- This paper examines the possibility of extending the principles of ISO/IEC 25000 (SQuaRE) to a quality model for Artificial Intelligence. Some results are described by comparing ISO/IEC 25000 product quality models to unstructured documents and guidelines concerning quality aspects of Artificial intelligence. The analysis shows some possible reuse of the models examined in the context of AI. The emerging important need of a general Governance of AI, including management, processes and organizational aspects, are outside the scope of SQuaRE.
INTRODUCTION
In this paper we examine the results of the analysis
produced in six documents and guidelines concerning AI
recently published by the European Commission, CEN
CENELEC, ETSI, IEEE, the Italian authorities AGID and
MISE [
QUALITY CHARACTERISTICS CONSIDERED
The methods adopted consists in the logic comparison
between SQuaRE quality characteristics and
characteristics mentioned in the documents analysed [
The number of characteristics defined in SQuaRE for each model are:
The number of quality characteristics mentioned in the
documents analysed [
The different numbers (SQuaRE and Aspect of AI) suggest taking more into account the data characteristics and services in future documents concerning AI and in ongoing projects. In fact, during the mentioned conference, the majority of the speakers (in total about 100) pointed out the importance of data management and data quality, particularly within machine learning and specific algorithms. It is also true that not all SQuaRE quality characteristics for software, services and quality in use have been considered in the documents and that new features have been added.
The Conference was born as a cultural and interdisciplinary proposal inspired by a polyphony of knowledge and skills, with the aim of stimulating the broadest possible debate on AI. For these reasons have been involved sociologists, philosophers, educators, psychologists, students, programmers, technicians, managers, exponents of private companies and public institutions.
The topics covered by the speakers (the most part by Italy) have been:
Ethics, communication, culture, education (28%) Technology, quality aspects, security (22%) Governance (17%) Health, therapy, pharmacology application (17%) Legislation, laws, economy, social aspects (16%) The diversity of themes suggests new perspectives of standardization with multiple group and experts.
III.
ELICITATION OF NEW ASPECTS OF QUALITY FOR PRODUCTS AND GOVERNANCE ACTIVITY In the following list are reported the SQuaRE quality characteristics and the aspects of quality detected by experts and organizations mentioned in the Guidelines for software, data, services, quality in use and government.
ISO/IEC 25010 (par. 4.2): - Functional suitability - Performance efficiency - Compatibility - Usability - Reliability (Maturity) - Security (Accountability) - Maintainability (testability) - Portability Guidelines: - Sustainability - Equity - Accountability - Transparency - Reliability - Safety - Robustness - Compliance - Testing - Reliability Aspects in common between ISO/IEC 25010 and Guidelines:
- Trustworthy - Health and well-being - Environment - Economic impacts Aspects in common between ISO/IEC 25000 and Guidelines: - Freedom for risk (Economic, Health,
Environment)
In addition to the quality aspects of the product counted in previous Tables, have been highlighted in the documents analyzed many elements useful for a definition of the activity of Governance, management and processes, issues that are outside the scope of ISO/IEC 25000 SQuaRE series.
Governance and process are out of the scope, except some aspects related to quality management. - evaluation of the impact of algorithms - law and regulations - software engineering processes - human vigilance - data governance - legality - ethics - non-discrimination and fairness - responsibility - human and machine roles - decision making - digital sovereignty Aspects in common between ISO/IEC 25000 and Guidelines: - laws and regulations (limited to data quality characteristic about compliance) The software/systems life cycle processes are the main working area which SC7/WG7 actively is working including integration and acquisition processes. In addition, the related tools and methodologies are run by SC7/WG26 and WG4, thus activities for AI related quality have to be considered by work products developed by other SCs and SC7/WGs including SC42 also.
It may be interesting in the future to extend SQuaRE
quality models to new technologies and to a deepen
relationships and definitions between models by examining
synonyms as well of the new specific quality characteristics
for AI. Works under development [
Using for example the SADT (Structural Analisys and
Design Technique) method [
The components of AI are multiple, sometimes considered independently of each other, sometimes seen as a whole. In many sectors, efforts are being made to incorporate further quality characteristics by experts that could be included in the quality models being prepared by the ISO commissions, particularly by ISO/IEC JTC1 SC42 Working Group 3 with the support of SC7 liaison.
Many organizations are active for a systemic accepted worldwide view of AI (European Commission, International and national Authorities, Industries, Universities and Associations).
Many experts and institutions are seeking to build a foundation of trust for this new technology, aiming at improving IT services in the area of the economy, health and environment.