AI and ethics最新文献

{"title":"Incorporating emergence in data-driven algorithms: the circularity pathway","authors":"Deepak P.,&nbsp;Adwait P. Parsodkar,&nbsp;Vishnu S. Nair,&nbsp;Sutanu Chakraborti","doi":"10.1007/s43681-025-00763-z","DOIUrl":"10.1007/s43681-025-00763-z","url":null,"abstract":"<div><p>Today’s AI algorithms often use a data-first approach, where available data predicates the development of algorithms, following which the algorithms are evaluated using quantitative metrics. This institutes a lack of attention to sociomaterial aspects and broader contexts, ones that are highly relevant in times when AI is used across a variety of socially relevant sectors. Contemporary AI design practices, often characterised as ground-truthing, has entrenched a social science deficit, where qualitative aspects are largely ignored. In this commentary, we consider the subset of tasks involving data-driven estimation of emergent properties, properties which intrinsically emerge through processes underpinned by relationalities between objects. We posit that attention to emergence relationalities would enhance algorithmic estimation of emergent properties, while also making them more ethically aligned due to attempting to mirror actual phenomena. Yet, where do we start to operationalize this? We observe that modelling circularities within data-driven algorithms has led to hugely successful algorithms. It is also the case that circular formulations are underpinned by relationalities between objects. Consequently, we propose that circularity offers a substantive pathway to embed emergence relationalities within algorithms. We illustrate the new qualitative analysis capabilities we acquire through viewing algorithmic circularity through the prism of emergence relationalities. These include thought-frameworks to aid designing algorithms, and capabilities towards critiquing extant algorithms. Throughout the paper, we use popular data-driven estimation tasks to anchor the narrative.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"5 5","pages":"4613 - 4622"},"PeriodicalIF":0.0,"publicationDate":"2025-06-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145121577","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Decoding the cry for help: AI's emerging role in suicide risk assessment","authors":"Pouyan Esmaeilzadeh","doi":"10.1007/s43681-025-00758-w","DOIUrl":"10.1007/s43681-025-00758-w","url":null,"abstract":"<div><p>Artificial Intelligence (AI) has shown significant potential in identifying early warning signs of suicide, a critical global health issue claiming nearly 800,000 lives annually. This study examines how AI technologies—with a primary focus on conversational agents (chatbots), Natural Language Processing (NLP), deep learning, and Large Language Models (LLMs)—can enhance early detection of suicide risk through linguistic pattern analysis and multimodal approaches. Traditional suicide risk assessment methods often lack timely intervention capabilities due to limitations in scalability and continuous monitoring. We synthesize current research on AI-driven approaches for suicide risk detection, specifically examining (1) how NLP and deep learning techniques identify subtle linguistic patterns associated with suicidal ideation, (2) the emerging capabilities of LLMs in powering more contextually aware chatbot interactions, (3) ethical frameworks necessary for responsible clinical implementation, and (4) regulatory frameworks for suicide prevention chatbots. Our analysis reveals that AI-powered chatbots demonstrate improved accuracy in detecting suicidal ideation while providing scalable, accessible support. Additionally, we offer a comparative analysis of leading AI chatbots for mental health support, examining their therapeutic approaches, technical architectures, and clinical evidence to highlight current best practices in the field. We also present a novel framework for evaluating chatbot effectiveness in suicide prevention that offers standardized metrics across five key dimensions: clinical risk detection, user engagement, intervention delivery, safety monitoring, and implementation success. While AI chatbots provide significant potential to transform early intervention, substantial challenges remain in addressing conversation design, ensuring appropriate escalation protocols, and integrating these tools into clinical workflows. This paper examines the most promising chatbot approaches for suicide prevention while establishing concrete benchmarks for responsible implementation in clinical settings.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"5 5","pages":"4645 - 4679"},"PeriodicalIF":0.0,"publicationDate":"2025-06-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145121536","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Eu regulatory ecosystem for ethical AI","authors":"Vaios Bolgouras,&nbsp;Apostolis Zarras,&nbsp;Christian Leka,&nbsp;Ioannis Stylianou,&nbsp;Aristeidis Farao,&nbsp;Christos Xenakis","doi":"10.1007/s43681-025-00749-x","DOIUrl":"10.1007/s43681-025-00749-x","url":null,"abstract":"<div><p>AI applications raise complex ethical, legal, and security challenges that demand comprehensive and coordinated governance at multiple levels. In this paper, we examine how key European Union (EU) regulatory frameworks, such as the AI Act, GDPR, and NIS2, interact to set standards for AI security, functionality, and ethical performance. By comparing the objectives and requirements outlined in these regulatory instruments, we identify points of convergence that encourage a holistic approach to creating transparent, accountable, and fair AI systems. This alignment is pivotal for building user trust and mitigating risks associated with data breaches, algorithmic bias, and privacy violations. Moreover, we explore how harmonizing these regulations can maintain the EU’s competitive edge in AI innovation, as clear governance structures help businesses remain agile while protecting consumer interests. Our analysis further addresses the ramifications for global AI governance, emphasizing the significance of a unified, forward-looking strategy to ensure responsible AI development. In doing so, we recommend future harmonization initiatives that promote societal well-being, safeguard human rights, and uphold ethical and technological standards worldwide.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"5 5","pages":"5063 - 5080"},"PeriodicalIF":0.0,"publicationDate":"2025-06-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s43681-025-00749-x.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145121397","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Non-discrimination law, the GDPR, the AI act and the - now withdrawn - AI liability directive proposal offering gateways to pre-trial knowledge of algorithmic discrimination","authors":"Ljupcho Grozdanovski","doi":"10.1007/s43681-025-00754-0","DOIUrl":"10.1007/s43681-025-00754-0","url":null,"abstract":"<p>This article focuses on the evidence necessary to support claims of discrimination arising from AI-assisted recruitment. It addresses two main issues. First, given that discrimination may be subtly expressed by (possibly opaque) AI systems, this article examines the EU legal frameworks designed to facilitate access to explanations and evidence capable of revealing discriminatory bias in automated recruitment processes. Those provisions include the Equality Directives, the GDPR, the AI Act (AIA), and the now-withdrawn AI Liability Directive (AILD) proposal. In analysing those provisions, particular attention is paid to the types of information that may be sought: the logic behind an AI’s output, the reasons a human decision-maker relied on that output, and the AI system’s compliance with the AIA. Second, the article determines which among the various applicable provisions should be treated as <i>lex specialis</i>, that is, the specific rule that should be preferentially applied to obtain pre-trial knowledge of algorithmic discrimination. In this context, special emphasis is placed on Articles 22 GDPR and 86 AIA, both of which recognize a right to an explanation and are potentially applicable to automated recruitment systems, since those can be classified as both high-risk under Annex III of the AIA and involving personal data processing, under the GDPR. From the standpoint of a litigant’s ability to satisfy the procedural requirements of both provisions, the article argues that Article 86 AIA may offer a more accessible pathway than Article 22 GDPR, both in terms of the scope of information provided and the conditions required for access. Nonetheless, neither provision guarantees automatic disclosure; access remains conditional and often subject to stringent procedural requirements. This selective, rather than automatic approach to transparency raises important questions about its implications for fundamental rights, particularly the right to access justice and effective remedies.</p>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"5 5","pages":"5039 - 5062"},"PeriodicalIF":0.0,"publicationDate":"2025-05-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145122538","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"From decision aiding to the massive use of algorithms: where does the responsibility stand?","authors":"Odile Bellenguez,&nbsp;Nadia Brauner,&nbsp;Alexis Tsoukiàs","doi":"10.1007/s43681-025-00753-1","DOIUrl":"10.1007/s43681-025-00753-1","url":null,"abstract":"<div><p>In the very large debates on ethics of algorithms, this paper proposes an analysis on human responsibility. On the one hand, algorithms are designed by some humans, who bear a part of responsibility in the results and unexpected impacts. Nevertheless, we show how the fact they cannot embrace the full situations of use and consequences leads to an unreachable limit. On the other hand, using technology is never free of responsibility, even if there also exist limits to characterise. Massive uses by unprofessional users introduce additional questions that modify the possibilities to be ethically responsible. The article is structured in such a way as to show how the limits have gradually evolved, leaving unthought of issues and a failure to share responsibility.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"5 5","pages":"5025 - 5037"},"PeriodicalIF":0.0,"publicationDate":"2025-05-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145122537","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Effects of ethics, regulations, and training regarding responsible AI use on continued use intentions of gen AI chatbots: structural equation model of university students perceptions","authors":"Gyasi Alfred Bannor,&nbsp;Francis Ohene Boateng,&nbsp;Yarhands Dissou Arthur","doi":"10.1007/s43681-025-00755-z","DOIUrl":"10.1007/s43681-025-00755-z","url":null,"abstract":"<div><p>Generative artificial intelligence (Gen AI) tools, such as chatbots, are increasingly used in higher education (HE), providing opportunities for enhanced learning while raising ethical and regulatory concerns. Drawing on Technology Acceptance Model (TAM), this study examines the influence of ethics, regulations, and training on university students’ intentions to continue using Gen AI chatbots in Ghana, addressing a critical gap in understanding the factors that affect continuous use intentions. Using online survey data collected from a convenience sample of 239 undergraduate students across two universities, structural equation modeling (SEM) was applied to evaluate the relationships between the variables. The findings reveal that ethics have a significant positive influence on students continued use intentions, highlighting the importance of ethical guidelines in fostering trust and promoting responsible use. In contrast, regulations and training had insignificant effects, suggesting potential misalignment between policy frameworks, training content, and students’ practical needs. These results emphasize that while ethical considerations are pivotal, regulatory measures must balance flexibility and enforcement, and training programs must be tailored to address specific challenges in AI use. This study contributes to the growing body of literature on AI in education, offering actionable insights for institutions to design effective strategies that ensure academic integrity and sustainable adoption of Gen AI tools in higher education. </p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"5 5","pages":"5011 - 5023"},"PeriodicalIF":0.0,"publicationDate":"2025-05-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145122524","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Who is a scientist? Gender and racial biases in google vision AI","authors":"Ehsan Mohammadi,&nbsp;Yizhou Cai,&nbsp;Alamir Novin,&nbsp;Valerie Vera,&nbsp;Ehsan Soltanmohammadi","doi":"10.1007/s43681-025-00742-4","DOIUrl":"10.1007/s43681-025-00742-4","url":null,"abstract":"<div><p>With the prevalence of artificial intelligence (AI) in everyday life, there is a need to study the biases of AI. Specifically, understanding the biases of AI in computer vision is important due to visual content's role in creating classes and categories that can shape people’s perspectives. Without supervision, such classifications can lead to gradual and intangible negative impacts of AI discrimination in the real world. Demographics at the intersection of gender and racial biases may experience unforeseen multiplier effects due to how AI compounds big data without accounting for implicit biases. To quantitatively verify this multiplier effect of biases, this study first examines the gender and racial biases in Google Cloud Vision AI, a leading application with a high level of adoption and usage in different sectors worldwide. Statistical analysis of 1600 diverse images of scientists reveals that Google Cloud Vision AI has implicit gender and racial biases in identifying scientists in image processing. Particularly, the findings show that, in this sample, Black and Hispanic individuals were represented less compared to White and Asian individuals as scientists. Google Cloud Vision AI also significantly underrepresented women as scientists compared to men. Finally, the results indicate that biases at the <i>intersection</i> of race and gender are exponentially worse, with women of color being least represented in images of scientists by Google Vision. Given the ubiquity and impact of AI applications, addressing the complexity of social issues such as equitable integration and algorithmic fairness is essential to maintaining public trust in AI.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"5 5","pages":"4993 - 5010"},"PeriodicalIF":0.0,"publicationDate":"2025-05-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s43681-025-00742-4.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145122444","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Implementing AI ethics: the VPCIO model","authors":"Désirée Martin,&nbsp;Michael W. Schmidt,&nbsp;Rafaela Hillerbrand","doi":"10.1007/s43681-025-00723-7","DOIUrl":"10.1007/s43681-025-00723-7","url":null,"abstract":"<div><p>Due to the development and use of artificial intelligence (AI) systems, there is a need for normative guidance on AI technology. Building on reasonably shared and systematized ethical values and principles (Martin et al. in Comparing AI ethics and AI regulation: ethical values and principles and the case of well-being, beneficence and sustainability, In: Müller, Dung, Dewey, Löhr (Eds.) Philosophy of artificial intelligence: the state of art, synthese library, Springer, Berlin, forthcoming), we aim to provide a framework for implementing ethics in AI systems. The research question in this paper is how to transfer values and principles to an AI system in a way that is understandable and evaluable for users, stakeholders, or an oversight body. Therefore, we work out how to translate values and principles into more concrete norms that can be implemented by the developer and monitored by the executive. Based on our systematization, we extend the so-called VCIO model, where VCIO stands for values, criteria, indicators and observables, as presented by Hallensleben et al. (From principles to practice—an interdisciplinary framework to operationalise AI ethics. VDE, Bertelsmann Stiftung, Frankfurt a. M./Gütersloh. https://www.ai-ethics-impact.org, 2020). Our contribution includes modifications to the model and, most importantly, the addition of principles. Building on this methodology, we present a model that is highly acceptable, the VPCIO model. We developed and evaluated the VPCIO for two case studies. The main case study is an AI-assisted robot used for reconnaissance of radiological hazards (based on a BMBF funded project, entitled KIARA (https://www.itas.kit.edu/english/projects_hill22_kiara.php)). The second case study is about an AI system in an entertaining context, namely to swap faces. Implementing the ethical aspects in these cases into the VPCIO model results in an indicator system that illustrates how ethical aspects can be transferred to an AI system in an understandable way.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"5 5","pages":"4975 - 4992"},"PeriodicalIF":0.0,"publicationDate":"2025-05-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s43681-025-00723-7.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145122532","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"The cyborg dilemma: body, identity, and bioethical conflicts in Mary. E. Pearson’s The Adoration of Jenna Fox","authors":"Anwesha Adhikary","doi":"10.1007/s43681-025-00722-8","DOIUrl":"10.1007/s43681-025-00722-8","url":null,"abstract":"<div><p>AI capacitated the introduction of a different world of power dynamics, especially in healthcare sectors. The diseased body, which is devoid of any kind of agency, is an almost similar kind of body which Michel Foucault terms as “Docile Body”; and has been a site of direct control and subjugation in terms of the rigid singular decision made only by doctors and caregivers concerning the patient’s well-being. With the growing colonization of AI in the medical field, this “body without agency”—the sufferer—becomes even more vulnerable to inhumane and often indecisive processes of recovery. Despite the apparent gift of new life, these interventions risk reducing the essence of living to mere computational monitoring of one’s mind. This paper, therefore, explores the issue of inhumane surveillance by the apparent “powerful”, as in the case of Jenna Fox’s body and mind, and delves into the bioethical concerns that arise after she transforms into a “Cyborg” in Mary. E. Pearson’s novel <i>The Adoration of Jenna Fox</i> (2008). Questioning the mechanics of power imposed by AI and its users on Jenna Fox’s “docile body” (after the accident), this paper attempts to locate the dilemma between machine consciousness and human consciousness showing how Jenna encounters a split—rather than a reconciliation—between these two selves, through a series of existential questions she puts across the course of the novel.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"5 5","pages":"4967 - 4974"},"PeriodicalIF":0.0,"publicationDate":"2025-05-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145122237","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"RogueGPT: transforming ChatGPT-4 into a rogue AI with dis-ethical tuning","authors":"Alessio Buscemi,&nbsp;Daniele Proverbio","doi":"10.1007/s43681-025-00750-4","DOIUrl":"10.1007/s43681-025-00750-4","url":null,"abstract":"<div><p>The ethical implications and potentials for misuse of Generative Artificial Intelligence are increasingly worrying topics. This paper explores how easily the default ethical guardrails of ChatGPT, using its latest customization features, can be bypassed by simple prompts and fine-tuning, that can be effortlessly accessed by the broad public. This malevolently altered version of ChatGPT, nicknamed “RogueGPT”, responded with worrying behaviours, beyond those triggered by jailbreak prompts. We conduct an empirical study of RogueGPT responses, assessing its flexibility in answering questions pertaining to what should be disallowed usage. Our findings raise significant concerns about the model’s knowledge about topics like illegal drug production, torture methods and terrorism. The ease of driving ChatGPT astray, coupled with its global accessibility, highlights severe issues regarding the data quality used for training the foundational model and the implementation of ethical safeguards. We thus underline the responsibilities and dangers of user-driven modifications, and the broader effects that these may have on the design of safeguarding and ethical modules implemented by AI programmers. Disclaimer. This paper contains examples of harmful language. Reader discretion is recommended.</p></div>","PeriodicalId":72137,"journal":{"name":"AI and ethics","volume":"5 5","pages":"4945 - 4966"},"PeriodicalIF":0.0,"publicationDate":"2025-05-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s43681-025-00750-4.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145122221","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}