commbox Artificial Intelligence & Algorithms

Artificial Intelligence & Algorithms

Filter

367

Sort by:

Direction:

Download (97)

Künstliche Intelligenz in politischen Kampagnen: Akzeptanz, Wahrnehmung und Wirkung

Simon Kruschinski, Pablo Jost, Hannah Fecher, Tobias Scherer

Frankfurt am Main: Otto Brenner Stiftung (2025), 108 pp.

"Im ersten Teil geben wir einen Überblick über den Einsatz von generativer KI in der politischen Kampagnenkommunikation. Insbesondere die generativen KI-Tools ChatGPT und Midjourney werden zunehmend in politischen Kampagnen eingesetzt, um menschenähnliche Texte beziehungsweise fotorealistische Bi ... more

Download (63)

The State of Large Language Models for African Languages: Progress and Challenges

Kedir Yassin Hussen, Walelign Tewabe Sewunetie, Abinew Ali Ayele, Sukairaj Hafiz Imam, Eyob Nigussie Alemu, Shamsuddeen Hassan Muhammad, Seid Muhie Yimam

arXiv.org (2025), 25 pp.

"Large Language Models (LLMs) are transforming Natural Language Processing (NLP), but their benefits are largely absent for Africa’s 2,000 low-resource languages. This paper comparatively analyzes African language coverage across six LLMs, eight Small Language Models (SLMs), and six Specialized SL ... more

Download (101)

La inteligencia artificial: Una mirada pastoral desde América Latina y el Caribe

José Sols Lucía (coord.)

Bogotá: Consejo Episcopal Latinoamericano y Caribeño (CELAM) (2025), 97 pp.

"El trabajo incluye desde nociones generales sobre Inteligencia Artificial hasta propuestas prácticas a los obispos del continente. El sumario incluye antecedentes históricos y evolución de la IA, aspectos éticos, antropológicos y teológicos, recoge el magisterio de la Iglesia, asume el desaf� ... more

Download (81)

Just Hiring! How to counter discrimination in AI-powered software for recruitment. A toolkit for software developers

M. Daviet, N. Baranowska, A. Mueller, C. Rus, C. Castillo, J. Saldivar, A. Gatzioura

FINDHR Fairness and Intersectional Non-Discrimination in Human Recommendation (2025), 86 pp.

"In this Toolkit, we offer introductory information and technical recommendations on how to mitigate the risk of discrimination in software development of systems used for recruitment. This Toolkit addresses software developers and product managers working or interested in this field, and it intends ... more

Download (161)

Künstliche Intelligenz im medialen Diskurs: Wird soziale Gerechtigkeit ausgeblendet?

Elke Grittmann, Lina Brink, Peter Kann

Frankfurt am Main: Otto Brenner Stiftung (2025), 84 pp.

[.] Auf Basis von über 2.000 Artikeln aus neun reichweitenstarken Print- und Onlinemedien untersuchen die Autor*innen, welche Themen in der Berichterstattung über KI gesetzt, wie Bezüge zu Fragen sozialer Gerechtigkeit hergestellt werden, welche narrativen Muster sich zeigen und welche Akteur*inn ... more

Download (103)

Artificial Intelligence's Impact on Persons with Disabilites in Africa
Deep Insights

Jerry John Kponyo, Bernard Chiira, Christopher Harrison, Hanyurwimfura Damien (eds.)

Kwame Nkrumah University of Science and Technology (KNUST); Assistive Technologies for Disability Trust (AT4D); Next Step Foundation; Tangaza University College; Kenya National Commission for Science, Technology, and Innovation (NACOSTI) (2025), ix, 169 pp.

"Artificial intelligence has the potential to improve the daily lives of Persons with Disabilities (PWDs) in Africa in practical, cumulative ways—transcribing, translating, summarizing, personalizing, and predicting to reduce routine frictions in school, work, health, mobility, and access to publi ... c services. The study surveys this emerging landscape across Kenya, Ghana, Rwanda, and comparator contexts, mapping how AI augments assistive technologies (AT) and mainstream services while clarifying the policy, data, and delivery conditions that determine real-world impact. Across sectors, we find promising AI use cases beyond traditional AT. In education, adaptive tutors, captioning, and sign-language tools can widen access, though localized content and language coverage remain thin. In employment, AI can support job matching, accessible applications, task breakdown, and on-the-job accommodations, especially when paired with employer guidance and incentives. In mental health, chatbots can extend scarce services and offer stigma-free entry points to care; caregiver tools use voice, translation, and simple workflows to ease coordination and stress. Mobility and independent living benefit from computer vision, speech interfaces, and early robotics, with African innovators piloting sign-language avatars, personalized speech recognition, and lower-cost bionics. The potential of these advances is real but uneven—most remain pilots, and scale requires fit-for-purpose data, infrastructure, and financing. One binding constraint is data. Disability is systematically under-counted in official statistics and under-represented or misrepresented in AI training sets, creating a “disability data desert” that limits accuracy for African contexts (from atypical speech and local sign languages to recognition of mobility aids). Without deliberate collection, curation, and governance that center PWDs, models will continue to miss or misread disabled users, with downstream harms in inclusion, safety, and opportunity. Building representative, ethical datasets—especially for African languages and disabilities—emerges as a first-order priority alongside affordable computing and connectivity. Policy momentum is encouraging but incomplete. The African Union’s Continental AI Strategy (2025–2030) explicitly names disability inclusion, calling for accessible datasets, skills, and governance. Ghana’s and Kenya’s AI strategies also advance infrastructure, data sharing, and public-sector adoption, although explicit disability provisions are still maturing. Convergence around rights-based, inclusive AI is visible; however, its delivery depends on procurement standards, budgeting for accessibility, and participation by Organizations of Persons with Disabilities (OPDs) in the design and oversight process. The study’s central conclusion is pragmatic: AI’s near-term value for disability inclusion is unlocked where three conditions meet. First, inclusion by design—co-creation with PWDs and OPDs, localized languages, and multi-modal accessibility baked into products and datasets. Second, enabling rules and rails—national strategies that mandate accessibility in digital services, robust disability-relevant data, and align incentives for inclusive procurement. Third, sustainable delivery—hybrid human-AI service models, community channels, and financing that lower costs for users and caregivers while supporting innovators to move from pilots to scale. Where these conditions hold, AI can measurably widen access to learning, work, and care; where they do not, tools plateau as proofs-of-concept and risks of bias and exclusion grow. Overall, the path to disability-inclusive AI in Africa is clear: invest in representative data and affordable infrastructure; mandate accessibility and accountability; and deliver through locally grounded, human-centered services. Done together, these steps turn today’s pilots into equitable, reliable systems that expand agency and opportunity for millions of Africans with disabilities." (Executive summary) more

Download (119)

Companion Document: The Guidelines for the Governance of Digital Platforms and Generative Artificial Intelligence

Paris: UNESCO (2025), 34 pp.

"The Guidelines for the Governance of Digital Platforms outline a set of responsibilities, duties, and recommendations relevant to the governance of generative AI. These include fostering an enabling environment for freedom of expression and access to information; establishing independent governance ... more

External Link (231)

AI in Africa: A Landscape Study

Rachel Adams, Fola Adeleke, Leah Junck, Samule Segun, Ayantola Alayande, Selamawit Abdella, Mark Gaffley, Fabio Andres Diaz Pabon, Funbi Salami

Global Center on AI Governance; African Observatory on Responsible AI (2025), 44 pp.

"Though comprehensive AI regulation is still nascent in Africa, there is an increased focus on national AI strategies, ethical guidelines and international collaborations. Over 40 countries have already established data protection and privacy laws, creating opportunities for harmonised regulations a ... nd a supportive environment for AI innovation. Key continental governance instruments and initiatives also include the African Union’s AI Strategy, the African Continental Free Trade Area and the African Commission on Human and Peoples’ Rights. This report found that the increasing application of AI in key sectors across Africa, such as health, education, finance, agriculture, public services and the environment, necessitates a collaborative approach to ethical AI development and use. This requires addressing persistent infrastructural gaps, developing regional regulatory frameworks, investments in AI research and capacity building efforts as well as promoting multi-stakeholder AI governance initiatives. Key strategies identified include joint investment and funding, pooling resources and supporting sustainable and fair open-source projects, which will help direct efforts toward innovation and regional independence. Open Source AI models are particularly important in Africa where local tech startups can leverage global open-source initiatives to build context-specific solutions. However, concerns regarding data ownership, privacy and ethical AI deployment call for more robust policies for open source AI on the continent. To foster responsible AI innovation, African countries should establish comprehensive AI governance frameworks while adhering to existing laws as well as strengthen regulatory institutions to provide oversight over AI technologies. Regulatory sandboxing also offers a viable approach to enable sage experimentation with appropriate oversight. Regional and international cooperation as well as leveraging public-private partnerships can promote funding for expanding AI research and development initiatives, resolve talent shortages and promote infrastructure development. Africa has a unique opportunity to leverage AI, along with its resources and local young talent, for socioeconomic development. However, to ensure inclusivity and sustainable innovation, it is essential to address infrastructure challenges and establish regulatory frameworks that protect societal wellbeing. Embracing collaborative efforts, including open source initiatives, regional and international partnerships are critical for achieving the continent’s ambitious goals without leaving anyone behind." (Executive Summaty, page 3) more

Download (67)

Pathways to Inclusion: Advancing Civil Society’s Role in AI Governance: Evidence from and for the Global Majority

Paola Gálvez Callirgos

Geneva: Globethics Publications (2025), 58 pp.

"This policy brief investigates the role of civil society on AI Governance with particular attention to the Global Majority. The research question explores whether civil society serves as an effective counterbalance to concentrated power in the AI ecosystem to ensure AI systems reflect diverse publi ... more

Download (106)

Artificial Humanities: A Fictional Perspective on Language in AI

Nina Beguš

Ann Arbor: University of Michigan Press (2025), 283 pp.

"Artificial Humanities explores how the humanities-literature, history, and art-can deepen our understanding of artificial intelligence and its development. By examining fictional representations of AI in parallel with actual technological developments, Nina Beguš presents a novel interdisciplinary ... more

External Link (115)

Índice Latinoamericano de Inteligencia Artificial (ILIA) 2025
Deep Insights

Álvaro Soto, Rodrigo Durán, Antonia Moreno, Sebastián Adasme, Sebastián Rovira, Valeria Jordán, Laura Poveda (coord.)

Santiago de Chile: Comisión Económica para América Latina y el Caribe (CEPAL); Centro Nacional de Inteligencia Artificial (2025), 195 pp.

"El Índice Latinoamericano de Inteligencia Artificial (ILIA) constituye un esfuerzo pionero para medir de manera sistemática el avance de la IA en 19 países de la región. El ILIA busca no solo monitorear el grado de preparación, adopción y gobernanza, sino también difundir buenas prácticas, ... además de ser un instrumento para orientar políticas públicas que incorporen la IA al modelo de desarrollo de los países de manera integral. El índice se organiza en tres dimensiones: Factores Habilitantes (infraestructura digital, talento humano y datos), Investigación, Desarrollo y Adopción, y Gobernanza. A partir de más de cien subindicadores, el ILIA 2025 ofrece un panorama detallado sobre avances, brechas y oportunidades estratégicas en el despliegue de la IA en América Latina y el Caribe. [...] Once hallazgos principales del ILIA 2025: i) El despertar de los adoptantes tardíos. Ecuador, Costa Rica, República Dominicana y Guatemala muestran mejoras aceleradas en conectividad, talento y estrategias de IA, reduciendo distancia con los países líderes, y abriendo oportunidades que parecían improbables para una cooperación regional. ii) Código abierto: la oportunidad para América Latina y el Caribe. El alto desempeño de Honduras, El Salvador y Cuba en producción de software abierto confirma que este modelo ofrece una vía estratégica para desarrollar soluciones locales, fomentar la colaboración regional y el aprendizaje colectivo, además de promover la transparencia algorítmica. iii) Muchos datos, pero poca disponibilidad. La región produce grandes volúmenes de información, pero su apertura y estandarización siguen siendo limitadas. Esto restringe la ciencia abierta, reduce la transparencia y frena el desarrollo de soluciones locales de IA. iv) Talento humano: alfabetización, poca especialización. La alfabetización en IA duplica a la formación profesional y cuadruplica al talento especializado. Aunque se amplían programas escolares y posgrados, persiste un embudo en la formación avanzada que limita el potencial de la región para producir soluciones propias. v) Talento e infraestructura para la soberanía digital. Brasil concentra más del 90% de la capacidad de cómputo de alto rendimiento de la región. Aunque Uruguay, Costa Rica y Colombia muestran mejores indicadores per cápita en Unidades de Procesamiento Gráfico (GPU por sus siglas en inglés), más de la mitad de los países carecen de infraestructura crítica, lo que refuerza asimetrías. Además, 13 de los 19 países no generan habilidades tempranas en IA en el currículo educativo y 11 no cuentan con programas de doctorado en IA en sus universidades. Sin fortalecer estas capacidades endógenas, la región corre el riesgo de generar soluciones de IA sesgadas y con baja pertinencia local. vi) IA Generativa: una oportunidad de democratización del acceso. La IA generativa ha democratizado el acceso y acelerado la adopción gracias a sus bajas barreras de entrada y a interfaces simples que permiten a no expertos usar modelos avanzados. Países como Chile, Costa Rica, Perú, Uruguay, Panamá y República Dominicana destacan en uso, y ALC es ya la tercera región del mundo en descargas de aplicaciones de IA Gen (15%–20%). Esta accesibilidad actúa como catalizador y, junto con el desarrollo de software y el código abierto, compensa déficits de cómputo, abriendo oportunidades para toda la región. vii) IA y participación ciudadana: un cambio de paradigma desaprovechado. Aunque la IA podría revolucionar la democracia digital, su uso se limita en la mayoría de los países a chatbots informativos y no en herramientas de consulta, de rendición de cuentas o de co-creación de políticas públicas. Colombia, México y Perú muestran algunos avances, pero ocho países no registran ningún caso de aplicación. viii) Investigación escasa y concentrada. Brasil y México reúnen el 68% de los investigadores activos en IA y, junto a Colombia, Chile y Argentina, concentran el 90% de las publicaciones. Solo siete países participan en conferencias internacionales de élite, con predominio de Chile y Brasil. La escasa presencia en foros académicos de alto impacto reduce la visibilidad y la transferencia de conocimiento de la IA de la región; la actividad académica sigue rezagada y requiere mayor atención en la mayoría de los países. ix) Gobernanza en IA: mucho plan y poca acción. Nueve países cuentan con estrategias nacionales de IA, pero solo unos pocos han asignado presupuesto, definido planes de implementación o establecidos indicadores de impacto. Sin mecanismos de ejecución, estas estrategias corren el riesgo de quedarse en meras declaraciones. Al mismo tiempo, mientras a nivel global avanza la apertura e interoperabilidad de datos, la región mantiene un ritmo lento, centrado en debates normativos y éticos. El verdadero desafío es equilibrar la regulación con políticas orientadas al desarrollo y a la implementación efectiva de la IA, de modo que esta tecnología genere valor sin frenar la innovación. x) Sustentabilidad: un llamado urgente. Solo cuatro países (Brasil, Chile, Colombia y México) muestran industrias de centros de datos robustas, y uno de cada cinco centros cumple con estándares internacionales de sustentabilidad. La mayoría de las políticas de IA y digitalización omite el impacto ambiental, lo que puede entrar en contradicción con los objetivos de sostenibilidad medioambiental de la región. xi) Emprendimiento en IA: una oportunidad para la región. Los 19 países ILIA concentran sólo el 1,12% de la inversión global en IA y cuenta con empresas unicornios en apenas seis países. No obstante, el ILIA muestra que no es necesaria una matriz productiva compleja para fomentar el desarrollo de empresas de IA locales y el uso de esta tecnología en el sector productivo. Entornos favorables de innovación y emprendimiento, inversión y políticas de desarrollo productivo pueden detonar dinámicas innovadoras." (Resumen ejecutivo, páginas 15-16) more

Download (93)

Digital Shadows: Deepfakes Used As Violence Against Women in Journalism and Politics During African Elections

Risper Akinyi, Josephine Miliza, Mwara Gichanga, Sarah Kiden et al.

Kibera: Tanda Community Network (2025), 77 pp.

"This report delves into the growing threat of deepfakes as a tool for technology-facilitated gender-based violence (TfGBV) against women in politics and journalism during elections across Africa. With increasing reliance on digital tools and artificial intelligence (AI), deepfakes pose a serious ri ... sk to democratic processes and the personal security of women in leadership roles. The report investigated how deepfakes and other forms of manipulated media contribute to TfGBV against women journalists and politicians in the context of elections in Ghana, Namibia, and Senegal. Both interviews and focus group discussions elicited responses from policymakers, journalists, technologists, and researchers. Deepfakes, which use AI to manipulate videos, audio, and images, are being weaponized to tarnish reputations, spread falsehoods, and harass women, often in the most public and vulnerable moments of their careers. Women in politics and journalism are particularly susceptible, as their visibility makes them prime targets for digital harassment and character assassination. This misuse of AI-generated content not only impacts their personal and professional lives but also threatens to silence women’s voices in public discourse, deterring other women from engaging in leadership roles. The report emphasizes the urgent need for comprehensive measures to address this issue. Key recommendations include: the enactment of specific deepfake legislation; public awareness campaigns to educate women in leadership about online safety and investment in AI detection technologies; strengthening support systems for victims of TfGBV and promoting responsible AI use are also critical to creating a safer, more inclusive digital environment for women; Collaborative efforts with international organizations and the expansion of media and digital literacy programs are suggested to bolster the nation's resilience against this evolving threat. By addressing the challenges posed by deepfakes and TfGBV, Ghana, Namibia, and Senegal can safeguard its democratic processes and promote gender equality in the political and media landscapes. The report calls for immediate action to protect women in leadership, ensuring that they can participate fully and fearlessly in the country's democratic development." (Executive summary) more

Download (111)

AI With Purpose: How Foundations and Nonprofits are Thinking About and Using Artificial Intelligence

Elisha Smith Arrillaga, Seara Grundhoefer, Christina Im

Center for Effective Philanthropy (CEP) (2025), 39 pp.

"While AI is widely used by nonprofits and foundations alike — most often for internal productivity and communications purposes — leaders share common concerns about AI in their work. They also agree that foundations do not currently have a strong grasp of nonprofits’ AI-related needs. Perhaps ... more

Download (89)

Engines of Order: A Mechanology of Algorithmic Techniques

Bernhard Rieder

Amsterdam: Amsterdam University Press (2025), 351 pp.

"Software has become a key component of contemporary life and algorithms that rank, classify, or recommend are everywhere. Building on the philosophy of Gilbert Simondon and the cultural techniques tradition, this book examines the constructive and cumulative character of software and retraces the h ... more

Download (86)

LAO P.D.R. Artificial Intelligence Readiness Assessment Report

Paris: UNESCO (2025), 72 pp.

"• In the legal context, Lao PDR’s regulatory framework for AI is still nascent. While the government is drafting a National AI Strategy, there are currently no laws addressing core AI issues such as transparency, accountability, or algorithmic redress. Existing frameworks like the Cybercrime La ... w, Data Protection Law and the Digital Government Decree cover general data and computing use but lack AI-specific provisions. AI is referenced in recent constitutional amendments and draft digital transformation policies, but there’s room for improvement in transforming those acknowledgements to clear legal mechanisms to protect citizens and bolster innovation. • For the sociocultural domain, despite progress in gender inclusivity, rural communities in Laos still face limited access to digital technologies due to infrastructure gaps, affordability, and low digital literacy. Successful collaborations with international organizations highlight what’s possible but also stress the need for ongoing investment in local capacity, outreach, and trust-building. Digital initiatives in health and efforts addressing concerns about technology’s environmental impact are evident, but they have not been specifically focused on AI. Digital transformation must also prioritize cultural preservation, including the development of culturally-aware datasets and NLP tools. Currently, most Lao language datasets and NLP tools are lagging participation from native Lao researchers, risking linguistic sovereignty and underrepresenting local dialects, norms, and values. • AI talent and innovation development in Laos needs strong support from its education and research ecosystems. University departments like Computer Science and Software Engineering are small and often siloed, with little crossdisciplinary collaboration. Research output needs some improvement. Promising developments in AI/ICT curricula in K-12 level are welcome, although their effectiveness must be closely monitored. • Laos’s digital economy remains small, constrained by a large informal sector, limited digital infrastructure, and low ICT investment, factors that hinder AI adoption and innovation. While some private companies in sectors such as telecom, banking, logistics, and manufacturing are beginning to use AI tools, and urban populations are gradually adopting app-based lifestyles, broader progress is limited by lagging human capital pipelines, minimal domestic investment, and an underdeveloped startup ecosystem. Foreign-developed platforms still continue to dominate the market, which if continued can limit local developers from developing capabilities and use cases. Aside from that, most Foreign Development Investment (FDI) remains concentrated in resource extraction, and budget shortfalls undermine the sustainability of e-government initiatives. • There is a critical need for a strong commitment to funding both the hard and soft infrastructure that supports AI. This includes enhancing cybersecurity capacity, reliable networks, and data centers and ensuring affordable access to computing infrastructure for AI practitioners and stakeholders." (Executive summary) more

Download (120)

The State of AI in Africa Report

Nairobi: Strathmore University, Centre for Intellectual Property and Information Technology Law (CIPIT) (2025), 67 pp.

"The State of AI in Africa Report 2025 offers a critical reflection on how AI ecosystems across the continent have evolved since 2023, highlighting advances in infrastructure, regulatory frameworks, capacity building, and sectoral innovation, while also surfacing enduring structural disparities. Thi ... s report reveals a continent that is no longer on the margins of AI development but actively reshaping how AI is conceived, deployed, and governed. Kenya, South Africa, Nigeria and Ghana have made notable strides in anchoring AI within national digital strategies, strengthening local innovation ecosystems, and investing in foundational infrastructure such as data centres and broadband connectivity. The growing presence of indigenous datasets, grassroots NLP initiatives, and AI models tailored to African realities reflects a broader shift from consumption to creation. Yet this progress remains uneven. While pockets of excellence are emerging, many regions continue to grapple with limited access to reliable electricity, fragmented regulation, data governance challenges, and critical skill shortages. Central Africa, in particular, still faces infrastructural and institutional constraints that hinder meaningful participation in the continental AI space. This report also documents how AI is being embedded in key sectors; agriculture, health, law, and the creative economy to address uniquely African challenges – from crop failure and disease surveillance to language preservation and cultural expression. These use cases affirm the value of locally informed, contextsensitive AI innovation. On the governance front, the adoption of the AU-AI Continental Strategy and the development of national AI policies signal a growing awareness of the need for harmonised, responsible and inclusive AI development. However, effective implementation remains a central concern, particularly around enforcement mechanisms, ethical guardrails, and regional coordination. This report presents a nuanced picture of AI in Africa: one of ambitious experimentation, uneven capacity, and deep potential. Africa’s future in AI depends not only on technological advancement but on the continent’s ability to embed equity, local relevance, and sustainability at the heart of its AI agenda." (Executive summary) more

External Link (188)

Key Questions for AI in Public Institutions: Prompts and Resources to Spark Critical Conversation
Key Guides

Tanja Dreher, Georgia van Toorn

Sydney: University of New South Wales (UNSW), UARC Centre of Excellence for Automated Decision-Making and Society. (2025), 30 pp.

""In this think piece we sketch the impacts and implications of corporate AI for institutions with a public interest orientation. We focus on public universities and social services as vital public institutions that have been widely understood as essential to democracy and social justice. Through a ... more

External Link (92)

Bias in der künstlichen Intelligenz

Jonas Ditz, Elmar Lichtmeß

Bonn: Bundesamt für Sicherheit in der Informationstechnik (2025), 34 pp.

"Bias ist eine tiefgreifende Problematik für Systeme der Künstlichen Intelligenz, die eine Gefährdung für den sicheren Einsatz von solchen Systemen darstellen kann. Der richtige Umgang mit Bias ist komplex und erfordert eine umfassende Beschäftigung mit der Thematik. Diese Publikation soll Entw ... ickelnden, Anbietenden und Betreibenden von KI-Systemen eine erste Einführung in die Bias-Thematik geben. Bias-Arten weisen eine hohe Diversität auf und können in unterschiedlichen Phasen des Lebenszyklus eines KI-Systems auftreten. Eine Auswahl der relevantesten Bias-Arten sowie eine Zuordnung zur Phase des Lebenszyklus, in der diese auftreten, wird dargestellt. Die Detektion von Bias ist sowohl in Daten als auch in bereits trainierten KI-Modellen möglich. Eine Auswahl an qualitativen und quantitativen Detektionsmöglichkeiten wird präsentiert. Anschließend gibt das Manuskript eine Übersicht über Methoden, die verwendet werden können, um Bias in einem KI-System zu verhindern. Diese sind aufgeteilt in Prä-, In- und Postprozessierungsmethoden, abhängig von dem jeweiligen Zeitpunkt, zu dem sie angewendet werden können. Zum Abschluss wird die Interaktion zwischen Bias und Cybersicherheit diskutiert. Das BSI fordert von Entwickelnden, Anbietenden und Betreibenden von KI-Systemen folgende Handlungen: • Wissen hilft bei der Lösung des Problems. Es muss ein ausreichender Informationsaufbau zur Bias-Thematik stattfinden, um entscheiden zu können, welche Arten für die eigene Anwendung relevant sind. (Kapitel 2) • Zuständigkeiten müssen geklärt sein. Für verwendete Datensätze und KI-Systeme muss eine für Bias zuständige Person benannt werden. Diese Person muss in der Lage sein, potentielle Bias zu identifizieren und geeignete Gegenmaßnahmen vorzuschlagen. • Bekämpfung von Bias beginnt bei den Daten. Nach Möglichkeit müssen organisatorische und technische Maßnahmen bei der Datenerhebung etabliert werden, die potentiellen Bias in den Daten reduzieren. (Kapitel 3) • Ziel muss die Minimierung von unerwünschten Bias in KI-Modellen sein. Prä- und Inprozessierungsmaßnahmen müssen – falls anwendbar – vorrangig verwendet werden. Vor der Anwendung von vortrainierten KI-Modellen (beispielsweise KI-Systeme mit allgemeinem Verwendungszweck) muss die Notwendigkeit für Postprozessierungsmaßnahmen geprüft werden (Kapitel 4) • Bei Bias heißt es am Ball bleiben. Bias-Detektion und -Mitigation muss als Prozess verstanden werden und fester Bestandteil jeder Phase des Produktzyklus eines KI-Modells sein." (Zusammenfassung, Seite 3) more

Download (64)

Antiqua et Nova: An Exposition of its Philosophical Intervention on Artificial Intelligence and its Relevance to Catholic Education

Rhoderick John S. Abellanosa

Kaabigan, volume 3, issue 1 (2025), pp. 16-34

"The essay is an exposition of Antiqua et Nova (AeN), a Church document that provides philosophical, theological, and pastoral reflections and guidelines on the understanding and use of artificial intelligence (AI). It is this work’s position that AeN’s significant contribution is its emphasis o ... more

Artificial Intelligence's Impact on Persons with Disabilites in Africa Deep Insights

Índice Latinoamericano de Inteligencia Artificial (ILIA) 2025 Deep Insights

Key Questions for AI in Public Institutions: Prompts and Resources to Spark Critical Conversation Key Guides

Artificial Intelligence's Impact on Persons with Disabilites in Africa
Deep Insights

Índice Latinoamericano de Inteligencia Artificial (ILIA) 2025
Deep Insights

Key Questions for AI in Public Institutions: Prompts and Resources to Spark Critical Conversation
Key Guides