AI: An Annotated Bibliography
Understanding AI’s Limits, Biases and Design
Preface
Below is an annotated bibliography I created for an MLIS class on AI and ethics. Fifteen published articles are summarized and briefly reviewed. This project examined current scholarship on artificial intelligence, ethics and the evolving role of information professionals. The works included span conceptual analyses, policy discussions, literature reviews and case studies.
Some studies emphasized transparency, accountability and governance; others explored practical applications in libraries, education and authorship. Taken together, they reflect a field in rapid transition—one that raises technical, ethical and institutional questions that remain open for debate.
Almeida, D., Shmarko, K., & Lomas, E. (2022). The ethics of facial recognition technologies, surveillance, and accountability in an age of artificial intelligence: A comparative analysis of US, EU, and UK regulatory frameworks. AI and Ethics, 2, 377–387. https://doi.org/10.1007/s43681-021-00077-w
Almeida et al. (2022) analyzes how facial recognition technology (FRT) is used by law enforcement in the United States (U.S.), the United Kingdom (U.K.), and the European Union (E.U.). The authors highlight the challenges of ethics and regulations that surround this technology while pointing out the need to balance public safety with human rights, specifically privacy. The paper compares and contrasts the laws and policies of the above mentioned countries and discusses the implications for accountability, transparency, and human rights.
In the U.S., FRT is advancing quickly. Yet, the laws remain inconsistent, often changing between states, and there is no overarching data privacy protection. By contrast, the U.K. and E.U. have in place policies such as the General Data Protection Regulation (GDPR) and the principle of Privacy by Design (PbD). However, there is still no global standard or unified human rights policy governing the use of FRT.
Almeida et al. argue for consistent global standards to ensure that FRT is used responsibly. Establishing such standards, they say, will help guide future policy, strengthen accountability, and better protect human rights.
The authors rightly observe that “the societal impacts of FRT misconduct are not to be underestimated” (p. 386). However, they could have been made better. Also, while the authors emphasize the importance of transparency, they offer little help on how it might be achieved. Moreover, the effectiveness of global regulatory bodies is doubtful, as illustrated by the United Nations’ ongoing challenges in enforcing meaningful regulations.
Bradley, F. (2022). Representation of libraries in artificial intelligence regulations and implications for ethics and practice. Journal of the Australian Library and Information Association, 71(3), 189–200. https://doi.org/10.1080/24750158.2022.2101911
Bradley (2022) investigates the ways in which libraries are represented in AI regulations and ethics, while identifying the opportunities for the library profession to contribute and thus ensure the establishment of trustworthy and ethical AI policies. Her paper is a literature and policy review, analyzing the existing national and international regulations of AI, ethics, and library initiatives.
According to Bradley, libraries are engaged in the development of some elements of AI, but their voice is not well represented in formal policies. She states that libraries have the potential to create ethical AI standards, especially in the protection of human rights and transparency.
Her findings indicate that, given their expertise in information rights and ethics, libraries should have a proactive role to play in AI regulation. Doing so will help ensure that future AI systems have values like intellectual freedom, privacy, and access.
Bradley provides a comprehensive review of AI policy and librarian views on the proactive role libraries could play in the development of ethical AI. Particularly, Bradley demonstrates an intersection of values between those of libraries and AI. However, being a review only it lacks the empirical data or case studies that might have been used to show exactly how libraries influence AI regulation in practice. This somewhat limits its practical use.
Cox, A. (2022). The ethics of AI for information professionals: Eight scenarios. Journal of the Australian Library and Information Association, 71(3), 201–214. https://doi.org/10.1080/24750158.2022.2084885
Cox (2022) reviews the literature regarding ethical issues surrounding the use of artificial intelligence by information professionals. He then provides eight real-life scenarios to illustrate the issues in a “concrete” (p. 201) form.
The scenarios Cox provides were developed by an in-depth review of the literature as well as contextual considerations and feedback from experts. The scenarios are intended to stimulate debate and reflection within the profession. Cox writes that these scenarios are tools to help information professionals navigate the complex ethics of AI. He emphasizes that said ethical challenges are multifaceted and context-dependent.
The strength of Cox’s paper is in the concreteness of the scenarios. He successfully contextualizes complex ethical ideas into a format that is understandable. His scenarios, therefore, are useful for training, policy development, and ongoing debates. They may also provide guidance for future research on the ethical integration of AI in the information industry.
One flaw, however, is the inability to create a scenario for every possible ethical challenge that may arise. Surely, Cox missed something. Empirical testing would also be beneficial. Still, the study’s emphasis on real-world, relatable dilemmas is valuable.
Cox, A. M., & Mazumdar, S. (2022). Defining artificial intelligence for librarians. Journal of Documentation, 78(8), 402–417. https://doi.org/10.1108/JD-12-2021-0251
Cox and Mazumdar (2022) explore how to define artificial intelligence from a librarian’s perspective. They do this by examining definitions of AI, analyzing the many types of technology that constitute AI, their potential uses in library settings, and the implications of AI on the profession. The paper is a conceptual piece that closely reviews the existing literature with a focus on readers interested in AI as a strategic priority rather than mere technology.
They found that nearly every formal definition of AI included the ability to perform tasks humans normally do. However, those definitions were fairly vague and did not mention specific technologies. Cox and Mazumdar conclude that AI is an evolving idea with significant cultural meaning.
The primary technology used in AI is machine learning, including the ability to train it, either supervised or unsupervised. Useful embodiments include virtual assistants (e.g., Amazon Alexa) and robots, both of which can interact with patrons or support library operations.
The paper identified five uses for AI in library operations: library back-end processes, library customer service, the creation of communities of data scientists, data and AI literacy, and patron behavioral management.
The paper concludes with a reflection on AI’s broader impact on the information science profession. While Cox and Mazumdar rightly note the potential for both disruption and improvement, their analysis shifts toward a sociopolitical critique that detracts from their discussion. The authors claim AI is seen as “white and male” while not pointing out the prevalence of female-coded virtual assistants - a major oversight. Moreover, their claim that “AI is often today being developed by very powerful, global tech companies which are embedded within the capitalist system with its historical links to patriarchy, colonialism and neo-liberalism” (p. 337) lacks any evidence; rather it relies on ideological assumptions more than facts.
Floridi, L. (2025). Distant writing: Literary production in the age of artificial intelligence. Minds and Machines, 35:30. https://doi.org/10.1007/s11023-025-09732-1
In wAIting: The Philosophy of Artificial Intelligence in the Age of Synthetic Reason, Floridi (2025) introduces the concept of distant writing. Distant writing is a new writing practice wherein a human author acts as a narrative architect, establishing the parameters of a story through prompts and refinement while an AI system writes or constructs the text. Drawing on experiments, Floridi examines the practice by understanding its theory, method and implications for authorship, creativity and narrative structure.
Floridi argues that distant writing will significantly change writing by shifting authorship from an individual writer to a narrative designer working with AI tools. Distant writing, therefore, has the potential to expand stories and democratize literary production. It also raises questions about authorship, creative agency, ownership, and human-technological relationships. The future of literature is likely to see increased collaboration between human creators and AI.
Distant writing could lower the barrier to entry by circumventing the traditional publishing gatekeepers prompting a wave of fresh material. However, said material will not undergo quality control either. This shift may require publishers to develop new ways for determining quality and marketability, leading to significant changes in the industry’s economics.
Moreover, distributing the writing process between human designers, AI systems, and original training data creators complicates authorship and accountability. It’s even further complicated by the use of copyrighted materials in AI’s training.
Unfortunately, Floridi overstates the ethical concerns. For instance, he argues that writing as design raises the issue of how to value different types of creative labor, such as whether the prompt designer deserves recognition equal to that of the traditional author. The answer is in the product: if the story resonates with its audience then it’ll likely get equal recognition. Is a novel any less valuable because its author used spell check instead of a dictionary?
Floridi, L. (2019). What the near future of artificial intelligence could be. University of Oxford. https://doi.org/10.2139/ssrn.3379206
Floridi (2019) attempts to delineate the possible futures of AI. He considers how the nature of data and the complexity of the problems AI may be asked to solve will shape its development and use. He pays close attention to ethical issues and how they might be managed. Floridi analyzes and synthesises the literature, computational theory, and other relevant philosophies.
According to Floridi, synthetic and hybrid (i.e., synthetic plus historical) data will become the primary source for AI training. Perhaps more importantly, Floridi writes that governance, ethical considerations, and strategic design are necessary to guide AI development toward the ethical even though its future is uncertain. His insights help inform policymakers, developers, and researchers on how to develop AI responsibly.
This paper is a thorough exploration of computational theory, ethics, and design. Its emphasis on data quality, ludification, and governance provides guidance for researchers and policy makers to create ethical AI systems. However, the paper is also a conceptual piece that analyzes philosophies. As such, it has limited practical use. Even so, the study provides unique insights into possible futures of AI and offers ideas for consideration regarding its responsible development.
Hodonu‑Wusu, J. O. (2024). The rise of artificial intelligence in libraries: The ethical and equitable methodologies, and prospects for empowering library users. AI and Ethics. https://doi.org/10.1007/s43681-024-00432-7
Hodonu‑Wusu (2024) examines the ethical and equitable use of AI in libraries, specifically how it can empower users, and what librarians need to think about in its implementation. The author reviews 1,499 articles published between 2018 and 2023 selected from the EBSCO database. He focuses on ethical and fair methodologies in AI applications in libraries.
Hodonu‑Wusu concludes that using AI in libraries has the potential for personalized service, improved resource management, and increased accessibility. However, he also notes the importance of implementing AI with a strong ethical foundation, one that prioritizes transparency, protects privacy, and guarantees equal access for all users. The paper guides future research on “Trustworthy AI, Fairness in AI, Explainable AI, and Human-in-the-loop” (p. 10) and how these issues may affect libraries.
The paper’s main strength is its comprehensive review of the literature and, as Hodonu‑Wusu notes, can be a springboard for future research. On the other hand, it relies on secondary sources and does not have empirical data, restricting its real-world usefulness. The use of words like inclusivity and equity may indicate an ideological framing as well, which could bias AI design if implemented. However, it is not clear if Hodonu‑Wusu is, in fact, advocating for Critical Social Justice.
Izuchukwu, A. C., Omeje, E. O., & Ozonwu, C. U. (2024). Digital wellness: The implications of the use of ChatGPT for teaching and learning in library schools in Enugu State. Journal of Applied Information Science and Technology, 17(1), 109-120.
Izuchukwu et al. (2024) explore the effects of ChatGPT on teaching and learning in library schools in Enugu State, Nigeria. The authors focus on digital wellness, defined as the physical, emotional and social well-being in a digital context. The paper mixes qualitative and quantitative methods, utilizing surveys and questionnaires and a sample size of 150 students from three universities.
The authors found that ChatGPT enhances critical thinking, information retrieval, and problem solving skills. However, it also increases screen time, possesses biases, and potentially risks privacy. It may also replace human librarians. Izuchukwu et al., therefore, recommend schools establish clear guidelines for AI use, and provide training and support. Schools should also encourage collaborative learning and emphasize the need for human librarians. The study concludes with a call for a balanced approach in integrating AIs like ChatGPT into library school curricula. Their findings could guide policy creation, curriculum design, and further research on AI in education.
The strengths of this study lie in its use of both qualitative and quantitative methods. The results are displayed using basic percentages, standard deviations, and means as analyzed with Statistical Package for the Social Sciences (SPSS) (p. 113). Thus, their findings have relevance and applicability. However, the study also relies on self reported data within a narrow geographical context (e.i., Enugu State, Nigeria). This may limit its generalizability. Moreover, a sample size of 150 students from three universities is not large enough to draw any concrete conclusions. Still, this research is helpful; it offers real-life insight into how AI can be responsibly integrated into educational settings.
Michalak, R. (2023). From ethics to execution: The role of academic librarians in artificial intelligence (AI) policy-making at colleges and universities. Journal of Library Administration, 63(7), 928–938. https://doi.org/10.1080/01930826.2023.2262367
Michalak (2023) argues that academic librarians have special expertise that makes their involvement in the development of ethical AI policies indispensable. He also raises challenges like limited awareness, institutional resistance, resource limitations, collaboration, and the evolution of AI technologies. Finally, he introduces the Ethical AI Policy Development (ALF Framework).
Michalak reviews existing discussions regarding the role of librarians in AI policy development. He concludes that the involvement of academic librarians in said development is essential for promoting social responsibility, human rights, and the common good. These librarians have expertise in information ethics, privacy, and intellectual freedom which enables them to guide the discussions on ethical research, address bias, and develop data policies, ensuring AI systems are transparent, fair, and accountable.
Michalak makes a convincing argument. Academic librarians, indeed, have expertise that make their input on the development of ethical AI systems important, although perhaps not indispensable. His ALF framework is a well-defined means by which librarians should go about it. Michalak even offers methods by which it can be introduced at universities. However, the paper is just an argument. It does not offer any empirical evidence of the benefits potentially gained from librarians’ involvement. Still, it highlights the ethical importance and practical means of engaging librarians while also serving as a roadmap for university administrators to foster ethical AI use.
Nyholm, S. (2018a). The ethics of crashes with self‐driving cars: A roadmap, I. Philosophy Compass, 13(7), e12507. https://doi.org/10.1111/phc3.12507
In this article, Nyholm (2018a) provides an “opinionated” (p. 2) review of the main discussions regarding the ethics of self-driving cars, specifically how autonomous vehicles should be programmed to respond in crash scenarios. He asks the question, “Should cars be programmed to always prioritize their owners, to minimize harm, or to respond to crashes on the basis of some other type of principle?” (p.1). He aims to introduce readers to the main ideas on this topic so far while also critically assessing those ideas so as to indicate where the conversation might go. As a conceptual review rather than an empirical study, Nyholm summarizes and evaluates the existing literature, including philosophical thought experiments, psychological research, and ethics.
Nyholm identifies three areas of debate:
Whether users should choose their ethical settings using an “ethical knob” (p. 3) or if a standard ethical setting should be mandated;
The usefulness versus the limitations of modeling self-driving car crash dilemmas on the philosophical trolley car problem; and
How can traditional moral theories (utilitarianism, Kantianism, virtue ethics, and contractualism) answer this new ethical dilemma?
He argues that utilitarian and consequentialist views ignore responsibility and public acceptance. Instead, Nyholm favors a pluralistic approach, combining the solutions of several moral traditions.
The article’s message goes beyond the philosophical; it has implications for public policy and car design. Nyholm writes that ethical programming decisions must be transparent, democratically accountable, and informed by both moral reasoning and social trust. Nevertheless, his survey is mostly theoretical and stops short of actual guidelines that could be put into place. Rather, it merely shows how classical moral theories can be applied to the new ethical dilemmas of autonomous driving.
Nyholm, S. (2018b). The ethics of crashes with self‐driving cars: A roadmap, II. Philosophy Compass, 13(7), e12506. https://doi.org/10.1111/phc3.12506
In this continuation of his earlier paper, Nyholm (2018b) dives into the legal and moral responsibility of autonomous car crashes. He asks the following questions:
When there are crashes involving self‐driving cars, who should be held responsible? If a fully automated car is seemingly operating as some form of artificial autonomous agent on the road, can any human beings fairly be held responsible for any harm the car might cause? And if it really is the case that self‐driving cars will prove to be much safer than regular cars, what new ethical duties might this create for people who use cars? (p. 2).
The legal literature suggests that responsibility should be divided between manufacturers, software developers, the vehicle owners depending on the circumstances of the crash. But Nyholm suggests the autonomy of these cars challenges the norm of individual responsibility and, therefore, a human-robot collaboration framework would make more sense. The issue is further complicated if the self-driving car is regarded as its own autonomous agent, in which case any human responsibility would be absurd.
If autonomous vehicles prove to be safer than standard cars, Nyholm suggests motorists have a moral duty to adopt the safer technology. Moreover, there may be a societal duty to create and enforce policies that promote the widespread use of these vehicles so as to prevent accidents and injuries.
Nyholm’s argument that self-driving cars challenge the norm of individual responsibility could have been developed further. The legal literature he cites seems to have the issue of responsibility under control. To use a non AI example, if the vehicle’s brakes fail the responsibility will fall on whoever caused the failure, be that the manufacturer for producing faulty brakes, the technician for installing them improperly, or the car owner who didn’t have them checked.
Osterlund, C., Jarrahi, M. H., Willis, M., Boyd, K., & Wolf, C. T. (2020). Artificial intelligence and the world of work: A co-constitutive relationship. Journal of the Association for Information Science and Technology, 71(4), 411–426. https://doi.org/10.1002/asi.24388
The breakthroughs in AI research will change the way work is organized, argues Osterlund et al. (2020), but the integration remains an open challenge. Their paper seeks to understand the relationship between AI and work through an interdisciplinary lens. They draw on a 2019 iConference workshop about Work in the Age of Intelligent Machines (WAIM). This interactive workshop used group discussions and reflective analysis to examine how AI integrates with and transforms work practices.
A recurring theme the authors found at the workshop is the need to move beyond superficial connections and assumptions about AI while remaining open to multiple potential futures. Investigating these futures requires understanding both how AI is integrated into the workplace and how the nature of work itself shapes AI development. Work sneaks into AI through the big data that powers its algorithms as well as through data management practices, embedded biases from past practices, and the ethical considerations.
Osterlund et al.’s discussion is mostly conceptual, with limited empirical evidence to substantiate their claims, which reduces its usefulness. Nevertheless, the study is useful for suggesting future research directions and promoting dialogue, though more empirical research is needed to make these insights actionable.
Siau, K., & Wang, W. (2020). Artificial intelligence (AI) ethics: Ethics of AI and ethical AI. Journal of Database Management, 31(2), 74–87. https://doi.org/10.4018/JDM.2020040105
Siau and Wang (2020) attempt to differentiate between the ethics of AI and ethical AI. The former refers to the moral principles guiding the design, development, and governance of AI systems as well as ethical concerns caused by AI (e.g., unemployment). The latter refers to AI that behaves ethically in its interactions with humans, other machines, and society. The authors use a conceptual and comparative analysis in lieu of original empirical research drawing on philosophical traditions, applied ethics, and policies developed by major organizations such as IEEE, UNESCO, and the European Commission.
The ethics of AI are still in its infancy, Siau and Wang argue. Perennial concerns include autonomy, privacy, transparency, bias, accountability, and potential societal harms like job displacement. They further argue that an ethical AI cannot exist without first establishing a strong sense of ethics in AI development. The ethics of an AI must include respect for human rights, data protection, and fairness. The goal is to develop ethical frameworks for AI that not only guide AI development but also deepen our understanding of human ethics, improve existing principles, and foster better interactions with AI.
The ethical issues related to AI can be placed into three categories: the features of AI that may lead to ethical challenges, human factors contributing to ethical risks, and the societal impacts resulting from ethical concerns associated with AI. The last of these includes job displacement, which is debated but may lead to significant societal upheaval.
While Siau and Wang bring to fore serious complications regarding AI, they underestimate the need for commercial viability for its development. They lament AI developers and researchers prioritizing measurable performance metrics like safety, reliability, usability, and customer satisfaction over the more abstract ideas of ethics. Further, they write, “The most important factors influencing consumer’s purchasing decisions are still the price and quality” (p.84). For better or worse, this is an indispensable part of life.
Slimi, Z., & Carballido, B. V. (2023). Navigating the ethical challenges of artificial intelligence in higher education: An analysis of seven global AI ethics policies. TEM Journal, 12(2), 590-602. https://doi.org/10.18421/TEM122-02
Slimi and Carballido (2023) investigate the challenges the use of AI poses in higher education. They pay particular attention to biased algorithms, AI-powered decision-making, and human displacement. They examined seven global AI ethics policies - including documents from UNESCO, China, the European Commission, Google, MIT, Sanford HAI, and Carnegie Mellon - with the intent to find common themes and assess how these policies might affect AI in higher education. They conclude that stakeholders must hold AI practices responsible for transparency, accountability, and fairness.
Their findings, if accepted, call for international standards on the ethical use of AI in higher education that may guide policymakers, educators, and developers toward AI systems that enhance education while preserving human rights and employment.
However, the study is severely limited by an assumed ideology, one that accepts claims of systemic bias and structural oppression without requiring evidence. For example, the authors assert that black patients are placed in longer wait-time appointments (p. 591), but they do not consider other possible contributing factors. Moreover, they do not show scheduling practices are, in fact, the cause of missed appointments.
Slimi, Z., & Carballido also do not question if their proposed solutions may reduce academic freedom, introduce new forms of gatekeeping, or lead to political conformity in institutions. Policies framed as promoting equity can easily become tools for institutional gatekeeping, compelled consensus, and the policing of acceptable viewpoints under the banner of “responsible AI.”
Tait, E., & Pierson, C. M. (2022). Artificial intelligence and robots in libraries: Opportunities in LIS curriculum for preparing the librarians of tomorrow. Journal of the Australian Library and Information Association, 71(3), 256–274. https://doi.org/10.1080/24750158.2022.2081111
Tait and Pierson (2022) evaluate the challenges and opportunities AI and robotics pose to library and information science (LIS) education. They argue that integrating AI and robotics into LIS educational programs is essential for LIS education to remain relevant and updated. To achieve this, the authors reviewed subject descriptions and curricula from Australian universities accredited by the Australian Library and Information Association (ALIA), alongside an ALIA document that outlines core knowledge, skills, and attributes for information professionals.
They found that only one course mentioned AI and none mentioned robotics. They also identified several ways to incorporate AI and robotics into the five categories of the ALIA foundational knowledge, while also allowing for the special needs of various programs and institutions. Tait and Pierson’s study suggests that LIS education in Australia is not adequately preparing LIS students for the coming “fourth industrial revolution” (p. 256), a term referring to the proliferation of AI and robotic technologies.
This study’s use of publicly available course descriptions - and only course descriptions - make its findings preliminary at best. Still, the methodology is sound, and the findings reliable as far as they go. Moreover, this study has real-world applications (namely, the integrating AI courses into LIS educational programs), which is more than can be said about many conceptual studies on this topic. It is, however, limited to an Australian context.
.