of both data and its accuracy and reliability in order to create a secure information technology. A very important issue is the adequacy of historical data to create correct training samples, which affects both the manageability of AI and social responsibility. Group “Manageability (Controllability)” describes the principles that must been followed to create AI software that is manageable, efficient, understandable to the end user, and controllable by the end user. The principles of this group also imply a cautious attitude toward the prospective capabilities of the AI system are being creating, which have not yet been fully clarified by the developer. The development should been conducted at a high scientific and technical level. The system must be reproducible under different conditions. The developer must take into account all risks in operating conditions.
The group “Social Responsibility and law” includes principles that characterize the AI system to be developed in terms of compliance with social and legal norms of society. Group “Benefits” describes the principles for assessing the usefulness of the developed AI system from both a tangible and intangible point of view for the user and society as a whole.
There is a group of principles (bottom of Figure 2): data neutrality, representative data, accuracy, reliability, openness, diversity, accessibility, accountability, auditability that are common to the three groups. These principles relate to data, but their implementation lays the foundation for addressing safety, manageability, and social responsibility. Using the developed classifiers of challenges and principles, let us introduce an index of AI system ethics (1):
N EAI Pi Pi IN ; , (1)
i1 lim Pi max;Pi 0
EAI max negative; Pi - is an evaluation of the principles that the AI system fulfills, (+) means positive impact, (-) – P IN i
- is a principle (initiative, IN) necessary to compensate for negative influences.
The main condition for using formula (1): the principles of development must correlate with the challenges posed by the development and minimize their negative impact on society from an ethical point of view. For example, automation can lead to increased productivity and, as a consequence, to job losses. This is a serious challenge with the highest ethical rating. We can rate the principle – P 5 fundamental rights and freedoms (right to work) on a maximum scale, for example – 5 points, i . At the same time, it will reduce the length of the working day, improving social conditions (if there is such a point in state programs, for example), so we can assess this initiative also on the maximum scale
P IN 5 – i . Thus, we are talking about the implementation of initiatives (recommendations for legislatures and businesses), which can minimize the impact of negative factors (challenges).
The author's top-level classification of AI ethics principles could been detailed by independent
research and introduced as a standard after agreement with all stakeholders. The Index of AI Ethics is
been considered as a general approach to the evaluation of developed AI systems. It requires further
study in terms of a detailed classification of principles in the form of a final list, which will be included
in the upper level groups proposed by the author. The development of evaluation scales for quantitative
assessment of the ethics of AI systems is also been envisaged. The results of research in the field of
ethics of artificial intelligence are closely intertwined with research and approaches to quantify the
technological singularity proposed in work [