=Paper=
{{Paper
|id=Vol-2853/short18
|storemode=property
|title=Formal Model of Trustworthy Artificial Intelligence Based on Standardization
|pdfUrl=https://ceur-ws.org/Vol-2853/short18.pdf
|volume=Vol-2853
|authors=Eduard Manziuk,Olexander Barmak,Iurii Krak,Olexander Mazurets,Tetiana Skrypnyk
|dblpUrl=https://dblp.org/rec/conf/intelitsis/ManziukBKMS21
}}
==Formal Model of Trustworthy Artificial Intelligence Based on Standardization==
https://ceur-ws.org/Vol-2853/short18.pdf
Formal Model of Trustworthy Artificial Intelligence Based on
Standardization
Eduard Manziuka, Olexander Barmaka, Iurii Krakb,c, Olexander Mazuretsa and Tetiana
Skrypnyka
a
Khmelnytskyi National University, Institutska str., 11, Khmelnytskyi, 29016, Ukraine
b
Taras Shevchenko National University of Kyiv, Volodymyrska str., 60, Kyiv, 01601, Ukraine
c
Glushkov Cybernetics Institute, Academician Glushkov Avenue, 40, Kyiv, 03187, Ukraine
Abstract
The widespread and rapid distribution and application of artificial intelligence (AI) systems
requires the development of formalized approaches and the construction of basic principles
for the functioning of domain areas of AI use. This need is embodied in the development of
recommendations and standards to obtain maximum benefits from the use of AI and
minimize possible risks. The regulatory framework is being built on a human-centric basis.
Accordingly, the developed standards should form the basis for further activities aimed at the
use of AI and be applicable at all stages of creating practical solutions. Therefore, an
important stage is the formalization of requirements, principles and provisions of legal and
ethical norms in the form of practical template approaches for practical application. With this
method, models and ontology of standardized concept of AI credibility are developed within
the research. This made it possible to identify the main concepts that allow forming a position
of trust, are a meaningful part of the concept of trustworthy AI, determine the need for its
existence and pose a threat to it. On the basis of ontology of the domain area, models were
developed and further decomposition of structural substantive concepts was carried out. In
the future, the characteristics of the concept of trustworthiness formation are defined.
Keywords 1
Human-centric AI, ethic AI, ontology, model, trustworthiness, standardization AI.
1. Introduction
The development of artificial intelligence systems today is undergoing significant growth and is
finding implementation in a wide range of practical tasks in various spheres of human life. The
problems solved by artificial intelligence (AI) are quite diverse, for example, such as object
recognition, languages, classification, clustering, etc. Prospects for the practical application of AI in
the areas of automatic car management, diagnosis of diseases in medicine, in the field of finance,
safety are achievable and everywhere activities are carried out on their implementation with the help
of subject information technology.
2. Related Works
Thus, the approaches used in AI and can be summarized to it are widely used in the field of
healthcare [1, 2], robotics and automated systems [3], classification and detective systems [4,5],
IntelITSIS’2021: 2nd International Workshop on Intelligent Information Technologies and Systems of Information Security, March 24–26,
2021, Khmelnytskyi, Ukraine
EMAIL: eduard.em.km@gmail.com (E. Manziuk); alexander.barmak@gmail.com (O. Barmak); krak@univ.kiev.ua (I. Krak);
exechong@gmail.com (O. Mazurets); tkskripnik1970@gmail.com (T. Skrypnyk)
ORCID: 0000-0002-7310-2126 (E. Manziuk); 0000-0003-0739-9678 (O. Barmak); 0000-0002-8043-0785 (I. Krak); 0000-0002-8900-0650
(O. Mazurets); 0000-0002-8531-5348 (T. Skrypnyk)
© 2021 Copyright for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR Workshop Proceedings (CEUR-WS.org)
telecommunication systems [6,7,8], cyber-security [9]. The scope of application is rapidly expanding
and constantly taking new forms [10,11] and is embodied in new approaches [12,13], specific areas of
use [14,15].
At the same time, the level of development of artificial intelligence systems indicates that AI can
make decisions that are unfair, biased, false, and are also unsuspected and misleading. This is not
significant under the conditions of application in tasks that have no signs of increased responsibility,
for example, image improvement, optimization of strategies in games, selection of offers, etc.
However, if decision-making has vital signs for human life and activity, there is a question of
trustworthy AI decisions. This is due to various areas of human life [16-19]. Today technologies are
being developed with the aim of integrating humans into the process of obtaining machine solutions
[20]. In the end, there is a problem of trust in problems and solutions of critical importance. The lack
of resolution of this issue greatly reduces the potential of widespread use. Rapid development of AI
applications requires actualization of efforts towards formality regarding law, ethics, control, security
of AI application [21]. The goal is to ensure the development and implementation of AI systems
through a control system. Lack of control affects trustworthiness and is therefore perceived as a risk
that limits the adaptation and use of such systems. Assessing the advantages and prospects of AI
technologies, AI regulation is carried out both at the national levels of countries and within the
framework of international cooperation [22-24].
3. Domain area concepts
The construction of artificial intelligence (AI) systems should be based on certain principles
according to which it is necessary to ensure the implementation of meeting the needs and according to
which this system is built. It is necessary to determine the value principles within which the project is
implemented. Thus, clear links between abstract principles of value priorities are necessary initial and
basic positions according to which applied concrete solutions will be developed. Formalization of
these norms and principles is implemented by developing standards according to the relevant norms.
The main idea is that compliance with norms and standards is the main framework on which the
design of AI systems is implemented. According to this approach, the principles laid down in the
norms will be the basis on which AI systems are built. This is the basis for the human-centric
development of AI. The widespread use of AI should be due to trust in processes, data, results,
decision robust and security.
As a result of active cooperation and fruitful work, a number of requirements and regulations to AI
are being developed. It is necessary to develop, implement and use AI systems that meet the ethical,
legal, security norms of man and society. That is, there is such a form of AI as human-centric AI. The
transition to the practical field of use forms the conditions and the need to create human-oriented
information technologies. It is based on the provision of fundamental human rights based on
collective, social and constitutional principles, in which individual freedom and respect for human
dignity are both practically possible and meaningful, and does not suggest an overly individualistic
understanding of man [21]. This approach is the basis and should be implemented at the stages of
development, implementation, use and monitoring of AI systems.
The need to develop the basic fundamental foundations of AI systems led to a significant number
of scientific publications and the work of expert groups in order to formally present them. So the EU
Commission's High-Level Expert Group on AI (AI HLEG) European AI Alliance in "Ethics
Guidelines for Trustworthy AI" [21] defined the concept of "trustworthy AI" as the basis for the
widespread use of AI based on a human-centric approach. To ensure trustworthiness, it is necessary to
implement seven key requirements, which are determined on the basis of critical discourses regarding
the content of the concept. Thus, the increase in the social need for the benefits of AI use, due to the
increase in scientific publications and wide discourse, is formalized by the results of the work of
expert groups. Subsequently, implemented in the form of appropriate standards. For example, several
of them: ISO/IEC NP TR 24027 (Information technology - Artificial Intelligence (AI) - Bias in AI
systems and AI aided decision making), ISO/IEC AWI TR 24368 (Information technology - Artificial
intelligence - Overview of ethical and societal concerns), ISO/IEC NP TR 24030 (Information
technology - Artificial Intelligence (AI) - Use cases) and others. One of the basic is
ISO/IEC TR 24028 [25] (Information technology - Artificial Intelligence (AI) - Overview of
trustworthiness in Artificial Intelligence, which aims to analyze the factors that shape and influence
AI systems.
The ISO/IEC TR 24028 standard describes an overview of AI trust issues. The standard aims to
analyze the factors that influence the formation of trustworthy AI and the decisions it generates. The
document briefly presents well-known approaches today that allow increasing confidence in technical
systems and potential uses in AI systems. There are also possible vulnerabilities and approaches that
reduce their impact on AI credibility. The structure of the standard and certainty of concepts make it
possible to form ontology of the concept of trustworthiness, which we will present in a simplified
form. The basis for the creation of ontology is the focus on determining the concepts, ensuring the
implementation of which in practical technical means and information technology using AI, will
ensure the parameter of trustworthy AI solutions.
In this direction, it is necessary to focus efforts on the development of models, methods of
improvement, information technologies and other areas that will give a significant advantage in the
practical application of systems. This will enable and formulate conditions for the use of AI in
responsible and critical areas.
Figure 1: Ontology of trustworthy Artificial Intelligence standard ISO/IEC TR 24028 (simplified
representation)
The central concept of the standard is trustworthy AI, which is defined as the ability of AI to meet
the expectations of stakeholders, is verifiable and has the ability to implement such a check. The
concept of threats, vulnerabilities and challenges is determined, the active influence of which on AI
weakens the trust of stakeholders. This circumstance ultimately calls into question the need to use AI
systems. The concept for preventing and reducing such an impact is determined, because such an
impact is destructive and weakens the concept of trustworthiness and ultimately destroys it. The
purpose of this concept is to reduce the magnitude of threats, vulnerabilities, challenges, and weaken
their impact on trustworthiness. Thus, it is the concept whose structural elements allow forming
trustworthiness. That is, the concept of "Mitigation measures" is a structural element for the formation
of the concept "Trustworthiness of AI". The purpose of the "Mitigation measures" concept is to build
trustworthiness, but it does not determine its content, that is, the basic concepts on which
trustworthiness is based. To determine the content of trustworthiness, the standard establishes the
concept of "High-level concerns". The relevant concept describes at the general level the main
components of the content of trustworthiness. The presented attitudes of concepts that stakeholders
rely on, expect from AI, hope to meet needs and invest in the content of trustworthy AI. At the same
time, the purpose of this concept is a sprawling interpretation, understanding and representations that
are formed on the expectations of stakeholders. The existence of the trustworthiness concept is
conditioned and solely determined by the needs of stakeholders. It should be noted that one of the
important concepts of the system is "Vulnerabilities, threats and challenges". It is this concept that
causes, stimulates and forms the need of stakeholders for the need for trustworthiness. The importance
of the existence of this concept is determined in terms of negative impact and difficulty in meeting
their needs by stakeholders. In order for stakeholders to be able to use AI systems from a position of
trust in them, that is, to meet the requirements of safety, control, sustainability of decisions and others
(is an integral part of the "High-level concerns" concept), development, research and implementation
at the level of practical information systems and technical solutions of the "Mitigation measures"
concept components is necessary. The evolutionary development of this concept is the main factor in
building trustworthiness and its constant increase in the direction of improving the satisfaction of
stakeholder needs.
4. AI trustworthiness ontology model
The model is based on the standard defined by groups of characteristics that together determine the
trust in decisions made using AI approaches. The Model "Information Technologies - Artificial
Intelligence (AI)" (Overview of trustworthiness in Artificial Intelligence) is defined within the
concepts of the relevant domain area. The standard defines high-level concepts that form the concept
of trustworthiness. The set and concept of stakeholders as the main objects need in the concept of
trustworthiness are also determined. Potential vulnerabilities in AI systems and the threats associated
with them are identified in the concepts of vulnerabilities, threats, and challenges. To meet the needs
of stakeholders in trustworthiness, the concept of mitigation measures is determined – possible
controls, recommendations and guidelines that can reduce the impact of known AI vulnerabilities.
Model of the domain area of the standard ISO / IEC TR 24028 "Information Technologies -
Artificial Intelligence (AI) - Overview of trustworthiness in Artificial Intelligence" based on ontology
will be presented in the form.
SbDmTrw = Con, Rel , (1)
where Con - a set of concepts defined within the standard;
Rel - a set of relationships between concepts.
The set of concepts (Fig. 1) is formed on the basis of standard information
Con = {{Vul , Thr , Chal}, {MtMs}, {HLConc}, {Sth}, {TrW }} and the set of relationships between them
Rel = {" reduce", " weaken", " form", " define", " need "}.
Thus, the concept of "Mitigation measures" ( MtMs ) forms the trustworthiness of AI. Accordingly,
the constituent elements of this concept form a set of characteristics and determine the functional
relationship of trustworthiness TrW = f (MtMs ) . Trustworthiness ( TrW ) according to the standard is
a function of the main categories (phrase categories). The structure of “Mitigation measures”
( MtMs ) is as follows
MtMs = {Ct , subCt , Ch}, (2)
where Ch - characteristics, Ct - subcategories, subCt - categories. The elements of the set are
formed in a hierarchy Ch ⊆ subCt ⊆ Ct and are functionally related to the attributes of
trustworthiness Ct , subCt , Ch = f ({atri }in=1 ) . Phrase categories are sometimes defined by a set of
components and the form of representation is defined as follows Ct ≡ {subcti }in=1 or Ct ≡ {ch}in=1 .
Each category is a set of characteristics, as combined by a standard according to semantic meaning,
and forming semantic clusters. There are 10 categories, many of which in the general case can be
represented {⋅, {}
⋅ } , due to the representation of some categories at the level of subcategories or
characteristics. To form uniformity and convenience, we will present the category in the form of a
category of the group that is defined {{}⋅ } . An example of representing categories from the standpoint
of homogeneity: {{ Transparency },{ Reliability, Resilience, Robustness },…}. In this case, the
homogeneity of the structure at the level of categories is formed. The model of the domain area of the
standard is also built on this principle. Given the hierarchical structures of concept formation in the
standard, the most convenient is to obtain a higher hierarchy of concepts, which can be represented as
a homogeneous structure of phrase categories. The analysis showed that within the standard it is the
level of subcategories. List of concept subcategories MtMs : Transparency, Explainability,
Controllability, strategies for reducing Bias, Privacy, Reliability, Resilience, Robustness, mitigating
system hardware Faults, functional Safety, Testing, Evaluation, Use, Applicability. There are many
mitigation measures at the subcategory level MtMs = {mtmsi }in=1 , n = 14 . There n - number of
phrasal subcategories of the concept. The set of relations between the components of the concept
MtMs consists of the relation "isPartOf", ie. Rel MtMs = {"isPartOf "} . Then the concept model
MtMs will take the form
Trs, Exp, Cont , Bsred , Prv, Rlb, Rsl , Rbs, Flmit , (3)
M MtMs = ,
Sffun, Tst , Evl ,Us, Apl , "isPartOf "
where Trs - transparency, Exp - explainability, Cont - controllability, Bsred - strategies for
reducing bias, Prv - privacy, Rlb - reliability, Rsl - resilience, Rbs - robustness, Flmit - mitigating
system hardware faults, Sffun - functional Safety, Tst - testing, Evl - evaluation, Us - use, Apl -
applicability.
The proposed models for the use of standard information are the next step in formalizing the
requirements for AI. This makes it possible to expand and simplify the use of standards in order to
accelerate their implementation in AI systems.
5. Experiment, Results and Discussions
In order to verify the correctness of the chosen approach to determining the form-forming
components of the trustworthiness concept, experimental studies of the degree of conformity of the
proposed structure to the information domain area were carried out. Trustworthiness's concept is
defined in the ethics domain of the human-centric approach. Therefore, for this purpose, we will use
methods of direct conformity and covering the completeness of the use of proximity functions and
interpretation of principles. It is based on research and data obtained in Jobin A. (2019) [27], which
globally analyzes the requirements, technical standards of ethical guidelines AI. On the basis of
content analysis, ethical principles and documents in which they are meaningful are defined. The
value of the selected data is that it consists of documents that are the result of extensive discussion
and consolidated opinion. That is, this is not the opinion of an individual scientist, researcher, and so
on, but is data of a high level of generalization. The corps consists of documents private companies,
governmental agencies respectively, private agencies, academic and research institutions, inter-
governmental or supra-national organizations, non-profit organizations, professional
associations/scientific societies, private sector alliances, research alliances, science foundations,
federations of worker unions, political parties.
The analysis showed that the method of direct comparison coincides at the level of categories
{Transparency}, {Privacy}, which is the relative importance of 16% and 11% in the amount of 27%.
Using the method of partial complete coincidence and meaningful correspondence {Explainability}
has full coincidence and is the content of {Transparency}, by the method of including codes of direct
coincidence - {strategies for reducing Bias} is the code {Justice & fairness}, {functional Safety} is
the code {Non -maleficence}, which is 15% and 14% in relative importance, respectively, and 56% in
the amount of the previous principles. According to the method of full compliance, the partial content
of {Controllability} contains the full meaning and has the phraseological equivalent {Freedom &
autonomy}, which is 8% in relative importance and 64% in the sum with the previous principles. In
terms of partial full match and full coverage of content {Reliability, Resilience, Robustness} in the
subcategory Reliability completely coincides with partial {Trust}, and in the category {Testing,
Evaluation} completely coincides with partial {Justice & fairness} with complete coincidence of
interpretation of content. Accordingly, the relative importance of {Trust} is 6%, in sum with the
previous principles 70%.
The categories {mitigating system hardware Faults} and {Use, Applicability} are represented at
the level of proximity functions of content interpretation. This method also covers the
{Responsibility} principle, which is 14% in relative importance, and 84% in sum with the previous
principles . Principles such as Beneficence, Sustainability, Dignity, Solidarity, the proximity functions
of content interpretation in the standard ISO / IEC TR 24028 are defined at the level according to
which they do not have full substantive compliance and amount to 16%. The experiment showed the
effectiveness of the proposed approach to determine the model of forming the content of the concept
of trustworthiness based on the principles and guidelines for ethical AI. The standard presents the
concepts of trustworthiness formation which are confirmed by the content analysis of the corps. All
the concepts of the ontology of trustworthiness formation have been confirmed at the level of the
main principles of ethics as the global landscape.
100,00% 86,90%
90,00% 80,95%
80,00% 71,43% 71,43%
70,00% 55,95%
60,00% 48,81%
50,00% 40,48%
40,00% 33,33%
30,00% 16,67% 15,48%
20,00% 7,14%
10,00%
0,00%
a)
Dignity Solidarity
Sustainability 3% 1%
3%
Trust
Freedom & 6% Transparency
autonomy 16%
8%
Beneficence Justice & fairness
9% 15%
Non-
Privacy
maleficence
11%
14%
Responsibility
14%
b)
Figure 2: Based on data Jobin A. (2019) [26] distribution of AI ethical principles: a) the share of
documents in which the principle is defined; b) the relative importance of the principle
This suggests that the standard is a relevant reflection of global trends. It should be noted that
certain ethical principles identified in the present documents and regulations, which have a relative
importance of 16% are not presented in the standard. This indicates that the ISO / IEC TR 24028 has a
specialized purpose. The information content revealed by the content analysis at the level of
identification of ethical principles is broader than the purpose of the ISO / IEC TR 24028 standard.
6. Conclusions
The rapid development of the AI information field is reviewed in the work of expert groups and
international standards - International Organization for Standardization (ISO). Concepts, principles,
approaches, scope, concepts of ethics and others are formalized. This is an important step in
organizing, determining in the direction of practical use of exponential growth of the information field
of the AI. The AI-based technical solutions market has gone beyond niche use and has acquired a
value that has a significant impact on a person with prospects for rapid growth. Accordingly, this
requires the development and implementation of technical AI standards. A long term is needed to
develop standards competing with the rapid development of AI. Formalization of all stages, principles
and approaches of AI requires in-depth development of scientific foundations. This, along with the
acceleration of informatization of the whole society, is sometimes competitive. Formalization from
this position is an important direction of systematization of domain field information. The
development of model ontologies based on human-centric principles is an important step in the
development of AI systems for important and critical applications.
It should be noted that the proposed model is developed within the standard and is limited by the
standard itself. As the direction of trust in AI is at the stage of development and standardization,
certain aspects may go beyond the considered standard and are accordingly not presented in the
model. This can probably be considered a shortcoming, and it will be further refined with the
evolution of trustworthiness AI. At the moment, the proposed model can be considered basic.
7. References
[1] S. M. Lee, J. B. Seo, J. Yun, Y. H. Cho, J. Vogel-Claussen, M. L. Schiebler & N. Kim, Deep
learning applications in chest radiography and computed tomography. Journal of thoracic
imaging, 34(2), 2019, pp. 75-85. doi: 10.1097/RTI.0000000000000387.
[2] A. S. Heinsfeld , A. R. Franco, R. C. Craddock, A. Buchweitz, & F. Meneguzzi, Identification of
autism spectrum disorder using deep learning and the ABIDE dataset. NeuroImage: Clinical, 17,
2018, pp. 16-23. doi: 10.1016/j.nicl.2017.08.017.
[3] S. Grigorescu, B. Trasnea, T. Cocias, & G. Macesanu, A survey of deep learning techniques for
autonomous driving. Journal of Field Robotics, 37(3), 2020, pp. 362-386. doi:10.1002/rob.21918
[4] A. Sahba, A. Das, P. Rad, & M. Jamshidi, Image graph production by dense captioning. In 2018
World Automation Congress (WAC), 2018, pp. 1-5. doi: 10.23919/WAC.2018.8430485.
[5] O. V. Barmak, Y. V. Krak, E. A. Manziuk, Characteristics for choice of models in the ansables
classification. CEUR Workshop Proceeding 2139, 2018, pp. 171-179.
https://doi.org/10.15407/pp2018.02.171.
[6] J. Boiko, I. Pyatin, O. Eromenko, & O. Barabash, Methodology for Assessing Synchronization
Conditions in Telecommunication Devices. Advances in Science, Technology and Engineering
Systems Journal, 5(2), 2020, pp. 320-327. doi: 10.25046/aj050242.
[7] B. Zhurakovskyi, J. Boiko, V. Druzhynin, I. Zeniv, & O. Eromenko, Increasing the efficiency of
information transmission in communication channels. Indonesian Journal of Electrical
Engineering and Computer Science, 19(3), 2020, pp. 1306-1315.
http://doi.org/10.11591/ijeecs.v19.i3.pp1306-1315.
[8] O. Nedashkivskiy, Y. Havrylko, B. Zhurakovskyi, & J. Boiko, Mathematical Support for
Automated Design Systems for Passive Optical Networks Based on the β-parametric
Approximation Formula. International Journal of Advanced Trends in Computer Science and
Engineering, 9(5), 2020, pp. 8207-8212. https://doi.org/10.30534/ijatcse/2020/186952020.
[9] J. Li, Cyber security meets artificial intelligence: a survey. Frontiers of Information Technology
& Electronic Engineering, 19(12), 2018, pp. 1462-1474. https://doi.org/10.1631/FITEE.1800573.
[10] I. Krak, O. Barmak, E. Manziuk, Using visual analytics to develop human and machine-centric
models: A review of approaches and proposed information technology, Computational
Intelligence, 2020, pp. 1-26. https://doi.org/10.1111/coin.12289.
[11] I. Krak, O. Barmak, E. Manziuk, & A. Kulias, Data Classification Based on the Features
Reduction and Piecewise Linear Separation. In International Conference on Intelligent
Computing & Optimization. Springer, Cham, 2019, pp. 282-289. https://doi.org/10.1007/978-3-
030-33585-4_28.
[12] R. Damasevicius, J. Toldinas, A. Venckauskas, S. Grigaliunas, & N. Morkevicius, Technical
Threat Intelligence Analytics: What and How to Visualize for Analytic Process. In 2020 24th
International Conference Electronics, 2020, pp. 1-4. doi:
10.1109/IEEECONF49502.2020.9141613.
[13] I. Krak, O. Barmak, E. Manziuk, & H. Kudin, Approach to piecewise-linear classification in a
multi-dimensional space of features based on plane visualization. In International Scientific
Conference “Intellectual Systems of Decision Making and Problem of Computational
Intelligence”, 2019, pp. 35-47. https://doi.org/10.1007/978-3-030-26474-1_3.
[14] Ö. N. Kenger, & E. A. Özceylan, Frequency-Based Approach For Multi-Class Data
Classification Problem. In 2020 International Conference on Electrical, Communication, and
Computer Engineering (ICECCE), 2020, pp. 1-4. doi: 10.1109/ICECCE49384.2020.9179335.
[15] J. Boiko, L. Karpova, O. Eromenko, & Y. Havrylko, Evaluation of phase-frequency instability
when processing complex radar signals. International Journal of Electrical and Computer
Engineering, 10(4), 2020, pp. 4226-4236. doi: 10.11591/ijece.v10i4.pp4226-4236.
[16] T. Neskorodіeva, E. Fedorov, I. Izonin, Forecast Method for Audit Data Analysis by Modified
Liquid State Machine. CEUR-WS vol. 2623, 2020. pp. 25-35.
[17] A. Nicheporuk, O. Savenko, A. Nicheporuk, Y. Nicheporuk, An android malware detection
method based on CNN mixed-data model, CEUR Workshop Proceedings, Vol. 2732, 2020, pp.
198–213
[18] G. Markowsky, O. Savenko, S. Lysenko, A. Nicheporuk, The Technique for Metamorphic
Viruses’ Detection Based on its Obfuscation Features Analysis, CEUR Workshop Proceedings,
Vol. 2104, 2018, pp. 680-687.
[19] S. Lysenko, K. Bobrovnikova & O. Savenko, A botnet detection approach based on the clonal
selection algorithm. In 2018 IEEE 9th International Conference on Dependable Systems,
Services and Technologies (DESSERT). IEEE (2018) 424-428.
[20] A.V. Barmak, Y.V. Krak, E.A. Manziuk, & V.S. Kasianiuk, Information technology separating
hyperplanes synthesis for linear classifiers. Journal of Automation and Information Sciences,
51(5), 2019, pp.54-64. doi: 10.1615/JAutomatInfScien.v51.i5.50.
[21] Ethics Guidelines for Trustworthy AI (AI HLEG, European AI Alliance, European
Commission), 2019. URL: https://ec.europa.eu/futurium/en/node/6945#_ftnref50
[22] Council of Europe: AI Initiatives, 2020. URL:
https://docs.google.com/spreadsheets/d/1mU2brATV_fgd5MRGfT2ASOFepAI1pivwhGm0VCT
22_U/edit#gid=0
[23] EU Commission: Declaration of Cooperation on Artificial Intelligence, 2018. URL:
https://ec.europa.eu/jrc/communities/sites/jrccties/files/2018aideclarationatdigitaldaydocxpdf.pdf
[24] EU Commission Lead DG (Responsible Unit) CNECT: Inception Impact Assessment: Proposal
for a legal act of the European Parliament and the Council laying down requirements for
Artificial Intelligence, 2020. URL: https://ec.europa.eu/info/law/better-regulation.
[25] ISO/IEC TR 24028 Information technology. Artificial intelligence. Overview of trustworthiness
in artificial intelligence (2020).
[26] A. Jobin, M. Ienca, & E. Vayena, The global landscape of AI ethics guidelines. Nature Machine
Intelligence, 1(9), 2019, pp. 389-399. https://doi.org/10.1038/s42256-019-0088-2.