Jun 2019· arXiv.org· Vol abs/1906.12307· 2 citations· 23 references
Computer Science
TL;DR
This paper provides a baseline for ethics in AI based software development by reporting results from an industrial multiple case study on AI systems development in the health care sector, and explores the current state of practice out on the field in the absence of formal methods and tools for ethically aligned design.
Abstract
Solutions in artificial intelligence (AI) are becoming increasingly widespread in system development endeavors. As the AI systems affect various stakeholders due to their unique nature, the growing influence of these systems calls for eth-ical considerations. Academic discussion and practical examples of autonomous system failures have highlighted the need for implementing ethics in software development. However, research on methods and tools for implementing ethics into AI system design and development in practice is still lacking. This paper be-gins to address this focal problem by providing a baseline for ethics in AI based software development. This is achieved by reporting results from an industrial multiple case study on AI systems development in the health care sector. In the context of this study, ethics were perceived as interplay of transparency, re-sponsibility and accountability, upon which research model is outlined. Through these cases, we explore the current state of practice out on the field in the ab-sence of formal methods and tools for ethically aligned design. Based on our data, we discuss the current state of practice and outline existing good practic-es, as well as suggest future research directions in the area.
A comprehensive model for integrating ethical standards into the phases of the Software Development Life Cycle (SDLC) is proposed, founded on the pillars of fairness, transparency, accountability, and sustainability, offering practical recommendations aimed at developers, organizations, and policymakers.
A. Alaswad· Al-Farooq Journal of Science...· 0 citations
Artificial Intelligence (AI) is rapidly entering scientific research programmes on the assumption that the established ethical frameworks are enough to address failure risks. But that assumption is wrong: AI systems are structurally different from the tools that conventional research ethics was designed to govern. As illustrated by mainstream models, AI systems can be characterized by being stochastic, opaque, resource-intensive, and operate at population scale, meaning that ethical breaches can extend far beyond individual research projects, with potential consequences for public trust in democratic institutions. Taking the case of EU research as reference, this paper proposes the ARCHON framework,a set of ten requirements derived from the technical architecture of AI systems, to facilitate ethical compliance of scientific research for experts and practicioners engaging in roles such as proposal evaluators, project officers, ethics board members, or research administrators. The paper includes a mapping table intended to facilitate the application of the framework to evaluate ethical compliance, both for specific projects and general research programmes. Applied consistently, the ARCHON framework enables substantive rather than merely declarative AI ethical compliance, contributing this way for scientific research policy to continue addressing societal needs while remaining democratically accountable.
Jesus Manuel Benitez Baleato· Open Research Europe· 0 citations
As Artificial Intelligence (AI) is incorporated into the workflow, several ethical issues and risks which can cause actual harms are being introduced. Despite the well-established frameworks of ethical AI principles, there are few practice-oriented approaches for structured interdisciplinary assessment of the ethical aspects associated with AI use and design. To supply the principles and address the increasing focus on AI-risks from an ethical perspective, an Ethical risk assessmeNt of Ai iN pracTice (ENACT) methodology is proposed. To develop ENACT together with a cross-sectoral, interdisciplinary consortium of Norwegian private and public businesses, the core principles of Design-Based Research (DBR) were applied including real context orientation, collaborative partnership and focus on testing and multiple interactions. Four aspects of the ENACT methodology were collaboratively developed and tested and are proposed in this paper including format, structure, scope and support tools. This paper describes and details three steps of the ENACT methodology and discusses its potential and limitations for qualitative ethical risk assessment of AI in organisational settings.
N. Murashova, Leonora Onarheim Bergsjø, Heidi Dahl et al.· Nordic Machine Intelligence· 0 citations
Abstract This article presents the AI Ethics Canvas: a graphical tool to guide public administrations in mainstreaming ethics in the development, deployment, and monitoring of artificial intelligence (AI) tools for their own everyday use. Our canvas is the result of an action-research effort conducted within the scope of a project involving the Autonomous Province of Trento (PAT), Italy. We initially developed it based on existing ethical and legal frameworks. Then, we conducted qualitative interviews with PAT functionaries and technologists developing software on behalf of the PAT to ensure the canvas’s content was clear and applicable to public administration work. While originally elaborated for the PAT, the canvas presented is applicable to all public administrations within the European Union who wish to adopt AI in their operations, as it references to core European legislation such as General Data Protection Regulation, AI Act, Data Act, and Data Governance Act. The article presents the AI Ethics Canvas both as a final product and in its creation process to enhance transparency in AI ethics tool development.
Riccardo Nanni, P. G. Bizzaro, Albana Celepija et al.· Data & Policy· 0 citations
Artificial intelligence (AI) is increasingly explored as a tool to support institutional decision-making, including within research ethics committees (RECs). However, empirical evidence remains limited regarding how REC members define the acceptable scope of AI integration in ethics review. This study examines how members of Spanish Research Ethics Committees (CEI/CEIm) conceptualise AI use within institutional oversight. A national cross-sectional survey was distributed to 202 committee members and technical secretariat staff, yielding 63 responses (31.2%). The questionnaire combined closed-ended and open-ended items addressing perceived usefulness, acceptable levels of automation, governance concerns, and implementation barriers. Quantitative data were analysed descriptively with exploratory chi-square tests, and qualitative responses were examined using inductive thematic analysis. Respondents expressed strong interest in AI for administrative and documentary support but consistently rejected fully automated ethical decision-making. A statistically significant association was observed between annual workload and interest in AI implementation (χ
2
= 6.20,
p
= 0.045), with higher-workload committees reporting greater openness. Years of experience and institutional role were not significantly associated with attitudes. Across responses, continuous human oversight, accountability, transparency, and data protection emerged as central conditions for acceptability. The findings indicate that AI integration in research ethics oversight is shaped not only by technological feasibility but by institutional concerns regarding responsibility, legitimacy, and deliberative authority. The study contributes empirical insight into how oversight institutions negotiate the acceptable boundaries of AI-assisted decision support in ethically sensitive governance contexts.
Daniel Wang, Marta Guix Arnau, Cristina Llop Julià et al.· Research Ethics· 0 citations
Ethical artificial intelligence (AI) is emerging as a widespread approach that refers to the development and use of AI technologies in a manner compatible with human values. However, the tendency of AI companies in Europe to relocate their operations to the United States is increasing debates about the applicability of this approach. One of the main reasons for this is the European Union's General Data Protection Regulation (GDPR); the costs, controls, and restrictions imposed by this regulation on data use may lead global technology firms to move outside of Europe. The high economic value-generating potential of AI creates pressure for the relaxation of these regulations, presenting a choice between ethical principles and economic interests. This study aims to provide a conceptual evaluation within the framework of the question of whether the economic gains provided by AI can outweigh human-centered ethical values. Accordingly, the research adopts a qualitative approach based on the literature and examines the current debates at a conceptual level. The assessments indicate that the European Union's desire to compete with the United States and China in the global artificial intelligence race creates a context that may make prioritizing ethical principles difficult. In this context, it is considered that the development of an ethical understanding of artificial intelligence cannot be considered independently of economic and political dynamics, and while it presents various limitations under current conditions, it offers a structure that can be improved with appropriate governance mechanisms.
Tolgay Ercan· Düzce Üniversitesi Sosyal Bi...· 0 citations
Computer-use AI agents struggle with multi-step workflows like email and customer support. Echoverse trains agents in realistic environments rather than simply providing more training tasks, helping them improve as the tasks, tests, and environments evolve. The post Echoverse: Deep, evolving environments for computer-use agents appeared first on Microsoft Research.
MIT News · Artificial Intelligence· news.mit.eduJul 14, 2026
Through research and entrepreneurship, Professor Devavrat Shah is helping to design methods that can handle constant decision-making using limited computational resources.
MIT News · Artificial Intelligence· news.mit.eduJun 3, 2026
The new ChartNet training dataset could improve the accuracy of vision-language models that help analyze business trends or interpret scientific figures.
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
