Join Our Discord (630+ Members)

What Does It Mean to Be a Responsible AI Practitioner: An Ontology of Roles and Skills

Content License: cc-by

What does it mean to be a responsible AI practitioner: An ontology of roles and skills

Papers is Alpha. This content is part of an effort to make research more accessible, and (most likely) has lost some details from the original. You can find the original paper here.


With the rapid growth of the AI industry, the need for AI and AI ethics expertise has also grown. Companies and governmental organizations are paying more attention to the impact AI can have on our society and how AI systems should be designed and deployed responsibly. From 2015 onward, a series of AI ethics principles, in-depth auditing toolkits, checklists, codebases, standards and regulationshave been proposed by many different international actors. Several communities of research and practice such as FATE (Fairness, Accountability, Transparency, and Ethics), responsible AI, AI ethics, AI safety and AI alignment have emerged. This general movement towards responsible development of AI has created new roles in the industry referred to as responsible AI practitioners in this paper. The primary mandate of these roles is understanding, analyzing, and addressing ethical and social implications of AI systems within the business context. The emergence of these roles challenges technology companies to curate these roles and teams. Leaders in AI-related organizations need to identify, recruit and train appropriate candidates for such roles. As the demand to fill such roles continue to increase, educators need effective means to train talent with the right set of skills.

Recently, scholars examined the common roles responsible AI practitioners serve, explored the challenges that they face, and criticized the problematic nature of the accountability mechanisms that relate to these roles. Moreover, others highlight the myriad practical challenges facing the development of a comprehensive training program to fill such roles. However, there is a lack of empirical research investigating the types of roles, corresponding responsibilities, and qualifications that responsible AI practitioners have in the industry. To address these gaps, we examine the following research questions:

  • RQ1: What are the types of roles and responsibilities that responsible AI practitioners hold in the industry?
  • RQ2: What are the skills, qualifications, and interpersonal qualities necessary for holding such roles?

We address these questions by conducting a two-part qualitative study. We examined 79 job postings from March 2020 to March 2022 and conducted expert interviews with 14 practitioners who currently hold these roles in the industry. Learning from fields of competency-based recruitment and curriculum development, we propose an ontology of different occupations and an accompanying list of competencies for those occupations.

As illustrated in Figure fig-teaser, our competency framework outlines seven occupations that responsible AI practitioners hold in the industry: researcher (of two kinds), data scientist, engineer, director/executive, manager, and policy analyst. For each occupation, the ontology includes a list of responsibilities, skills, knowledge, attitudes, and qualifications. We find that while the roles and responsibilities held by responsible AI practitioners are wide-ranging, they all have interdisciplinary backgrounds and are individuals who thrive in working with individuals from different disciplines. We discuss how educators and employers can use this competency framework to develop new curricula/programs and adequately recruit for the rapidly changing field of responsible AI development.


With the increased media reporting and regulation requirements around social and ethical issues of AI-based products and services, the role of a responsible AI practitioner has emerged as a demanding position in the technology industry. In this section, we provide an overview of debates about these roles and existing educational programs that aim to train future responsible AI practitioners. We discuss how existing competency frameworks treat the role of a responsible AI practitioner and highlight the gaps we address in this work.

Emergence of the responsible AI practitioners

Considering the nascency of AI ethics as a domain, only a few scholars have characterized occupations held by responsible AI practitioners. For instance, Gambelin frames the role of an AI ethicist as “an individual with a robust knowledge of ethics” who has the responsibility and the ability to “apply such abstract concepts (i.e. ethical theories) to concrete situations” for the AI system. According to Gambelin, an AI ethicist in the industry also needs to be aware of existing policy work, have experience in business management, and possess excellent communication skills. Gambelin identifies bravery as the most important characteristic of an AI ethicist as they often need to “shoulder responsibility” for potential negative impacts of AI in the absence of regulation.

Moss and Metcalf investigated practices and challenges of responsible AI practitioners in Silicon Valley and described them as “ethics owners” who are responsible for “handling challenging ethical dilemmas with tools of tech management and translating public pressure into new corporate practices”. Echoing Moss and Metcalf’s seminal work on examining AI industry practices, a growing body of empirical work highlights that responsible AI practitioners face challenges such as misalignment of incentives, nascent organizational cultures, shortage of internal skills and capability, and the complexity of AI ethics issues when trying to do their day-to-day tasks. Furthermore, only large technology companies often have the necessary resources to hire responsible AI practitioners. Small and medium-sized companies struggle to access such expertise and rely on openly available information or hire external consultants/auditors as needed. This has given rise to AI ethics as consulting and auditing service.

While challenges in operationalizing responsible AI practices are an active area of research, there is a gap in understanding the role and necessary competencies of responsible AI practitioners in the industry.

Qualifications to be a responsible AI practitioner

The emergence of auditors in the field of responsible AI emphasizes the need for formal training and certification of such roles in the industry. This raises a few practical questions: Who is qualified to take these roles? How should these individuals be trained? Are existing computer science, engineering, and social science curricula prepare individuals for such roles?

Educators responded to this need by developing a range of educational programs and curricula. In a survey of the curricula for university courses focused on AI ethics, Garrett et al. emphasize that such topics should be formally integrated into the learning objectives of current and new courses. On the other hand, as Peterson et al. describe, discussing social and ethical issues in computer science courses remains a challenge. They propose pedagogues for fostering the emotional engagement of students in the classroom as a solution.

Recognizing the importance of interdisciplinary approaches in AI ethics, Raji et al. argue that computer science is currently valued significantly over liberal arts even in the research area of fairness of machine learning systems. Furthermore, they state that the perceived superiority culture in computer science and engineering has created a “new figure of a socio-technical expert”, titled “Ethics Unicorns” - full stack developers, who can solve challenging problems of integrating technology in society.

This overemphasis on computer science expertise and the trend toward integrating ethics content in existing technical curricula may be problematic if these efforts do not match the skills and disciplinary needs of the industry. It raises questions about whether the educational backgrounds of responsible AI practitioners today are indeed in computer science. In this work, we inform the curriculum development efforts across a diverse range of disciplinary areas by understanding these roles in the industry and outlining the attributes, qualifications, and skills necessary for holding them.

Competency frameworks in AI and AI ethics

Competency frameworks are useful tools for human resource management (i.e. recruitment, performance improvement) and educational development (i.e. new training programs and curriculum development in universities). Competency frameworks highlight different competencies required for a profession and link these competencies to skills and knowledge. According to Diana Kramer “competencies are skills, knowledge and behaviours that individuals need to possess to be successful today and in the future”. This definition frames our discussion of competency in this paper.

Competency frameworks help governmental and non-governmental organizations keep track of the type of skills their employees/general public need in the short and long term. Educators use these frameworks to update existing curricula and develop appropriate learning objectives. On the other hand, business leaders and human resource professionals use these frameworks for their recruitment practices.

Today’s existing competency frameworks do not sufficiently represent roles and competencies of a responsible AI practitioner. For example, ONET is United State’s national program for collecting and distributing information about occupations. ONET-SOC is a taxonomy that defines 923 occupations and they are linked to a list of competencies. Searching the taxonomy for “ethics”, “machine learning”, “data”, “security”, and “privacy” leads to minimal results such as “information security analysis”, “data scientist and “database architect”. The dataset do not include occupation titles such as machine learning engineer/researcher or data/AI ethics manager.

ESCO, the European skills, competencies, qualifications, and occupation is the European and multilingual equivalent of US’s O*NET. ESCO contains 3008 occupations and13890 skills. Searching for the above terms leads to more relevant results such as computer vision engineer, ICT intelligent system designer, policy manager, corporate social responsibility manager, ethics hacker, data protection officer, chief data officer, and ICT security manager. However, emerging occupations relevant to AI and AI ethics have not been well-represented in these established, Western competency frameworks.

As a response, a number of new AI competency frameworks have recently been developed. One such enabler is the series of projects funded by the Pôle montréalais d’enseignement supérieur en intelligence artificielle (PIA), a multi-institutional initiative in Montreal, Canada aimed to align educational programs with the needs of the AI industry. Six projects related to AI competency frameworks were funded – including the work presented in this paper. This resulted in an overarching AI competency for postsecondary education that includes ethical competencies, and a competency framework specific to AI ethics skills training. Bruneault et al., in particular, created a list of AI ethics competencies based on interviews of university instructors/professors already teaching courses related to AI ethics across North America.

Our work complements these collective efforts by providing a framework that represents the needs of the industry expressed in recent AI ethics-related job postings and the realities of the jobs AI ethics practitioners hold in nonprofit and for-profit corporations today.


Practitioners and scholars of different domains typically create competencies frameworks using a process most appropriate for their needs. However, many follow a version of the process highlighted by Sanghi. The steps of the process are: 1) Define the purpose and performance objective of a position, 2) Identify the competencies and behaviors that predict and describe superior performance in the job, 3) Validate selected competencies, 4) Implement/integrate competencies and 5) Update competencies.

In this work, we focus on answering questions raised in the first two steps about the objectives of responsible AI practitioner roles and skills/qualities required to perform well in these positions. We take a two-pronged approach to understand the nature of emerging roles under the broad category of responsible AI practitioners in the industry. Firstly, we reviewed and analyzed job postings related to our working definition of responsible AI practitioner . Secondly, we interviewed individuals who are responsible AI practitioners in the industry today. We then synthesized data collected from these two sources through thematic analyses. We present our proposed competency framework in Section finding. This study was approved by the Research Ethics Board of our academic institution.

AI Ethicist Job Postings Review

We collected and analyzed 94 publicly available job postings over the period of March 2020 to March 2022. The job postings included a range of job titles, including researcher, manager, and analyst.The following sections describe the process for collecting, selecting, and analyzing these job postings that led to the development of the ontology of responsible AI practitioner roles and skills.

Collection of job postings

To collect “AI ethicist” job postings, we searched and scraped three job-finding websites, including LinkedIn,, and SimplyHired, every two months from March 2020 to March 2022. We used the following search terms: AI ethics lead, Responsible AI lead, AI ethics researcher, data OR AI ethicist and fairness OR transparency researcher/engineer. Considering that search results only showed a few relevant job postings, we also collected job postings that came through referrals, including mailing lists such as FATML,, and roboticsworldwide.

After scanning all the resulting job postings with the inclusion criteria, we gathered a total of 79 job postings for thematic analysis. We included the job postings that were published within our data collection period, were situated in the industry (including not-for-profit organizations), and outlined responsibilities with regards to implementing AI ethics practices in a given sector.[The table outlining the inclusion and exclusion criteria is in the supplemental material.]


Using Braun and Clarke’s thematic analysis methodology, we analyzed the job postings with the coding scheme illustrated in Table tab-jp-coding.The lead author created this coding scheme after reviewing all the postings. The coding scheme was also informed by frequently used categories across competency frameworks explained earlier in section competencyframework.

The codes were generally split into four key elements: the company environment, responsibilities in the given occupation, qualifications, and skills. The codes of “company environment” and “qualifications - interdisciplinarity” are unique to this coding scheme due to their prevalence in the postings’ content.

After developing the first draft of the coding scheme, a student researcher was trained to use this scheme and coded 10% of the job postings. The student researcher’s analysis using the coding scheme was consistent with the lead researcher’s analysis of the same set of job postings. The discussion between the lead and student researcher helped clarify the description and examples for each code. However, there were no new codes that were added to the scheme. The lead author updated the coding scheme and coded the entire set of postings using the new scheme.

Coding scheme for Job Posting Analysis

Table Label: tab-jp-coding

Download PDF to view table

Expert interviews

The job postings provide a high-level analysis of the required skills and competencies expressed by recruiters; however, they may not represent the reality of these roles. Therefore, we conducted 14 interviews with experts who currently hold responsible AI practitioner positions in the industry. The focus of the interviews was on understanding the responsibilities, qualifications, and skills necessary for these roles. Considering the objective of this research project on the type of roles and skills, we did not acquire any demographic information about the participants in these roles. This also ensured that we can maintain the anonymity of these participants considering that a limited number of people hold these positions.


We compiled a list of potential interview candidates through (a) referrals within the authors’ professional network and (b) we used similar search terms as the ones highlighted for job postings to look for people who currently hold these positions. Moreover, we also considered people from the industry who had accepted papers at relevant conferences such as FAccT and AIES in 2020 and 2021. The suitable participants:

  • worked for a minimum of three months in their role;
  • held this position in the industry or worked mainly with industrial partners;
  • held managerial, researcher, technical positions that are focused on implementing responsible AI practices within the industry.

We did not interview researchers or professors in academic institutes and only interviewed those holding positions at nonprofit and for-profit companies. While we only used the search terms in English to find interview participants for practical reasons, we did not limit our recruitment efforts to a geographical region given the limited number of individuals holding these roles across the industry. We recruited and conducted interviews from June 2021 - February 2022.

Interview protocol

The primary researcher conducted all fourteen interviews. All of the interviews were 45 to 60 minutes in length. The interviewer first described the project and obtained the participant’s consent. The interview was semi-structured with ten questions focused on exploring the following four topics:[The detailed interview protocol is included in the supplementary material.]

  • Background and current role
  • Situation your work, projects in AI ethics
  • Skills, knowledge, values
  • Looking into the future

Data Analysis

The interviews were audio recorded and transcribed. The primary researcher checked and corrected the transcriptions manually afterwards. The authors analyzed the interviews deductively and inductively. The lead author applied the coding scheme derived from the job postings () to the interview transcripts. Furthermore, a reflexive thematic analysiswas done by the lead author to capture the rich and nuanced details that were not represented in the coding scheme for the job postings. The results from the reflexive analysis were used to contextualize and improve the coding scheme from the job postings. However, no new codes were added to the scheme considering the focus of the research objective addressed in this paper. The lead author presented the coding scheme to the full research team, including a student researcher who had independently reviewed the data to identify general trends. The coding scheme was iteratively revised and refined based on the feedback from the research team.

Author reflexivity and limitations

We recognize that this research reflects our positionality and biases as academics in North America. Furthermore, the data we collected were all in English and they were representative of job postings and positions in companies situated in North America and Europe. We were not able to collect data on job postings and candidates representing existing efforts in Asia and the Global South. Furthermore, we recognize that the roles in this field are continually shifting. Therefore, this ontology is only a snapshot of the roles and skills that responsible AI practitioners have and are recruited for today. Further iterations on these types of frameworks will be necessary in the future as these roles evolve. Finally, this study focuses on examining responsibilities, qualifications, and skills required of today’s practitioners independent of their demographic factors (e.g., gender, age). We recognize the importance of representing a demographically diverse group of individuals and their experiences in qualitative research such as ours. Once responsible AI practitioners become a common occupation held by many, future studies should include demographic factors as part of similar investigations.

Proposed competency framework for responsible AI practitioners

From our analysis, we developed a preliminary competency framework that captures seven classes of existing occupational roles and several emerging classes of occupations. Figures postingchart and interviewchart show how each occupation type was represented in the job postings and interviews. Three of the occupations require technical expertise (researcher, data scientist, and engineer), two require policy expertise (researcher, policy analyst), and the remaining two are managerial (manager, director). In the following sections, we provide a detailed description of the responsibilities, skills, qualifications, and qualities for each of these roles.

Distribution of occupations represented in the job postings datasetpostingchart

Distribution of occupations represented in the job postings dataset

Distribution of occupations represented in the interviewsinterviewchart

Distribution of occupations represented in the interviews

Researcher (technical)

The most common class of occupations found in the job postings was that of a researcher focused on technical aspects of fairness, explainability, safety, alignment, privacy and auditability of AI systems (24 job postings, 2 interviews). Employers represented in this dataset were looking to hire researchers at varying levels of seniority (assistant, associate and principal). The main responsibilities of these researchers are split into four main categories: conducting research, communicating their findings, working with other teams (internally and externally), and developing novel solutions for identified problems. As expected, research directions set by these researchers need to support company-specific needs, and there is an emphasis on communication between researchers and product, legal and executive teams.

Skills The researchers in this group need to have a mix of technical skills (i.e. software engineering and programming languages such as Python), research skills (i.e. analytical thinking and synthesis of complex ideas ), and leadership skills (i.e. leading and guiding fellow researchers). The dataset from the job postings emphasized equally all these skills, and more senior positions emphasized leadership skills. A senior researcher explained that they look for “different research skills” depending on the project; however, they generally look for “some background in machine learning, statistics, computer science or something of that nature” and hire candidates that have some “interdisciplinary background” . The data from the postings and the interviews show a strong emphasis on good verbal and written communication skills. Participants highlighted the ability to publish in academic venues and some emphasized the ability to communicate with different audiences internally (i.e. product teams and executives) and externally (policy-makers and executives). A technical researcher emphasized the importance of_“convincing stakeholders”_ and creating “strategic collaborations” by communicating with practitioners with “diverse” backgrounds.

Qualifications The job postings mainly aim to attract candidates who have a PhD in computer science or a related field. Few of the job postings accept a master’s in these fields, whereas some do not highlight a specific degree and mainly focus on necessary skills and knowledge. The majority of postings have a heavy emphasis on the required experience. Interview participants also emphasized the importance of experience. A research manager expressed that they are not necessarily looking for a “PhD in computer science” . They are looking for candidates with experience in “leading and executing a research agenda” , working with different people and teams, synthesizing and “communicating challenging concepts” , and practicing software engineering. Some postings highlight experience with implementing AI ethics-related concepts. However, this was often listed as a preferred qualification rather than a required one. Similarly, researchers we interviewed, echoed the importance and value of having a publication record in “Fairness, Accountability, Transparency, and Ethics (FATE) communities” such as ACM Conference on Fairness, Accountability, and Transparency (FAccT) and AAAI/ACM Conference on AI, Ethics, and Society (AIES).

Interpersonal Qualities The most common attitude/value was the aptitude and interest to collaborate and work in an interdisciplinary environment . A researcher emphasized that the current conversations are “engineering focused” and they actively incorporate perspectives from_social science and philosophy by collaborating with experts in these areas_ . The most desired value was “curiosity to learn about [responsible AI] problems” . Many of the participants highlight other values and attitudes such as “passion” towards building safe and ethical AI systems, willingness to manage uncertainty and challenges , creativity , and resourcefulness .

Data scientist

The data scientist occupation is represented in 10 job postings in our dataset, and none in the interviews. The job postings seek to fill traditional data scientist roles with an added focus on examining responsible AI-related issues. The common responsibilities outlined for these positions are a) to collect and pre-process data, and b) to develop, analyze, and test models – these are typical of existing data science roles. However, the job postings emphasize the position’s responsibility to test machine learning models for AI ethics concerns such as fairness and transparency. Data scientists who work in the responsible AI domain have additional non-conventional roles. These roles include understanding and interpreting existing regulations, policies, and standards on the impact of AI systems and testing the systems’ capability for elements covered in these policies. They also need to work with technical and non-technical stakeholders to communicate findings, build capacity around responsible AI concepts and engage them as needed.

Skills The job postings put a heavy emphasis on advanced analytical skills and the ability to use programming languages such as R, Python and SQL for basic data mining. The ability to learn independently in a new domain and master complex code base is also listed as one of the key skills. A few of the postings list project management and organizational skills; however, this is not common. When it comes to the knowledge required, the focus shifts from the technical domain to an understanding of fields such as sociology, critical data studies, and AI regulations. Many postings highlight that potential candidates need to be familiar with concepts such as AI/ML auditing, algorithmic impact assessments, assessment of fairness in predictive models, explainability, robustness, and human-AI interaction. Technical knowledge, such as understanding transfer-based language models and logistic regression model development, is also highlighted in the posting. Lastly, the job postings outline the need for strong interpersonal, verbal, and written communication skills. However, experience publishing and presenting at academic venues is not mentioned.

Qualifications The majority of the job postings require a bachelor’s degree in quantitative fields such as data science and computer science and prefer higher degrees (master’s or Ph.D.). Companies are looking for candidates who have experience in data science, software engineering, and worked with large language models. Moreover, they are looking for experience in putting responsible AI principles into practice, evaluating the ethics of algorithms, and having basic familiarity with law and policy research. The ability and experience to translate AI ethic principles into practice are heavily emphasized throughout these job postings.

Interpersonal Qualities The job postings emphasize the ability to work with people from different backgrounds. However, these job postings do not include a comprehensive list of values. A few postings mention being a self-starter, working collaboratively to resolve conflict, and caring deeply about the data used to train ML models as key attitudes. Being flexible, innovative, curious, adaptive, and passionate about tackling real word challenges are also some of the sought-after values.


The engineer occupation is represented in 8 of the job postings. None of our interview participants belong to this category. The key responsibility of an engineer practicing AI ethics is to help establish a safety culture and system within an organization by developing technical tools. They are tasked with developing a workflow for modeling and testing for issues such as bias, explainiability, safety, and alignment of AI systems. As part of this, engineers need to create code bases that could be used across the AI system development pipeline based on existing and evolving best practices.

Skills and Qualifications Job postings for engineers place a significant emphasis on experience-based qualifications and skills. The companies represented in this dataset are looking for skills and experience in software development, dataset production pipelines, researching fairness and safety implications of ML systems, and the development of large language models. They are also looking for experience working in a fast-paced technology company. Based on these qualifications, the main set of skills are programming and AI/ML development skills and this needs to be supported by knowledge and familiarity with foundational concepts in AI/ML, fairness, explainability, system’s safety, and safety life cycle management. Lastly, most of the job descriptions do not have a heavy emphasis on communication skills. Only a few mention excellent written and oral communication skills as a requirement.

Interpersonal Qualities In contrast to the lack of emphasis on communication skills, these postings have a particular focus on the attitude and values of ideal candidates more so than any other occupation category. These attitudes include being result-oriented, willingness to contribute as needed (even if not specified) and keen to learn new concepts. They are looking for people who value working on challenging problems and care about the societal impact of their work.

Researcher (law, philosophy and social sciences)

The second most frequent category of postings belongs to researchers that focus on topics such as policy, sociotechnical issues, and governance (14 job postings, 3 interviews). We created a separate group of positions as their responsibilities, skills, and qualities are sufficiently different from the technical researcher position. Candidates in this category need to conduct research, perform ethics or impact assessments of AI systems, act as a liaison and translator between research, product, policy, and legal teams, and lastly, advise on policy, standards, and regulations-related matters internally and externally. When conducting research, two different focus areas come up in the job postings: testing and evaluating AI system to inform policy and researching existing policies/regulations, and translating them into practice.

Skills The job postings highlight two sets of distinct skills for this group of researchers. Firstly, these researchers require a basic level of programming, advanced analytics, and data visualization skills. Few positions highlighted the need for even more advanced ML and AI skills. It is noteworthy that despite these researchers’ focus on policy, governance and sociotechnical issues, the postings still require them to have some data analytic skills. Secondly, these researchers need to have excellent facilitation, community-building, and stakeholder engagement skills. These two skills need to be complemented by strong leadership and management skills. The job postings heavily emphasize strong communication skills for this group of researchers. Besides the conventional skill of presenting and publishing papers, this group of researchers need to effectively work across different functionalities and disciplines. On a similar trend, these researchers need to have expertise in a variety of areas. They need to have a good understanding of “qualitative and quantitative research methods” , reliably know the current and emerging_“legal and regulatory frameworks and policies”_ , be “familiar with AI technology” and have a good knowledge of practices, process, design, and development of AI technology. This is a vast range of expertise and often “very difficult to recruit” for as highlighted by our expert interviewees.

Qualifications Just over half of the job postings list a Ph.D. in relevant areas as a requirement, including human-computer interaction, cognitive psychology, experimental psychology, digital anthropology, law, policy, and quantitative social sciences. Two postings require only a bachelor’s or a master’s in the listed areas. Similar to the technical researcher occupation, some positions do not specify any educational requirements and only focus on experience and skills. Our expert interviewees in this category are from a range of educational backgrounds ranging from a master’s in sociotechnical systems, a law degree combined with a background in statistics, and a master’s in cognitive systems.

Besides experience in research, companies are looking for experience in translating research into design, technology development, and policy. A researcher explained that they need to do a lot of “translational work” between the academic conversation and product teams in companies. A good candidate for this occupation would have “project management” , “change management” , “stakeholder engagement” , and “applied ethics” experience in a “fast-paced environment”. All four of these skills do not appear in all of the job postings and interview discussions. However, a permutation of them appears throughout the job posting data and participants’ responses.

Interpersonal Qualities

As emphasized strongly in both of the datasets, ideal candidates in this category need to have a “figure-it-out somehow” or “make it happen” attitude as explained by a participant. They are “driven by curiosity and passion towards” issues related to responsible AI development and are excited to engage with the product teams. Participants noted that ideal candidates in these roles are “creative problem solvers” who can work in a “fast-changing environment” .

Policy analyst

Policy analyst occupation is the least represented [1 expert interview, 4 job postings] in our data sources; however, considering the consistent list of competencies, we decided to include it within the proposed framework. The role of a policy analyst is to understand, analyze and implement a given policy within an organization. Moreover, they need to engage with policymakers and regulators and provide feedback on existing policies.

Skills and Qualifications A policy analyst needs to have proven knowledge of laws, policies, regulations, and precedents applicable to a given technology when it comes to AI ethics-related issues. Moreover, all of the job postings highlight the importance of familiarity with AI technology. According to the job postings, a good candidate would have experience in interpreting policy and developing assessments for a given application. They also need to be skilled in management, team building, and mentorship. This finding echoes remarks from expert interviews. Even though none of the job postings specify an educational degree requirement, the expert we interviewed was a lawyer with a master’s in technology law.

Interpersonal Qualities The job postings in this category heavily emphasized values and attitudes. A good analyst needs to have sound judgment and outstanding personal integrity. They should be caring and knowledgeable about the impact of technology on society. Moreover, they enjoy working on complex multifaceted problems and are passionate about improving governance of AI systems. The expert interviewee’s perspective closely matches these attributes. Participants elaborated that they needed to be “brave” and “step up to ask questions and challenge status quo consistently over a long time” . As expected communications skills are considered critical for success. The expert interviewee significantly emphasized the importance of “networking as a key factor” in succeeding in their role.


We analyzed 7 management-related job postings and 5 expert interviewees in this category. The product managers take the role of incorporating responsible AI practices in the product development process. In contrast, program managers are often leading and launching a new program on establishing AI ethics practices within the organization. These programs often involve building an organization’s capacity to manage responsible AI issues.

Skills For both streams of management, the potential candidates need to have strong business acumen and a vision for the use/development of AI technology within an organization. Some of the key management skills highlighted in the job postings include the ability to manage multiple priorities and strategically remove potential blockers to success. Another sought-after skill is the ability to effectively engage stakeholders in the process. Expert interviewees also echoed the importance of this skill as their roles often involve getting people “on board with new ways of thinking and creating” . According to the job postings, good candidates for management need to have a practical understanding of the AI life cycle and be familiar with integrating responsible AI practices into a program or a product. Our interviewees note that they continuously need to “learn and keep up with the fast-paced development of AI” .

Qualifications Not many postings have highlighted educational qualifications and instead focused on experience qualifications. However, the main educational qualification is a bachelor’s degree with a preference for higher degrees. The postings have primarily highlighted a degree in a technical field such as computer science or software engineering. Interestingly the interviews reflect a different flavor of educational backgrounds. All of the experts we interviewed had at minimum a master’s degree and the majority of them completed their studies in a non-technical field such as philosophy, media studies, and policy. However, these individuals had acquired a significant level of expertise in AI ethics through “self-studying” and “engaging with the literature” and the responsible AI “community” . For example, two of the participants trained in technical fields and had a significant level of industry experience. Similarly, they had learned about responsible AI through their own initiative.

On the other hand, the job postings heavily focus on experience , including a significant amount of technical know-how, experience focused on ML development, product and program management , and implementation of ethical and social responsibility practices within fast-paced technology companies. The interview participants had been “working in the industry for some time” before taking on these management roles. However, their range of experiences do not cover all of the required experiences outlined in the job descriptions. As expected, excellent communication skills are noted in the job descriptions and strongly echoed by the experts as well. The job postings do not necessarily elaborate on the nature of communication skills; however, the experts note that the “ability to listen” , understand , and sometimes “persuade different stakeholders” is key in such roles.

Interpersonal Qualities Few of the job postings make remarks about attitudes/values and highlight that managers need to value designing technology for social good and cooperation with other stakeholders. A good candidate for management shouldfoster a growth mindset and approach their work with agility, creativity, and passion. All of the participants expressed their passion for developing ethical technologyand indicate that they took a lot of initiative to learn and contribute to the field within their company and externally before they could take on their management roles.


The job descriptions dataset has 4 postings for director positions and 2 of the expert interviewees have directorship roles. According to the job postings, director responsibilities include at least three of the following: a) lead the operationalization of AI ethics principles, b) provide strategic direction and roadmap towards enterprise-wide adoption and application of ethical principles and decision frameworks, and c) build internal capacity for AI ethics practice and governance. Depending on the nature of the organization and its need to incorporate AI ethics practices, these responsibilities vary in scope. For example, a director within a technology start-up will only be able to commit “limited amount of time to operationalizing AI ethics principles and building internal capacity” compared to a director within a larger technology company.

Skills and qualifications According to the job postings, the key skill for being a director is having the ability to build a strong relationship with a broad community that helps define and promote best practice standards of AI ethics. An ideal director can effectively pair their technical skills/know-how with their management skills and policy/standards knowledge to develop strategic plans for the company. Experience in directing and leading teams, particularly in social responsibility practices within technology companies is highly valued for such positions. Only one job posting specifies an educational (a bachelor’s related to policy development and implementation). Others only highlight experience. The two interviewees hold master’s degrees in business and information systems respectively. They also had extensive industry experience that was not directly in AI ethics. However, their experience involved “translation of policy within a technology application” .

Interpersonal Qualities As expected, according to the job postings a good candidate for directorship needs to have exceptional written and verbal communication skills, need to be able “to articulate complex ideas” to technical and non-technical audiences, “engage and influence stakeholders” and “collaborate with people from different disciplines, and cultures” . This set of skills was reflected in our expert interviews. Both interviewees emphasized how they maintain a good flow of communication with the employees and how they remain always open to having conversations on a needs basis. This allowed them to build trust within the company and pursue moving forward with their strategic plan. The job postings highlight the ability to earn trust in relationships as a sought-after value for a directorship role. A director should also be able to challenge the status quo, be passionate about good technology development, be comfortable with ambiguity , and adapt rapidly to changing environment and demands . Most importantly, a director needs to have “a strong and clear commitment to the company values” as they set the tone for others within the organization.

Emerging occupations

Besides the abovementioned classes of occupations, we found a few other positions that do not map easily to any of the existing categories. Considering the limited number of these positions, they do not justify a category of their own. However, we note these emerging roles to understand how they might shape up the responsible AI profession. These occupation titles include data ethicists (2 in job postings), AI ethics consultants (2 in interviews), dataset leads (2 in job postings), communication specialist (1 in job postings), safety specialist (1 in job posting) and UX designer (1 in job postings). The following describes the main function of these positions:

  • Data ethicist: manage organizational efforts in operationalizing AI ethics practices through policy and technology development work. This role has similarities to the role of a policy analyst and data scientist.
  • AI ethics consultant: apply their expertise in AI ethics to solve pain points for consulting clients.
  • Dataset lead: curate datasets while accounting for fairness and bias-related issues.
  • Safety specialist: use and test large language model-based systems to identify failures and errors.
  • AI ethics communication specialist: write communication pieces that focus on AI ethics issues.
  • UX designers: design user interfaces with ethics in mind.

Future of the responsible AI profession

Our interview participants shared a variety of responses to the question “what will the future of their job be like?”. Some participants thought that eventually, “everyone in a company will be responsible” for understanding ethical and social issues of AI as part of their job. In this scenario, everyone would need to have the appropriate knowledge and skillset to apply responsible AI practices in their work or at least know when they need to ask for advice from internal or external experts.

On the contrary, many participants expressed that “dedicated roles” need to be recruited. These participants elaborate that recruitment for these roles is and will “continue to be challenging” as it is difficult to find people with interdisciplinary backgrounds and established industry work experience. Many of the managers we interviewed have chosen “to build teams that come from different disciplinary backgrounds” and provide “professional development opportunities” on the job. However, they also described that hiring people into these roles is challenging since corporate leaders are not always willing to invest a lot of resources in AI ethics. This often can lead to “exhaustion and burn-out” for individuals who currently hold these roles - this is especially true for small and medium-sized technology companies. According to participants, this will likely change with a progressive shift in the regulatory landscape.


Educators and employers play a pivotal role in shaping a responsible AI culture. In our efforts to create a competency framework that outlines the range of roles for responsible AI practitioners, we find that such frameworks can not only guide corporate leaders to recruit talent but also help grow their responsible AI capacity.

We find that the ability to work in an interdisciplinary environment, communicate and engage with diverse stakeholder groups, and the aptitude for curiosity and self-learning are consistently highlighted for all of the roles. This emphasizes the need to foster an environment where students and existing employees in different roles are encouraged to adopt interdisciplinary approaches/collaboration and explore responsible AI content.

In this section, we articulate how an interdisciplinary environment can be fostered, the importance of organizational support for responsible AI practitioners, and the need to proactively monitor the rapidly changing occupational demand and landscape for these roles.

Being able to work in an interdisciplinary environment is critical

Our results show that many of the responsible AI practitioners today come from non-traditional, non-linear, and interdisciplinary educational and work backgrounds to their current positions. The educational and work experiences of these participants span a multitude of fields and allowed them to develop a strong set of skills in navigating disciplinary boundaries and understanding problems from diverse perspectives. The participants often described their role as a translator and facilitator between different groups and disciplines within the organization. For instance, they remarked that a concept such as fairness, transparency, or ethically safe has completely different meanings depending on the personal and professional backgrounds of their audience. The participants often needed to translate what these concepts mean across different disciplinary boundaries (i.e. statistics and law).

Notably, while the job postings asked for a diverse array of skills and qualifications from multiple disciplines, those who hold such positions today are often specialized in one or two disciplines. However, they had been exposed to and worked across multiple disciplines in their professional career. The most important asset that our interviewees emphasized was being able to work across disciplinary boundaries. The candidates who successfully hold such positions are_not “ethical unicorn, full stack developers”_. However, they have honed the skills necessary to translate and create solutions to responsible AI issues across multiple disciplines. Building on existing proposal to improve responsible AI practicesand education, we posit that AI team leaders need to pay a special attention to hiring individuals with the capability to create, critique and communicate across multiple disciplines . Consider Furthermore, educators can get inspiration from education models in highly interdisciplinary fields such as healthcare and create curricula/spaces where students work with peers from different academic backgrounds.

Responsible AI practitioners are advocates - but they need organizational support

We find that responsible AI practitioners are often highly driven and motivated to make a positive impact. These individuals often hold a strong sense of valuing social justice and want to ensure that AI technology is developed in a way that is good for society’s well-being. One of the most consistent ideas that came through in the interviews is the attitude that the participants had toward their careers. Many of the interview participants took the time to immerse themselves in learning new topics and expressed that they were self-motivated to do so. This is especially true for the individuals who are taking some of these first positions in the industry. When looking at the career trajectory of many of the participants, we observe that they often created their own roles or came into a newly created role. Moreover, these individuals often needed to start their own projects and create relationships with others in the organization to measure their own progress and establish credibility.

Similar to any emerging profession many of the participants act as champions for ethical and safe development of AI. They are often working in an environment that questions and challenges the need for considering AI ethics principles. As some of the participants remarked, they often have to answer questions such as “why do we need to pay for ethics assessments?” ,“what is the value of considering AI ethics in a start-up?” , or “why should we put in the time? what is the value added?” . This act of advocating for AI ethics is even more challenging when existing regulations do not have proper enforcement mechanisms for responsible AI practices. Many of the participants assume the role of an advocate and often use their excellent communication skills to build relationships and capacity within their organization.

For the successful implementation of responsible AI practices, it is important that business leaders pay attention and support the advocacy efforts of these practitioners. Many of today’s responsible AI practitioners are working with limited resources, have critical responsibilities, and are experiencing burn-out. Whenever possible, leaders in AI companies need to create appropriate incentive structures, provide the necessary resources and communicate the value of establishing responsible AI practices to their employees so that these practitioners have the necessary support for the effective execution of their responsibilities. Recognizing the nature of these roles, educators can learn from existing methodsand integrate leadership training into their curricula when addressing responsible AI-related content.

Educators and employers need to monitor and plan for the rapidly changing landscape of responsible AI roles

The nature of occupations in the AI industry is continually growing and shifting. The rapid technological development, upcoming regulationsand global economic conditionsimpact how companies recruit and retain responsible AI expertise. Furthermore, there is a need for new educational efforts and programs for preparing new graduates to take on responsible AI practices. The proposed ontology provides a synthesis of roles that have emerged in responsible AI practice and it can serve as a planning tool for corporate leaders and educators.

Corporate leaders can use this ontology to build internal capacity for individuals who currently hold researcher, data scientist, engineer, policy advisor, manager, and director roles in their institutions. Depending on these companies’ responsible AI needs and resources, business executives can work towards creating interdisciplinary teams for establishing responsible AI practice by recruiting individuals with the competencies outlined for each of these roles. Besides recruiting and fostering for responsible AI competencies, these leaders need to communicate the importance of these practices and start by creating the appropriate organizational incentives and resources for adapting responsible AI practices. Government and non-governmental organizations could support such efforts, particularly small and medium-size companies, by formally recognizing such roles in their taxonomies of occupationsand providing resources.

Current computer science and engineering education focuses primarily on teaching professional ethics. There is minimal focus and resources on cultivating skills and knowledge required for cultivating the skills that focus on ethics in design. On the other hand, there is a lack of clarity of how much students in social and political sciences need to work on their technical acumen to become skilled responsible AI practitioners. Educators could use the list of competencies to develop a set of learning objectives and examine the efficacy of different teaching pedagogies in supporting these objectives. Moreover, Educators can use the competency framework as a tool for acquiring resources for further curricula and program development.

Notably, the proposed ontology primarily focuses on type of roles, responsibilities and skills without addressing other important factors in recruitment and education efforts such as diversity of individual who get to learn about responsible AI issues or take such roles in the industry. Therefore, it is critical that users of this ontology, consider factors that are not captured in the scope of this ontology. Furthermore, considering the rapidly changing conversation around responsible AI practices, the type of roles in this onotlogy will shift and expand. We invite the community of researchers , practitioners and educators to reflect on these roles and build on this ontology.


With the increased regulatory activities in the industry, companies have the incentive to ensure responsible AI development. In this work, we found seven different type of roles and their corresponding responsibilities, skills, qualifications, and interpersonal qualities expected in today’s responsible AI practitioner. We propose a preliminary competency framework for responsible AI practitioners and highlight the importance of creating interdisciplinary teams and providing adequate organizational support for individuals in these roles.

We thank our study participants for taking the time to share their experiences, expertise, and feedback. We also thank our anonymous reviewers for their deep engagement and valuable feedback on this paper. This work benefitted greatly from the data collection and analysis assistance from our collaborators Sandi Mak, Ivan Ivanov, Aidan Doudeau, and Nandita Jayd at Vanier College, Montreal. We are grateful for their contributions. Finally, this work was financially supported by the Natural Sciences and Engineering Research Council of Canada and Pôle montréalais d’enseignement supérieur en intelligence artificielle.

Supplementary Material

Interview protocol

Consent Process Thank you for reading and signing the Human Subjects Consent Form for this project..

Introduction Thank you for agreeing to take part in this study. My name is [interviewer] and I will be conducting this interview. I am a research assistant working with [advisor]. We have invited you to take part in this today because of your current role. The purpose of this study is to examine the experiences of professionals including ethicists, technologists and business leaders who are dealing with ethical and social implications of particular AI technology through development and implementation.

Today I am playing two roles: that of a interviewer and that of a researcher. At this time I would like to give a brief overview of the project and the consent form. [5min]

Background and current role

  • Please tell us about your role at your current company.
  • What is your official job title, and what are you main responsibilities?
  • Could you please tell me about your background, expertise and experience that led you to take on your current role?
  • Who do you work most closely with at your company? Who do you manage?Who do you report to? Who are clients? Who are your partners?

Situation your work, projects in AI ethics

  • How do you situate your work within the broader field of AI ethics? What types of challenges are you working on in this field? Please feel free to share any specific examples from your projects.
  • What are the main projects related to AI ethics that you are working on that you can tell us about?
  • What types of resources (academic papers, academic experts, standards, guidelines) do you use in your AI ethics practice?
  • Do you use the guidelines on AI ethics practice that have published in this field over the past 5 years? If so, which one and what does following the guideline look like at your company?
  • What are the most important challenges to implementing ethics principles at your work?

Skills, knowledge, values

  • What are the most important skillsets, knowledge base and values that you currently use at your job? What are you currently developing and will need in the future?
  • If you decide to hire someone to replace you in your current role, what would you look for? What skills or background would your ideal candidate have?

Looking into the future

  • From your perspective, what roles do you think are necessary in the field of AI ethics in academia, industry, governmental and non-governmental organizations? Please elaborate.

Relevant tables

Coding scheme for Job Posting Analysis

Table Label: tab-jp-coding-full

Download PDF to view table

Inclusion and exclusion criteria for job postings

Table Label: tab-jp-criteria

Download PDF to view table

Competency framework for AI ethics

Table Label: tab-framework

Download PDF to view table


   2  note = {},
   3  url = {},
   4  doi = {10.1145/1188913.1188915},
   5  pages = {36--44},
   6  year = {2007},
   7  month = {January},
   8  number = {1},
   9  volume = {50},
  10  journal = {Communications of the ACM},
  11  title = {The patent holder's dilemma: Buy, sell, or troll?},
  12  author = {Patricia S. Abril and Robert Plant},
  16  acmid = {1219093},
  17  url = {},
  18  doi = {10.1145/1219092.1219093},
  19  year = {2007},
  20  month = {April},
  21  number = {2},
  22  volume = {54},
  23  numpages = {50},
  24  articleno = {5},
  25  journal = {J. ACM},
  26  title = {Deciding equivalances among conjunctive aggregate queries},
  27  author = {Sarah Cohen and Werner Nutt and Yehoshua Sagic},
  31  note = {},
  32  month = {},
  33  series = {},
  34  number = {},
  35  volume = {},
  36  editor = {},
  37  edition = {2nd.},
  38  address = {New York, NY},
  39  year = {2001},
  40  publisher = {Wiley},
  41  title = {Understanding Policy-Based Networking},
  42  author = {David Kosiur},
  46  note = {},
  47  month = {},
  48  number = {},
  49  editor = {},
  50  url = {},
  51  doi = {10.1007/3-540-09237-4},
  52  publisher = {Springer-Verlag},
  53  address = {New York, NY},
  54  volume = {68},
  55  series = {Lecture Notes in Computer Science},
  56  title = {First-Order Dynamic Logic},
  57  year = {1979},
  58  author = {David Harel},
  62  note = {},
  63  month = {},
  64  type = {},
  65  number = {},
  66  pages = {},
  67  chapter = {},
  68  url = {},
  69  doi = {10.1007/3-540-09237-4},
  70  publisher = {University of Chicago Press},
  71  edition = {1st.},
  72  address = {Chicago},
  73  volume = {9},
  74  year = {2007},
  75  series = {The name of the series one},
  76  subtitle = {The book subtitle},
  77  title = {The title of book one},
  78  editor = {Ian Editor},
  79  author = {},
  83  note = {},
  84  month = {},
  85  type = {},
  86  number = {},
  87  pages = {},
  88  chapter = {100},
  89  volume = {},
  90  url = {},
  91  doi = {10.1007/3-540-09237-4},
  92  publisher = {University of Chicago Press},
  93  edition = {2nd.},
  94  address = {Chicago},
  95  year = {2008},
  96  series = {The name of the series two},
  97  subtitle = {The book subtitle},
  98  title = {The title of book two},
  99  editor = {Ian Editor},
 100  author = {},
 104  note = {},
 105  month = {},
 106  type = {},
 107  series = {},
 108  number = {},
 109  volume = {},
 110  url = {},
 111  doi = {10.1145/90417.90738},
 112  pages = {19--33},
 113  editor = {Sape Mullender},
 114  chapter = {},
 115  edition = {2nd.},
 116  year = {1990},
 117  address = {New York, NY},
 118  publisher = {ACM Press},
 119  booktitle = {Distributed Systems},
 120  title = {Achieving application requirements},
 121  author = {Asad Z. Spector},
 125  note = {},
 126  month = {},
 127  type = {},
 128  number = {},
 129  edition = {},
 130  url = {},
 131  doi = {10.1007/3-540-65193-4_29},
 132  pages = {368--394},
 133  editor = {Grzegorz Rozenberg and Frits W. Vaandrager},
 134  chapter = {},
 135  year = {1998},
 136  volume = {1494},
 137  address = {London},
 138  publisher = {Springer-Verlag},
 139  booktitle = {Lectures on Embedded Systems},
 140  series = {Lecture Notes in Computer Science},
 141  title = {Statecarts in use: structured analysis and object-orientation},
 142  author = {Bruce P. Douglass and David Harel and Mark B. Trakhtenbrot},
 146  note = {},
 147  month = {},
 148  series = {},
 149  number = {},
 150  volume = {},
 151  editor = {},
 152  edition = {},
 153  address = {},
 154  year = {1997},
 155  publisher = {Addison Wesley Longman Publishing Co., Inc.},
 156  title = {The Art of Computer Programming, Vol. 1: Fundamental Algorithms (3rd. ed.)},
 157  author = {Donald E. Knuth},
 161  note = {(book)},
 162  month = {},
 163  number = {},
 164  editor = {},
 165  url = {},
 166  doi = {},
 167  publisher = {Addison Wesley Longman Publishing Co., Inc.},
 168  address = {},
 169  edition = {3rd},
 170  volume = {1},
 171  series = {Fundamental Algorithms},
 172  title = {The Art of Computer Programming},
 173  year = {1998},
 174  author = {Donald E. Knuth},
 178  editors = {Z. Ghahramani and R. Cowell},
 179  month = {January},
 180  publisher = {The Society for Artificial Intelligence and Statistics},
 181  booktitle = {Proceedings of Tenth International Workshop on Artificial Intelligence and Statistics, {\rm The Barbados}},
 182  year = {2005},
 183  title = {Structured Variational Inference Procedures and their Realizations (as incol)},
 184  author = {Dan Geiger and Christopher Meek},
 188  note = {},
 189  organization = {},
 190  month = {},
 191  number = {},
 192  url = {},
 193  doi = {99.9999/woot07-S422},
 194  pages = {422--431},
 195  address = {Milan Italy},
 196  publisher = {Paparazzi Press},
 197  year = {2010},
 198  volume = {3},
 199  editor = {Reginald N. Smythe and Alexander Noble},
 200  series = {LAC '10},
 201  booktitle = {Proceedings of the 3rd. annual workshop on Librarians and Computers},
 202  title = {An experiment in bibliographic mark-up: Parsing metadata for XML export},
 203  author = {Stan W. Smith},
 207  numpages = {9},
 208  articleno = {Paper 7},
 209  address = {Berkley, CA},
 210  publisher = {USENIX Association},
 211  series = {WOOT '07},
 212  booktitle = {Proceedings of the first USENIX workshop on Offensive Technologies},
 213  title = {Catch me, if you can: Evading network signatures with web-based polymorphic worms},
 214  year = {2007},
 215  author = {Matthew Van Gundy and Davide Balzarotti and Giovanni Vigna},
 219  pages = {99-100},
 220  numpages = {2},
 221  articleno = {7},
 222  address = {Berkley, CA},
 223  publisher = {USENIX Association},
 224  series = {WOOT '08},
 225  booktitle = {Proceedings of the first USENIX workshop on Offensive Technologies},
 226  title = {Catch me, if you can: Evading network signatures with web-based polymorphic worms},
 227  year = {2008},
 228  author = {Matthew Van Gundy and Davide Balzarotti and Giovanni Vigna},
 232  pages = {90--100},
 233  address = {Berkley, CA},
 234  publisher = {USENIX Association},
 235  series = {WOOT '09},
 236  booktitle = {Proceedings of the first USENIX workshop on Offensive Technologies},
 237  title = {Catch me, if you can: Evading network signatures with web-based polymorphic worms},
 238  year = {2009},
 239  author = {Matthew Van Gundy and Davide Balzarotti and Giovanni Vigna},
 243  note = {},
 244  organization = {},
 245  month = {},
 246  number = {},
 247  volume = {},
 248  editor = {},
 249  url = {},
 250  doi = {10.1145/567752.567774},
 251  pages = {226--236},
 252  address = {New York, NY},
 253  publisher = {ACM Press},
 254  year = {1979},
 255  series = {POPL '79},
 256  booktitle = {Proceedings of the 6th. ACM SIGACT-SIGPLAN symposium on Principles of Programming Languages},
 257  title = {Predicate Path expressions},
 258  author = {Sten Andler},
 262  note = {},
 263  month = {},
 264  address = {Cambridge, MA},
 265  number = {TR-200},
 266  type = {MIT Research Lab Technical Report},
 267  institution = {Massachusetts Institute of Technology},
 268  title = {LOGICS of Programs: AXIOMATICS and DESCRIPTIVE POWER},
 269  year = {1978},
 270  author = {David Harel},
 274  year = {2003},
 275  intitution = {FOI-R-0961-SE, Swedish Defence Research Agency (FOI)},
 276  school = {Royal Institute of Technology (KTH), Stockholm, Sweden},
 277  title = {Optimal Motion Control of a Ground Vehicle},
 278  author = {David A. Anisi},
 282  month = {},
 283  type = {},
 284  note = {UMI Order Number: AAT 8506171},
 285  address = {Palo Alto, CA},
 286  school = {Stanford University},
 287  title = {Algorithms for Closest-Point Problems (Computational Geometry)},
 288  year = {1985},
 289  author = {Kenneth L. Clarkson},
 293  url = {},
 294  lastaccessed = {June 7, 2006},
 295  title = {Stats and Analysis},
 296  month = {March},
 297  year = {2006},
 298  author = {Poker-Edge.Com},
 302  note = {},
 303  lastaccessed = {March 21, 2008},
 304  month = {March},
 305  url = {},
 306  day = {5},
 307  howpublished = {Video},
 308  title = {A more perfect union},
 309  year = {2008},
 310  author = {Barack Obama},
 314  lastaccessed = {},
 315  month = {August},
 316  howpublished = {},
 317  url = {},
 318  note = {Patent No. 12345, Filed July 1st., 2008, Issued Aug. 9th., 2009},
 319  title = {The fountain of youth},
 320  year = {2009},
 321  author = {Joseph Scientist},
 325  distincturl = {1},
 326  organization = {},
 327  series = {},
 328  number = {},
 329  volume = {},
 330  editor = {},
 331  howpublished = {Video},
 332  note = {},
 333  url = {},
 334  doi = {99.9999/woot07-S422},
 335  month = {March 21, 2008},
 336  pages = {4},
 337  address = {New York, NY},
 338  publisher = {ACM Press},
 339  year = {2003},
 340  booktitle = {ACM SIGGRAPH 2003 Video Review on Animation theater Program: Part I - Vol. 145 (July 27--27, 2003)},
 341  title = {Solder man},
 342  author = {Dave Novak},
 346  note = {},
 347  howpublished = {Video},
 348  url = {},
 349  doi = {10.1145/1057270.1057278},
 350  month = {Jan.-March},
 351  number = {1},
 352  volume = {3},
 353  eid = {4},
 354  journal = {Comput. Entertain.},
 355  title = {Interview with Bill Kinder: January 13, 2005},
 356  year = {2005},
 357  author = {Newton Lee},
 361  note = {To appear},
 362  howpublished = {},
 363  url = {},
 364  doi = {},
 365  articleno = {Article~5},
 366  month = {July},
 367  number = {3},
 368  volume = {12},
 369  journal = {Digital Libraries},
 370  title = {The Enabling of Digital Libraries},
 371  year = {2008},
 372  author = {Bernard Rous},
 376  address = {New York, NY, USA},
 377  publisher = {ACM},
 378  doi = {},
 379  pages = {11},
 380  issn = {1084-6654},
 381  year = {2000},
 382  volume = {5},
 383  journal = {J. Exp. Algorithmics},
 384  title = {(old) Finding minimum congestion spanning trees},
 385  author = {Werneck,, Renato and Setubal,, Jo\~{a}o and da Conceic\~{a}o,, Arlindo},
 389  address = {New York, NY, USA},
 390  publisher = {ACM},
 391  acmid = {384253},
 392  doi = {10.1145/351827.384253},
 393  url = {},
 394  articleno = {11},
 395  issn = {1084-6654},
 396  year = {2000},
 397  month = {December},
 398  volume = {5},
 399  journal = {J. Exp. Algorithmics},
 400  title = {(new) Finding minimum congestion spanning trees},
 401  author = {Werneck, Renato and Setubal, Jo\~{a}o and da Conceic\~{a}o, Arlindo},
 405  address = {Amsterdam, The Netherlands, The Netherlands},
 406  publisher = {Elsevier Science Publishers B. V.},
 407  doi = {},
 408  pages = {342--353},
 409  issn = {1566-2535},
 410  year = {2009},
 411  number = {4},
 412  volume = {10},
 413  journal = {Inf. Fusion},
 414  title = {(old) Distributed data source verification in wireless sensor networks},
 415  author = {Conti, Mauro and Di Pietro, Roberto and Mancini, Luigi V. and Mei, Alessandro},
 419  keywords = {Clone detection, Distributed protocol, Securing data fusion, Wireless sensor networks},
 420  address = {Amsterdam, The Netherlands, The Netherlands},
 421  publisher = {Elsevier Science Publishers B. V.},
 422  acmid = {1555162},
 423  doi = {10.1016/j.inffus.2009.01.002},
 424  url = {},
 425  numpages = {12},
 426  pages = {342--353},
 427  issn = {1566-2535},
 428  year = {2009},
 429  month = {October},
 430  number = {4},
 431  volume = {10},
 432  journal = {Inf. Fusion},
 433  title = {(new) Distributed data source verification in wireless sensor networks},
 434  author = {Conti, Mauro and Di Pietro, Roberto and Mancini, Luigi V. and Mei, Alessandro},
 438  keywords = {cscw, distributed knowledge acquisition, incentive design, online games, recommender systems, reputation systems, user studies, virtual community},
 439  address = {New York, NY, USA},
 440  publisher = {ACM},
 441  acmid = {1358946},
 442  doi = {10.1145/1358628.1358946},
 443  url = {},
 444  numpages = {6},
 445  pages = {3873--3878},
 446  location = {Florence, Italy},
 447  isbn = {978-1-60558-012-X},
 448  year = {2008},
 449  booktitle = {CHI '08 extended abstracts on Human factors in computing systems},
 450  title = {Portalis: using competitive online interactions to support aid initiatives for the homeless},
 451  author = {Li, Cheng-Lun and Buyuktur, Ayse G. and Hutchful, David K. and Sant, Natasha B. and Nainwal, Satyendra K.},
 455  address = {Upper Saddle River, NJ, USA},
 456  publisher = {Prentice Hall PTR},
 457  edition = {1st},
 458  isbn = {0130850845},
 459  year = {1999},
 460  title = {Visual Basic 6: Design, Specification, and Objects with Other},
 461  author = {Hollis, Billy S.},
 465  address = {Boston, MA, USA},
 466  publisher = {Addison-Wesley Longman Publishing Co., Inc.},
 467  edition = {1st},
 468  isbn = {0201433117},
 469  year = {1999},
 470  title = {The  Latex Web Companion: Integrating TEX, HTML, and XML},
 471  author = {Goossens, Michel and Rahtz, S. P. and Moore, Ross and Sutor, Robert S.},
 475  address = {Amherst, MA, USA},
 476  publisher = {University of Massachusetts},
 477  source = {\&id=oai%3Ancstrlh%3Aumass_cs%3Ancstrl.umassa_cs%2F%2FUM-CS-1987-018},
 478  year = {1987},
 479  title = {Vertex Types in Book-Embeddings},
 480  author = {Buss, Jonathan F. and Rosenberg, Arnold L. and Knott, Judson D.},
 484  address = {Amherst, MA, USA},
 485  publisher = {University of Massachusetts},
 486  source = {\&id=oai%3Ancstrlh%3Aumass_cs%3Ancstrl.umassa_cs%2F%2FUM-CS-1987-018},
 487  year = {1987},
 488  title = {Vertex Types in Book-Embeddings},
 489  author = {Buss, Jonathan F. and Rosenberg, Arnold L. and Knott, Judson D.},
 493  address = {New York, NY, USA},
 494  publisher = {ACM},
 495  order_no = {608085},
 496  location = {Florence, Italy},
 497  isbn = {978-1-60558-012-X},
 498  year = {2008},
 499  title = {CHI '08: CHI '08 extended abstracts on Human factors in computing systems},
 500  note = {General Chair-Czerwinski, Mary and General Chair-Lund, Arnie and Program Chair-Tan, Desney},
 501  author = {},
 505  address = {Stanford, CA, USA},
 506  school = {Stanford University},
 507  note = {AAT 8506171},
 508  year = {1985},
 509  title = {Algorithms for Closest-Point Problems (Computational Geometry)},
 510  advisor = {Yao, Andrew C.},
 511  author = {Clarkson, Kenneth Lee},
 515  address = {New York, NY, USA},
 516  publisher = {ACM},
 517  issue_date = {January/April 1984},
 518  number = {5-1},
 519  volume = {13-14},
 520  issn = {0146-4833},
 521  year = {1984},
 522  journal = {SIGCOMM Comput. Commun. Rev.},
 523  key = {{$\!\!$}},
 527  address = {Washington, DC, USA},
 528  publisher = {IEEE Computer Society},
 529  acmid = {1010128},
 530  doi = {},
 531  url = {},
 532  pages = {21--22},
 533  isbn = {0-7695-2167-3},
 534  year = {2004},
 535  series = {ICWS '04},
 536  booktitle = {Proceedings of the IEEE International Conference on Web Services},
 537  title = {IEEE TCSC Executive Committee},
 538  key = {IEEE},
 542  address = {New York, NY, USA},
 543  publisher = {ACM Press/Addison-Wesley Publishing Co.},
 544  isbn = {0-201-62427-3},
 545  year = {1993},
 546  title = {Distributed systems (2nd Ed.)},
 547  editor = {Mullender, Sape},
 551  address = {Austin, TX, USA},
 552  publisher = {University of Texas at Austin},
 553  source = {\&id=oai%3Ancstrlh%3Autexas_cs%3AUTEXAS_CS%2F%2FAI86-33},
 554  year = {1986},
 555  title = {New Algorithms for Dependency-Directed Backtracking (Master's thesis)},
 556  author = {Petrie, Charles J.},
 560  address = {Austin, TX, USA},
 561  school = {University of Texas at Austin},
 562  source = {\&id=oai%3Ancstrlh%3Autexas_cs%3AUTEXAS_CS%2F%2FAI86-33},
 563  year = {1986},
 564  title = {New Algorithms for Dependency-Directed Backtracking (Master's thesis)},
 565  author = {Petrie, Charles J.},
 569  year = {1981},
 570  publisher = {Addison-Wesley},
 571  title = {Seminumerical Algorithms},
 572  author = {Donald E. Knuth},
 576  address = {Hershey, PA, USA},
 577  publisher = {IGI Publishing},
 578  acmid = {887010},
 579  url = {},
 580  numpages = {24},
 581  pages = {51--74},
 582  isbn = {1-59140-056-2},
 583  year = {2001},
 584  booktitle = {E-commerce and cultural values},
 585  title = {The implementation of electronic commerce in SMEs in Singapore (as Incoll)},
 586  author = {Kong, Wei-Chang},
 590  address = {Hershey, PA, USA},
 591  publisher = {IGI Publishing},
 592  acmid = {887010},
 593  url = {},
 594  numpages = {24},
 595  pages = {51--74},
 596  isbn = {1-59140-056-2},
 597  year = {2001},
 598  title = {E-commerce and cultural values},
 599  chapter = {The implementation of electronic commerce in SMEs in Singapore (Inbook-w-chap-w-type)},
 600  type = {Name of Chapter:},
 601  author = {Kong, Wei-Chang},
 605  year = {2020},
 606  url = {},
 607  title = {{Toward Trustworthy AI Development: Mechanisms for Supporting Verifiable Claims}},
 608  mendeley-groups = {PhD/AI competency project},
 609  file = {:C$\backslash$:/Users/shala/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Brundage et al. - 2020 - Toward Trustworthy AI Development Mechanisms for Supporting Verifiable Claims.pdf:pdf},
 610  eprint = {2004.07213},
 611  author = {Brundage, Miles and Avin, Shahar and Wang, Jasmine and Belfield, Haydn and Krueger, Gretchen and Hadfield, Gillian and Khlaaf, Heidy and Yang, Jingying and Toner, Helen and Fong, Ruth and Maharaj, Tegan and Koh, Pang Wei and Hooker, Sara and Leung, Jade and Trask, Andrew and Bluemke, Emma and Lebensold, Jonathan and O'Keefe, Cullen and Koren, Mark and Ryffel, Th{\'{e}}o and Rubinovitz, JB and Besiroglu, Tamay and Carugati, Federica and Clark, Jack and Eckersley, Peter and de Haas, Sarah and Johnson, Maritza and Laurie, Ben and Ingerman, Alex and Krawczuk, Igor and Askell, Amanda and Cammarota, Rosario and Lohn, Andrew and Krueger, David and Stix, Charlotte and Henderson, Peter and Graham, Logan and Prunkl, Carina and Martin, Bianca and Seger, Elizabeth and Zilberman, Noa and H{\'{E}}igeartaigh, Se{\'{a}}n {\'{O}} and Kroeger, Frens and Sastry, Girish and Kagan, Rebecca and Weller, Adrian and Tse, Brian and Barnes, Elizabeth and Dafoe, Allan and Scharre, Paul and Herbert-Voss, Ariel and Rasser, Martijn and Sodhani, Shagun and Flynn, Carrick and Gilbert, Thomas Krendl and Dyer, Lisa and Khan, Saif and Bengio, Yoshua and Anderljung, Markus},
 612  arxivid = {2004.07213},
 613  archiveprefix = {arXiv},
 614  abstract = {With the recent wave of progress in artificial intelligence (AI) has come a growing awareness of the large-scale impacts of AI systems, and recognition that existing regulations and norms in industry and academia are insufficient to ensure responsible AI development. In order for AI developers to earn trust from system users, customers, civil society, governments, and other stakeholders that they are building AI responsibly, they will need to make verifiable claims to which they can be held accountable. Those outside of a given organization also need effective means of scrutinizing such claims. This report suggests various steps that different stakeholders can take to improve the verifiability of claims made about AI systems and their associated development processes, with a focus on providing evidence about the safety, security, fairness, and privacy protection of AI systems. We analyze ten mechanisms for this purpose--spanning institutions, software, and hardware--and make recommendations aimed at implementing, exploring, or improving those mechanisms.},
 618  year = {2019},
 619  title = {{Improving fairness in machine learning systems: What do industry practitioners need?}},
 620  mendeley-groups = {PhD/AI competency project},
 621  keywords = {Algorithmic bias,Empirical study,Fair machine learning,Need-finding,Product teams,UX of machine learning},
 622  isbn = {9781450359702},
 623  file = {:G$\backslash$:/My Drive/PhD/AI Competency Project/Literature Review/improving fairness in ML systems - what do industry practitioners need.pdf:pdf},
 624  eprint = {1812.05239},
 625  doi = {10.1145/3290605.3300830},
 626  booktitle = {Conf. Hum. Factors Comput. Syst. - Proc.},
 627  author = {Holstein, Kenneth and Vaughan, Jennifer Wortman and Daum{\'{e}}, Hal and Dud{\'{i}}k, Miroslav and Wallach, Hanna},
 628  arxivid = {1812.05239},
 629  archiveprefix = {arXiv},
 630  abstract = {The potential for machine learning (ML) systems to amplify social inequities and unfairness is receiving increasing popular and academic attention. A surge of recent work has focused on the development of algorithmic tools to assess and mitigate such unfairness. If these tools are to have a positive impact on industry practice, however, it is crucial that their design be informed by an understanding of real-world needs. Through 35 semi-structured interviews and an anonymous survey of 267 ML practitioners, we conduct the first systematic investigation of commercial product teams' challenges and needs for support in developing fairer ML systems. We identify areas of alignment and disconnect between the challenges faced by teams in practice and the solutions proposed in the fair ML research literature. Based on these findings, we highlight directions for future ML and HCI research that will better address practitioners' needs.},
 634  year = {2020},
 635  title = {{Activism by the AI Community}},
 636  pages = {15--21},
 637  mendeley-groups = {PhD/AI competency project},
 638  isbn = {9781450371100},
 639  file = {:G$\backslash$:/My Drive/PhD/AI Competency Project/Literature Review/activitist by AI ethics community.pdf:pdf},
 640  doi = {10.1145/3375627.3375814},
 641  booktitle = {AIES 2019 - Proc. 2019 AAAI/ACM Conf. AI, Ethics, Soc.},
 642  author = {Belfield, Haydn},
 643  abstract = {The artificial intelligence (AI) community has recently engaged in activism in relation to their employers, other members of the community, and their governments in order to shape the societal and ethical implications of AI. It has achieved some notable successes, but prospects for further political organising and activism are uncertain. We survey activism by the AI community over the last six years; apply two analytical frameworks drawing upon the literature on epistemic communities, and worker organising and bargaining; and explore what they imply for the future prospects of the AI community. Success thus far has hinged on a coherent shared culture, and high bargaining power due to the high demand for a limited supply of AI 'talent'. Both are crucial to the future of AI activism and worthy of sustained attention.},
 647  year = {2022},
 648  title = {{AI Ethics Training in Higher Education : Competency Framework}},
 649  number = {February},
 650  mendeley-groups = {PhD/AI competency project},
 651  file = {:G$\backslash$:/My Drive/PhD/AI Competency Project/Literature Review/22949-Cegep-Andre-Laurendeau-Referentiel-de-competence-V4-FINAL-Interactif (1).pdf:pdf},
 652  author = {Bruneault, Fr{\'{e}}d{\'{e}}rick and Andr{\'{e}}-laurendeau, C{\'{e}}gep and Laflamme, Andr{\'{e}}ane Sabourin and Fillion, Gabrielle and Abtroun, Neila and Freeman, Andrew},
 656  year = {2017},
 657  url = {{\_}paths/edison{\_}cf-ds-release2-v08{\_}0.pdf},
 658  title = {{EDISON Data Science Framework: Part1. Data Science Competence Framework Release 2}},
 659  pages = {1--59},
 660  number = {July},
 661  mendeley-groups = {PhD/AI competency project},
 662  file = {:G$\backslash$:/My Drive/PhD/AI Competency Project/Literature Review/edison{\_}cf-ds-release2-v08{\_}0.pdf:pdf},
 663  author = {{Yuri Demchenko, Adam Belloum}, Tomasz Wiktorski},
 667  year = {2021},
 668  title = {{Artificial Intelligence Competency Framework Table of Contents}},
 669  number = {September},
 670  mendeley-groups = {PhD/AI competency project},
 671  file = {:G$\backslash$:/My Drive/PhD/AI Competency Project/Literature Review/Corrected-FINAL{\_}PIA{\_}ConcordiaDawson{\_}AICompetencyFramework.pdf:pdf},
 672  author = {Blok, Sherry and Trudeau, Joel and Cassidy, Robert},
 676  year = {2021},
 677  volume = {5},
 678  title = {{Where Responsible AI meets Reality: Practitioner Perspectives on Enablers for Shifting Organizational Practices}},
 679  pages = {1--23},
 680  number = {CSCW1},
 681  mendeley-groups = {PhD/AI competency project},
 682  keywords = {industry practice,organizational structure,responsible ai},
 683  journal = {Proc. ACM Human-Computer Interact.},
 684  issn = {25730142},
 685  file = {:G$\backslash$:/My Drive/PhD/AI Competency Project/Literature Review/Where Responsible AI meets Reality - Practitioner's Perspective.pdf:pdf},
 686  eprint = {2006.12358},
 687  doi = {10.1145/3449081},
 688  author = {Rakova, Bogdana and Yang, Jingying and Cramer, Henriette and Chowdhury, Rumman},
 689  arxivid = {2006.12358},
 690  archiveprefix = {arXiv},
 691  abstract = {Large and ever-evolving technology companies continue to invest more time and resources to incorporate responsible Artificial Intelligence (AI) into production-ready systems to increase algorithmic accountability. This paper examines and seeks to offer a framework for analyzing how organizational culture and structure impact the effectiveness of responsible AI initiatives in practice. We present the results of semi-structured qualitative interviews with practitioners working in industry, investigating common challenges, ethical tensions, and effective enablers for responsible AI initiatives. Focusing on major companies developing or utilizing AI, we have mapped what organizational structures currently support or hinder responsible AI initiatives, what aspirational future processes and structures would best enable effective initiatives, and what key elements comprise the transition from current work practices to the aspirational future.},
 695  year = {2020},
 696  volume = {1},
 697  url = {},
 698  title = {{Brave: what it means to be an AI Ethicist}},
 699  publisher = {Springer International Publishing},
 700  pages = {87--91},
 701  number = {1},
 702  mendeley-groups = {PhD/AI competency project},
 703  keywords = {AI Ethics,Artificial intelligence,Bravery,Ethical decision making,ai ethics,artificial intelligence,bravery,ethical decision making},
 704  journal = {AI Ethics},
 705  issn = {2730-5953},
 706  isbn = {0123456789},
 707  file = {:G$\backslash$:/My Drive/PhD/AI Competency Project/Literature Review/Gambelin2020{\_}Article{\_}BraveWhatItMeansToBeAnAIEthici.pdf:pdf},
 708  doi = {10.1007/s43681-020-00020-5},
 709  author = {Gambelin, Olivia},
 710  abstract = {Despite there being a strong call for responsible technology, the path towards putting ethics into action is still yet to be fully understood. To help guide the implementation of ethics, we have seen the rise of a new professional title; the AI Ethicist. However, it is still unclear what the role and skill set of this new profession must include. The purpose of this piece is to offer a preliminary definition of what it means to be an AI Ethicist by first examining the concept of an ethicist in the context of artificial intelligence, followed by exploring what responsibilities are added to the role in industry specifically, and ending on the fundamental characteristic that underlies it all: bravery.},
 714  year = {2021},
 715  title = {{"you can't sit with us": Exclusionary pedagogy in AI ethics education}},
 716  pages = {515--525},
 717  mendeley-groups = {PhD/AI competency project},
 718  isbn = {9781450383097},
 719  file = {:G$\backslash$:/My Drive/PhD/AI Competency Project/Literature Review/Raji-pedagogy2021.pdf:pdf},
 720  doi = {10.1145/3442188.3445914},
 721  booktitle = {FAccT 2021 - Proc. 2021 ACM Conf. Fairness, Accountability, Transpar.},
 722  author = {Raji, Inioluwa Deborah and Scheuerman, Morgan Klaus and Amironesei, Razvan},
 723  abstract = {Given a growing concern about the lack of ethical consideration in the Artificial Intelligence (AI) field, many have begun to question how dominant approaches to the disciplinary education of computer science (CS) - -and its implications for AI - -has led to the current "ethics crisis". However, we claim that the current AI ethics education space relies on a form of "exclusionary pedagogy,"where ethics is distilled for computational approaches, but there is no deeper epistemological engagement with other ways of knowing that would benefit ethical thinking or an acknowledgement of the limitations of uni-vocal computational thinking. This results in indifference, devaluation, and a lack of mutual support between CS and humanistic social science (HSS), elevating the myth of technologists as "ethical unicorns"that can do it all, though their disciplinary tools are ultimately limited. Through an analysis of computer science education literature and a review of college-level course syllabi in AI ethics, we discuss the limitations of the epistemological assumptions and hierarchies of knowledge which dictate current attempts at including ethics education in CS training and explore evidence for the practical mechanisms through which this exclusion occurs. We then propose a shift towards a substantively collaborative, holistic, and ethically generative pedagogy in AI education.},
 727  language = {en},
 728  note = {Accessed: 2023-3-15},
 729  howpublished = {\url{}},
 730  year = {2023},
 731  author = {ORCAA Consulting},
 732  title = {ORCAA},
 736  language = {en},
 737  note = {Accessed: 2023-3-15},
 738  howpublished = {\url{}},
 739  year = {2019},
 740  month = {December},
 741  author = {AI Ethics Lab},
 742  title = {AI Ethics Lab},
 746  language = {en},
 747  note = {Accessed: 2023-3-15},
 748  howpublished = {\url{}},
 749  year = {2023},
 750  author = {Ethical AI Advisory},
 751  title = {Ethical AI Advisory},
 755  year = {2020},
 756  month = {January},
 757  abstract = {The rapid spread of artificial intelligence (AI) systems has
 758precipitated a rise in ethical and human rights-based frameworks
 759intended to guide the development and use of these technologies.
 760Despite the proliferation of these ``AI principles,'' there has
 761been little scholarly focus on understanding these efforts either
 762individually or as contextualized within an expanding universe of
 763principles with discernible trends.To that end, this white paper
 764and its associated data visualization compare the contents of
 765thirty-six prominent AI principles documents side-by-side. This
 766effort uncovered a growing consensus around eight key thematic
 767trends: privacy, accountability, safety and security,
 768transparency and explainability, fairness and non-discrimination,
 769human control of technology, professional responsibility, and
 770promotion of human values. Underlying this ``normative core,''
 771our analysis examined the forty-seven individual principles that
 772make up the themes, detailing notable similarities and
 773differences in interpretation found across the documents. In
 774sharing these observations, it is our hope that policymakers,
 775advocates, scholars, and others working to maximize the benefits
 776and minimize the harms of AI will be better positioned to build
 777on existing efforts and to push the fractured, global
 778conversation on the future of AI toward consensus.},
 779  author = {Fjeld, Jessica and Achten, Nele and Hilligoss, Hannah and Nagy,
 780Adam and Srikumar, Madhulika},
 781  title = {Principled Artificial Intelligence: Mapping Consensus in Ethical
 782and Rights-Based Approaches to Principles for AI},
 786  language = {en},
 787  year = {2019},
 788  month = {September},
 789  pages = {389--399},
 790  number = {9},
 791  volume = {1},
 792  publisher = {Nature Publishing Group},
 793  journal = {Nature Machine Intelligence},
 794  abstract = {In the past five years, private companies, research institutions
 795and public sector organizations have issued principles and
 796guidelines for ethical artificial intelligence (AI). However,
 797despite an apparent agreement that AI should be `ethical', there
 798is debate about both what constitutes `ethical AI' and which
 799ethical requirements, technical standards and best practices are
 800needed for its realization. To investigate whether a global
 801agreement on these questions is emerging, we mapped and analysed
 802the current corpus of principles and guidelines on ethical AI.
 803Our results reveal a global convergence emerging around five
 804ethical principles (transparency, justice and fairness,
 805non-maleficence, responsibility and privacy), with substantive
 806divergence in relation to how these principles are interpreted,
 807why they are deemed important, what issue, domain or actors they
 808pertain to, and how they should be implemented. Our findings
 809highlight the importance of integrating guideline-development
 810efforts with substantive ethical analysis and adequate
 811implementation strategies. As AI technology develops rapidly, it
 812is widely recognized that ethical guidelines are required for
 813safe and fair implementation in society. But is it possible to
 814agree on what is `ethical AI'? A detailed analysis of 84 AI
 815ethics reports around the world, from national and international
 816organizations, companies and institutes, explores this question,
 817finding a convergence around core principles but substantial
 818divergence on practical implementation.},
 819  author = {Jobin, Anna and Ienca, Marcello and Vayena, Effy},
 820  title = {The global landscape of AI ethics guidelines},
 824  location = { Hamburg, Germany},
 825  year = {2023},
 826  month = {April},
 827  volume = {1},
 828  publisher = {Association for Computing Machinery},
 829  author = {Rismani, Shalaleh and Shelby, Renee and Smart, Andrew and
 830Jatho, Edgar and Kroll, Josh A and Moon, Ajung and
 831Rostamzadeh, Negar},
 832  booktitle = {Proceedings of the 2023 {CHI} Conference on Human Factors
 833in Computing Systems ({CHI} '23)},
 834  title = {From Plane Crashes to Algorithmic Harm: Applicability of
 835Safety Engineering Frameworks for Responsible {ML}},
 839  eprint = {2210.05791},
 840  primaryclass = {cs.HC},
 841  archiveprefix = {arXiv},
 842  year = {2022},
 843  month = {October},
 844  abstract = {Understanding the landscape of potential harms from
 845algorithmic systems enables practitioners to better
 846anticipate consequences of the systems they build. It also
 847supports the prospect of incorporating controls to help
 848minimize harms that emerge from the interplay of
 849technologies and social and cultural dynamics. A growing
 850body of scholarship has identified a wide range of harms
 851across different algorithmic technologies. However,
 852computing research and practitioners lack a high level and
 853synthesized overview of harms from algorithmic systems
 854arising at the micro, meso-, and macro-levels of society. We
 855present an applied taxonomy of sociotechnical harms to
 856support more systematic surfacing of potential harms in
 857algorithmic systems. Based on a scoping review of computing
 858research $(n=172)$, we identified five major themes related
 859to sociotechnical harms - representational, allocative,
 860quality-of-service, interpersonal harms, and social
 861system/societal harms - and sub-themes. We describe these
 862categories and conclude with a discussion of challenges and
 863opportunities for future research.},
 864  author = {Shelby, Renee and Rismani, Shalaleh and Henne, Kathryn and
 865Moon, Ajung and Rostamzadeh, Negar and Nicholas, Paul and
 866Yilla, N'mah and Gallegos, Jess and Smart, Andrew and
 867Garcia, Emilio and Virk, Gurleen},
 868  title = {Identifying Sociotechnical Harms of Algorithmic Systems:
 869Scoping a Taxonomy for Harm Reduction},
 873  location = {Seoul, Republic of Korea},
 874  keywords = {responsible AI, language models, risk assessment, responsible
 875innovation, technology risks},
 876  address = {New York, NY, USA},
 877  year = {2022},
 878  month = {June},
 879  series = {FAccT '22},
 880  pages = {214--229},
 881  publisher = {Association for Computing Machinery},
 882  abstract = {Responsible innovation on large-scale Language Models (LMs)
 883requires foresight into and in-depth understanding of the risks
 884these models may pose. This paper develops a comprehensive
 885taxonomy of ethical and social risks associated with LMs. We
 886identify twenty-one risks, drawing on expertise and literature
 887from computer science, linguistics, and the social sciences. We
 888situate these risks in our taxonomy of six risk areas: I.
 889Discrimination, Hate speech and Exclusion, II. Information
 890Hazards, III. Misinformation Harms, IV. Malicious Uses, V.
 891Human-Computer Interaction Harms, and VI. Environmental and
 892Socioeconomic harms. For risks that have already been observed
 893in LMs, the causal mechanism leading to harm, evidence of the
 894risk, and approaches to risk mitigation are discussed. We
 895further describe and analyse risks that have not yet been
 896observed but are anticipated based on assessments of other
 897language technologies, and situate these in the same taxonomy.
 898We underscore that it is the responsibility of organizations to
 899engage with the mitigations we discuss throughout the paper. We
 900close by highlighting challenges and directions for further
 901research on risk evaluation and mitigation with the goal of
 902ensuring that language models are developed responsibly.},
 903  author = {Weidinger, Laura and Uesato, Jonathan and Rauh, Maribeth and
 904Griffin, Conor and Huang, Po-Sen and Mellor, John and Glaese,
 905Amelia and Cheng, Myra and Balle, Borja and Kasirzadeh, Atoosa
 906and Biles, Courtney and Brown, Sasha and Kenton, Zac and
 907Hawkins, Will and Stepleton, Tom and Birhane, Abeba and
 908Hendricks, Lisa Anne and Rimell, Laura and Isaac, William and
 909Haas, Julia and Legassick, Sean and Irving, Geoffrey and
 910Gabriel, Iason},
 911  booktitle = {2022 {ACM} Conference on Fairness, Accountability, and
 913  title = {Taxonomy of Risks posed by Language Models},
 917  keywords = {AI regulation; Labels; Certification; Self-regulation; Soft law},
 918  year = {2022},
 919  month = {April},
 920  pages = {105657},
 921  volume = {44},
 922  journal = {Computer Law \& Security Review},
 923  abstract = {AI regulation is emerging in the EU. The European authorities,
 924NGOs and academics have already issued a series of proposals to
 925accommodate the `development and uptake of AI' with an
 926`appropriate ethical and legal framework' and promote what the
 927European Commission has called an `ecosystem of trust'. In the
 928spring of 2020, the European Commission submitted a legislative
 929proposal for public consultation including four options ranging
 930from ``soft law only'' to a broad scope of mandatory requirements
 931and combinations thereof, for addressing the risks linked to the
 932development and use of certain AI applications. One year later,
 933the Commission unveiled on 21 April 2021 the EU Act on Artificial
 934Intelligence.11Regulation of the European Parliament and of the
 935Council laying down harmonised rules on artificial intelligence
 936(artificial intelligence act) and amending certain union
 937legislative acts, Brussels, 21.4.2021 COM (2021) 206 final. The
 938proposal primarily focusses on regulating 'high-risk' systems
 939through mandatory requirements and prohibition measures. This
 940approach leaves a wide range of AI-systems, with potentially
 941serious impact on fundamental rights, merely unregulated as
 942regards specifically AI related risks. This paper explores the
 943boundaries of the impact of the Act for primarily non-high-risk
 944systems and discuss the options for introducing a voluntary
 945labelling scheme for enhancing protection against the risks of
 946medium and low risk AI systems.},
 947  author = {Stuurman, Kees and Lachaud, Eric},
 948  title = {Regulating {AI}. A label to complete the proposed Act on
 949Artificial Intelligence},
 953  language = {en},
 954  note = {Accessed: 2023-2-15},
 955  howpublished = {\url{}},
 956  year = {2022},
 957  month = {October},
 958  publisher = {Fasken},
 959  title = {The Regulation of Artificial Intelligence in Canada and
 960Abroad: Comparing the Proposed {AIDA} and {EU} {AI} Act},
 964  language = {en},
 965  year = {2022},
 966  month = {October},
 967  journal = {MIT Technology Review},
 968  abstract = {Companies say they want ethical AI. But those working in the
 969field say that ambition comes at their expense.},
 970  author = {Heikkil{\"a}, Melissa},
 971  title = {Responsible {AI} has a burnout problem},
 975  year = {2009},
 976  month = {March},
 977  pages = {3--8},
 978  number = {1},
 979  volume = {21},
 980  publisher = {Taylor \& Francis},
 981  journal = {Engineering Management Journal},
 982  abstract = {Abstract:Leadership must be a key element advancing for the
 983engineering profession to remain relevant and connected in an
 984era of heightened outsourcing and global competition. Companies
 985intent on maintaining a competitive edge are calling upon
 986educators to produce engineers capable of leading
 987multidisciplinary teams, combine technical ingenuity with
 988business acumen, and produce graduates who have a passion for
 989lifelong learning. Industry is also challenging universities to
 990broaden curricula beyond the intellectual endeavors of design
 991and scientific inquiry to the greater domain of professional
 992leadership and entrepreneurship. Managers in industry are
 993similarly challenged to cultivate key leadership attributes in
 994junior engineers. This article explores the changing nature of
 995engineering in a globally competitive environment and addresses
 996why leadership must become a key issue in the career progression
 997of engineers. We will present a literature review of leadership
 998models along with some proposed solutions for cultivating
 999leadership skills as part of the career development process.
1000Lastly, we will present specific recommendations on how to
1001cultivate leadership attributes throughout an engineering
1003  author = {Farr, John V and Brazil, Donna M},
1004  title = {Leadership Skills Development for Engineers},
1008  year = {2009},
1009  month = {March},
1010  pages = {3--8},
1011  number = {1},
1012  volume = {21},
1013  publisher = {Taylor \& Francis},
1014  journal = {Engineering Management Journal},
1015  abstract = {Abstract:Leadership must be a key element advancing for the
1016engineering profession to remain relevant and connected in an
1017era of heightened outsourcing and global competition. Companies
1018intent on maintaining a competitive edge are calling upon
1019educators to produce engineers capable of leading
1020multidisciplinary teams, combine technical ingenuity with
1021business acumen, and produce graduates who have a passion for
1022lifelong learning. Industry is also challenging universities to
1023broaden curricula beyond the intellectual endeavors of design
1024and scientific inquiry to the greater domain of professional
1025leadership and entrepreneurship. Managers in industry are
1026similarly challenged to cultivate key leadership attributes in
1027junior engineers. This article explores the changing nature of
1028engineering in a globally competitive environment and addresses
1029why leadership must become a key issue in the career progression
1030of engineers. We will present a literature review of leadership
1031models along with some proposed solutions for cultivating
1032leadership skills as part of the career development process.
1033Lastly, we will present specific recommendations on how to
1034cultivate leadership attributes throughout an engineering
1036  author = {Farr, John V and Brazil, Donna M},
1037  title = {Leadership Skills Development for Engineers},
1041  language = {en},
1042  year = {2022},
1043  month = {November},
1044  journal = {Wired},
1045  abstract = {As part of a wave of layoffs, the new CEO disbanded a group
1046working to make Twitter's algorithms more transparent and fair.},
1047  author = {Knight, Will},
1048  title = {Elon Musk Has Fired Twitter's `Ethical {AI'} Team},
1052  location = {Oxford, United Kingdom},
1053  keywords = {scale barriers, ethics, startups, data, AI},
1054  address = {New York, NY, USA},
1055  year = {2022},
1056  month = {July},
1057  series = {AIES '22},
1058  pages = {92--106},
1059  publisher = {Association for Computing Machinery},
1060  abstract = {Artificial Intelligence startups use training data as direct
1061inputs in product development. These firms must balance numerous
1062tradeoffs between ethical issues and data access without
1063substantive guidance from regulators or existing judicial
1064precedence. We survey these startups to determine what actions
1065they have taken to address these ethical issues and the
1066consequences of those actions. We find that 58\% of these
1067startups have established a set of AI principles. Startups with
1068data-sharing relationships with high-technology firms or that
1069have prior experience with privacy regulations are more likely
1070to establish ethical AI principles and are more likely to take
1071costly steps, like dropping training data or turning down
1072business, to adhere to their ethical AI policies. Moreover,
1073startups with ethical AI policies are more likely to invest in
1074unconscious bias training, hire ethnic minorities and female
1075programmers, seek expert advice, and search for more diverse
1076training data. Potential costs associated with data-sharing
1077relationships and the adherence to ethical policies may create
1078tradeoffs between increased AI product competition and more
1079ethical AI production.},
1080  author = {Bessen, James and Impink, Stephen Michael and Seamans, Robert},
1081  booktitle = {Proceedings of the 2022 {AAAI/ACM} Conference on {AI}, Ethics,
1082and Society},
1083  title = {The Cost of Ethical {AI} Development for {AI} Startups},
1087  language = {en},
1088  year = {2021},
1089  month = {May},
1090  journal = {MIT Technology Review},
1091  abstract = {Artificial intelligence is changing every industry---from
1092manufacturing to retail. It's also changing the culture at
1093companies as they strive to keep up with accelerating digital
1095  title = {Embracing the rapid pace of {AI}},
1099  keywords = {Artificial intelligence;Ethics;Technology
1100management;Technological innovation;History;Computers;Process
1102  year = {2016},
1103  month = {September},
1104  pages = {682--693},
1105  publisher = {},
1106  abstract = {The development of emergent technologies carries with it ethical
1107issues and risks. We review ways to better manage the ethical
1108issues and risks of one emerging technology: Artificial
1109Intelligence (AI). Depending on how AI's development is managed,
1110it may have beneficial and/or deleterious effects. The
1111processing capacity of Tianhe-2, the world's fastest
1112supercomputer, by some measures, exceeds the processing capacity
1113of a single human brain. but at a prohibitive processing/power
1114consumption ratio and physical size. Given the current pace of
1115AI R\&D activities, some estimates in the literature suggest
1116that the technology could become capable of self-determination
1117and super intelligence in only a few decades. This demands a
1118serious analysis of the ethical implications of AI's development
1119and the risks it might pose, in addition to technology
1120management recommendations. We review the state of AI
1121development, the timeline and scope of its possible future
1122development, and potential ethical risks in its implementation.
1123Further, we briefly review ethics and risk management practices
1124as they relate to technology. Finally, we make technology
1125management recommendations, which may help to address the
1126ethical implications and to mitigate existential risks to
1127humanity-with the development and dissemination of AI-by guiding
1128its proper management.},
1129  author = {Meek, Taylor and Barham, Husam and Beltaif, Nader and Kaadoor,
1130Amani and Akhter, Tanzila},
1131  booktitle = {2016 Portland International Conference on Management of
1132Engineering and Technology ({PICMET})},
1133  title = {Managing the ethical and risk implications of rapid advances in
1134artificial intelligence: A literature review},
1138  note = {Accessed: 2023-3-12},
1139  howpublished = {\url{}},
1140  year = {2022},
1141  month = {November},
1142  author = {Goldman, Sharon},
1143  title = {Why Meta and Twitter's {AI} and {ML} layoffs matter},
1147  location = {Dallas, TX, USA},
1148  keywords = {soft skills, service learning in computer science, communication},
1149  address = {New York, NY, USA},
1150  year = {2011},
1151  month = {March},
1152  series = {SIGCSE '11},
1153  pages = {517--522},
1154  publisher = {Association for Computing Machinery},
1155  abstract = {Soft skills such as communication, teamwork, and organization
1156are important to students' future success in the working world.
1157Faculty members know it, students know it, and employers are
1158explicitly asking for these skills. Are computer science
1159departments responsible to teach these skills? If so, where in
1160the curriculum should they be covered? This paper explores the
1161soft skills that employers want, and possible places to include
1162the teaching of those skills in the curriculum. It then shows
1163how an extensive set of soft skills were incorporated into a
1164service learning course for the students in the Mathematical,
1165Information and Computer Sciences department at Point Loma
1166Nazarene University. Finally, it makes suggestions as to how
1167other service learning or capstone courses could be altered to
1168afford more opportunity for soft skill education.},
1169  author = {Carter, Lori},
1170  booktitle = {Proceedings of the 42nd {ACM} technical symposium on Computer
1171science education},
1172  title = {Ideas for adding soft skills education to service learning and
1173capstone courses for computer science students},
1177  language = {en},
1178  year = {2001},
1179  month = {September},
1180  pages = {867--875},
1181  number = {9},
1182  volume = {35},
1183  publisher = {Wiley Online Library},
1184  journal = {Med. Educ.},
1185  abstract = {PURPOSE: This article examines literature on interdisciplinary
1186education and teamwork in health care, to discover the major
1187issues and best practices. METHODS: A literature review of
1188mainly North American articles using search terms such as
1189interdisciplinary, interprofessional, multidisciplinary with
1190medical education. MAIN FINDINGS: Two issues are emerging in
1191health care as clinicians face the complexities of current
1192patient care: the need for specialized health professionals, and
1193the need for these professionals to collaborate.
1194Interdisciplinary health care teams with members from many
1195professions answer the call by working together, collaborating
1196and communicating closely to optimize patient care. Education on
1197how to function within a team is essential if the endeavour is
1198to succeed. Two main categories of issues emerged: those related
1199to the medical education system and those related to the content
1200of the education. CONCLUSIONS: Much of the literature pertained
1201to programme evaluations of academic activities, and did not
1202compare interdisciplinary education with traditional methods.
1203Many questions about when to educate, who to educate and how to
1204educate remain unanswered and open to future research.},
1205  author = {Hall, P and Weaver, L},
1206  title = {Interdisciplinary education and teamwork: a long and winding
1211  year = {2003},
1212  pages = {186},
1213  number = {4},
1214  volume = {24},
1215  publisher = {},
1216  journal = {Nurs. Educ. Perspect.},
1217  abstract = {making an informed decision about integrated curriculum
1218development and course implementation, multidisciplinary,
1219interdisciplinary, and transdisciplinary educational teams are
1220defined. Examples are offered that reflect these three
1221integrated educational team models. Finally, the benefits and
1222potential problem areas that result from team initiatives are
1223briefly reviewed....},
1224  author = {Dyer, Jean A},
1225  title = {Multidisciplinary, Interdisciplinary, and
1226{TransdisciplinaryEducational} Models and Nursing Education},
1230  year = {2018},
1231  month = {November},
1232  pages = {842--859},
1233  number = {6},
1234  volume = {43},
1235  publisher = {Taylor \& Francis},
1236  journal = {Eur. J. Eng. Educ.},
1237  abstract = {ABSTRACTToday, interdisciplinary education is a hot topic.
1238Gaining an insight into the nature of interdisciplinary
1239education may help when making design decisions for
1240interdisciplinary education. In this study, we argue that,
1241derived from interdisciplinary research, the choice of problem,
1242the level of interaction between different disciplines and
1243constructive alignment are variables to consider when designing
1244interdisciplinary education. Several models of analysis have
1245been used in two descriptive case studies to gain insight into
1246the design parameters for interdisciplinary education. In this
1247study, we AIM to describe (a) the level and nature of
1248integration, (b) the problem definitions as a guiding principle
1249for constructive alignment for (c) the design and execution of
1250interdisciplinary/transdisciplinary education.},
1251  author = {Klaassen, Renate G},
1252  title = {Interdisciplinary education: a case study},
1256  year = {2006},
1257  month = {January},
1258  pages = {77--101},
1259  number = {2},
1260  volume = {3},
1261  publisher = {Routledge},
1262  journal = {Qual. Res. Psychol.},
1263  abstract = {Thematic analysis is a poorly demarcated, rarely acknowledged,
1264yet widely used qualitative analytic method within psychology.
1265In this paper, we argue that it offers an accessible and
1266theoretically flexible approach to analysing qualitative data.
1267We outline what thematic analysis is, locating it in relation to
1268other qualitative analytic methods that search for themes or
1269patterns, and in relation to different epistemological and
1270ontological positions. We then provide clear guidelines to those
1271wanting to start thematic analysis, or conduct it in a more
1272deliberate and rigorous way, and consider potential pitfalls in
1273conducting thematic analysis. Finally, we outline the
1274disadvantages and advantages of thematic analysis. We conclude
1275by advocating thematic analysis as a useful and flexible method
1276for qualitative research in and beyond psychology.},
1277  author = {Braun, Virginia and Clarke, Victoria},
1278  title = {Using thematic analysis in psychology},
1282  language = {en},
1283  keywords = {Artificial Intelligence; Auditing; Conformity assessment;
1284European Union; Governance; Regulation; Technology},
1285  year = {2022},
1286  pages = {241--268},
1287  number = {2},
1288  volume = {32},
1289  journal = {Minds Mach.},
1290  abstract = {The proposed European Artificial Intelligence Act (AIA) is the
1291first attempt to elaborate a general legal framework for AI
1292carried out by any major global economy. As such, the AIA is
1293likely to become a point of reference in the larger discourse on
1294how AI systems can (and should) be regulated. In this article, we
1295describe and discuss the two primary enforcement mechanisms
1296proposed in the AIA: the conformity assessments that providers of
1297high-risk AI systems are expected to conduct, and the post-market
1298monitoring plans that providers must establish to document the
1299performance of high-risk AI systems throughout their lifetimes.
1300We argue that the AIA can be interpreted as a proposal to
1301establish a Europe-wide ecosystem for conducting AI auditing,
1302albeit in other words. Our analysis offers two main
1303contributions. First, by describing the enforcement mechanisms
1304included in the AIA in terminology borrowed from existing
1305literature on AI auditing, we help providers of AI systems
1306understand how they can prove adherence to the requirements set
1307out in the AIA in practice. Second, by examining the AIA from an
1308auditing perspective, we seek to provide transferable lessons
1309from previous research about how to refine further the regulatory
1310approach outlined in the AIA. We conclude by highlighting seven
1311aspects of the AIA where amendments (or simply clarifications)
1312would be helpful. These include, above all, the need to translate
1313vague concepts into verifiable criteria and to strengthen the
1314institutional safeguards concerning conformity assessments based
1315on internal checks.},
1316  author = {M{\"o}kander, Jakob and Axente, Maria and Casolari, Federico and
1317Floridi, Luciano},
1318  title = {Conformity Assessments and Post-market Monitoring: A Guide to the
1319Role of Auditing in the Proposed European {AI} Regulation},
1323  year = {2020},
1324  title = {{Ethics Owners: A new model of organizational responsibility in data-driven technology companies}},
1325  pages = {1--71},
1326  mendeley-groups = {PhD/AI competency project},
1327  file = {:G$\backslash$:/My Drive/PhD/AI Competency Project/Literature Review/Ethics-Owners{\_}20200923-DataSociety.pdf:pdf},
1328  booktitle = {Data Soc.},
1329  author = {Moss, Emanuel and Metcalf, Jacob},
1333  year = {2020},
1334  url = {},
1335  title = {{What Do We Teach When We Teach Tech Ethics? A Syllabi Analysis}},
1336  mendeley-groups = {PhD,PhD/AI competency project},
1337  keywords = {curriculum,ethics,professional responsibility,syllabi},
1338  isbn = {9781450367936},
1339  file = {:C$\backslash$:/Users/shala/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Fiesler, Garrett, Beard - 2020 - What Do We Teach When We Teach Tech Ethics A Syllabi Analysis.pdf:pdf},
1340  doi = {10.1145/3328778.3366825},
1341  booktitle = {SIGCSE '20},
1342  author = {Fiesler, Casey and Garrett, Natalie and Beard, Nathan},
1343  abstract = {As issues of technology ethics become more pervasive in the media and public discussions, there is increasing interest in what role ethics should play in computing education. Not only are there more standalone ethics classes being offered at universities, but calls for greater integration of ethics across computer science curriculum mean that a growing number of CS instructors may be including ethics as part of their courses. To both describe current trends in computing ethics coursework and to provide guidance for further ethics inclusion in computing, we present an in-depth qualitative analysis of 115 syllabi from university technology ethics courses. Our analysis contributes a snapshot of the content and goals of tech ethics classes, and recommendations for how these might be integrated across a computing curriculum. CCS CONCEPTS • Social and professional topics → Computing education.},
1347  year = {2020},
1348  volume = {1},
1349  url = {},
1350  title = {{Emerging challenges in AI and the need for AI ethics education}},
1351  publisher = {Springer International Publishing},
1352  pages = {61--65},
1353  number = {1},
1354  mendeley-groups = {PhD/AI competency project},
1355  keywords = {AI ethics,Artificial intelligence,Design ethics,Ethics education,Professional responsibility,ai ethics,artificial intelligence,design ethics,ethics education,professional responsibility},
1356  journal = {AI Ethics},
1357  issn = {2730-5953},
1358  isbn = {0123456789},
1359  file = {:G$\backslash$:/My Drive/PhD/AI Competency Project/Literature Review/Borenstein-Howard2021{\_}Article{\_}EmergingChallengesInAIAndTheNe.pdf:pdf},
1360  doi = {10.1007/s43681-020-00002-7},
1361  author = {Borenstein, Jason and Howard, Ayanna},
1362  abstract = {Artificial Intelligence (AI) is reshaping the world in profound ways; some of its impacts are certainly beneficial but widespread and lasting harms can result from the technology as well. The integration of AI into various aspects of human life is underway, and the complex ethical concerns emerging from the design, deployment, and use of the technology serves as a reminder that it is time to revisit what future developers and designers, along with professionals, are learning when it comes to AI. It is of paramount importance to train future members of the AI community, and other stakeholders as well, to reflect on the ways in which AI might impact people's lives and to embrace their responsibilities to enhance its benefits while mitigating its potential harms. This could occur in part through the fuller and more systematic inclusion of AI ethics into the curriculum. In this paper, we briefly describe different approaches to AI ethics and offer a set of recommendations related to AI ethics pedagogy.},
1366  year = {2020},
1367  title = {{Co-Designing Checklists to Understand Organizational Challenges and Opportunities around Fairness in AI}},
1368  pages = {1--14},
1369  mendeley-groups = {PhD/AI competency project},
1370  keywords = {"AI,ML,checklists",co-design,ethics,fairness},
1371  isbn = {9781450367080},
1372  file = {:C$\backslash$:/Users/shala/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Madaio et al. - 2020 - Co-Designing Checklists to Understand Organizational Challenges and Opportunities around Fairness in AI.pdf:pdf},
1373  doi = {10.1145/3313831.3376445},
1374  booktitle = {Comput. Hum. Interact.},
1375  author = {Madaio, Michael A. and Stark, Luke and {Wortman Vaughan}, Jennifer and Wallach, Hanna},
1376  abstract = {Many organizations have published principles intended to guide the ethical development and deployment of AI systems; however, their abstract nature makes them difficult to oper-ationalize. Some organizations have therefore produced AI ethics checklists, as well as checklists for more specific concepts , such as fairness, as applied to AI systems. But unless checklists are grounded in practitioners' needs, they may be misused. To understand the role of checklists in AI ethics, we conducted an iterative co-design process with 48 practitioners, focusing on fairness. We co-designed an AI fairness checklist and identified desiderata and concerns for AI fairness checklists in general. We found that AI fairness checklists could provide organizational infrastructure for formalizing ad-hoc processes and empowering individual advocates. We discuss aspects of organizational culture that may impact the efficacy of such checklists, and highlight future research directions.},
1380  year = {2020},
1381  title = {{Closing the AI accountability gap: Defining an end-to-end framework for internal algorithmic auditing}},
1382  pages = {33--44},
1383  mendeley-groups = {PhD/AI Ethics Metrics Paper/Acct/Resp,PhD/AI competency project},
1384  keywords = {Accountability,Algorithmic audits,Machine learning,Responsible innovation},
1385  journal = {FAT* 2020 - Proc. 2020 Conf. Fairness, Accountability, Transpar.},
1386  isbn = {9781450369367},
1387  file = {:C$\backslash$:/Users/shala/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Raji et al. - 2020 - Closing the AI accountability gap Defining an end-to-end framework for internal algorithmic auditing.pdf:pdf},
1388  eprint = {2001.00973},
1389  doi = {10.1145/3351095.3372873},
1390  author = {Raji, Inioluwa Deborah and Smart, Andrew and White, Rebecca N. and Mitchell, Margaret and Gebru, Timnit and Hutchinson, Ben and Smith-Loud, Jamila and Theron, Daniel and Barnes, Parker},
1391  arxivid = {2001.00973},
1392  archiveprefix = {arXiv},
1393  abstract = {Rising concern for the societal implications of artificial intelligence systems has inspired a wave of academic and journalistic literature in which deployed systems are audited for harm by investigators from outside the organizations deploying the algorithms. However, it remains challenging for practitioners to identify the harmful repercussions of their own systems prior to deployment, and, once deployed, emergent issues can become difficult or impossible to trace back to their source. In this paper, we introduce a framework for algorithmic auditing that supports artificial intelligence system development end-to-end, to be applied throughout the internal organization development lifecycle. Each stage of the audit yields a set of documents that together form an overall audit report, drawing on an organization's values or principles to assess the fit of decisions made throughout the process. The proposed auditing framework is intended to contribute to closing the accountability gap in the development and deployment of large-scale artificial intelligence systems by embedding a robust process to ensure audit integrity.},
1397  year = {2019},
1398  volume = {1},
1399  url = {},
1400  title = {{The global landscape of AI ethics guidelines}},
1401  publisher = {Springer US},
1402  pages = {389--399},
1403  number = {9},
1404  mendeley-groups = {PhD,ECSE 681,PhD/AI competency project},
1405  journal = {Nat. Mach. Intell.},
1406  issn = {2522-5839},
1407  isbn = {4225601900},
1408  file = {:C$\backslash$:/Users/shala/AppData/Local/Mendeley Ltd./Mendeley Desktop/Downloaded/Jobin, Ienca, Vayena - 2019 - The global landscape of AI ethics guidelines.pdf:pdf},
1409  doi = {10.1038/s42256-019-0088-2},
1410  author = {Jobin, Anna and Ienca, Marcello and Vayena, Effy},
1411  abstract = {In the past five years, private companies, research institutions and public sector organizations have issued principles and guidelines for ethical artificial intelligence (AI). However, despite an apparent agreement that AI should be ‘ethical', there is debate about both what constitutes ‘ethical AI' and which ethical requirements, technical standards and best practices are needed for its realization. To investigate whether a global agreement on these questions is emerging, we mapped and analysed the current corpus of principles and guidelines on ethical AI. Our results reveal a global convergence emerging around five ethical principles (transparency, justice and fairness, non-maleficence, responsibility and privacy), with substantive divergence in relation to how these principles are interpreted, why they are deemed important, what issue, domain or actors they pertain to, and how they should be implemented. Our findings highlight the importance of integrating guideline-development efforts with substantive ethical analysis and adequate implementation strategies. As AI technology develops rapidly, it is widely recognized that ethical guidelines are required for safe and fair implementation in society. But is it possible to agree on what is ‘ethical AI'? A detailed analysis of 84 AI ethics reports around the world, from national and international organizations, companies and institutes, explores this question, finding a convergence around core principles but substantial divergence on practical implementation.},
1415  year = {2019},
1416  title = {{Foresight into AI Ethics}},
1417  publisher = {Open Roboethics Institute},
1418  number = {October},
1419  mendeley-groups = {PhD/AI competency project},
1420  file = {:G$\backslash$:/My Drive/PhD/AI Competency Project/Literature Review/ORI-Foresight-into-Artificial-Intelligence-Ethics-Launch-V1.pdf:pdf},
1421  author = {Moon, AJung and Rismani, Shalaleh and Millar, Jason and Forsyth, Terralynn and Eshpeter, Jordan and Jaffar, Muhammad and Phan, Anh},
1425  year = {2020},
1426  url = {},
1427  title = {{Ethics and algorithms toolkit}},
1428  mendeley-groups = {PhD/AI competency project},
1429  author = {Anderson, David and Bonaguro, Joy and McKinney, Miriam and Nicklin, Andrew and Wiseman, Jane},
1433  year = {2021},
1434  url = {},
1435  title = {{Competence and competency frameworks - factsheets - CIPD}},
1436  mendeley-groups = {PhD/AI competency project},
1437  booktitle = {CIPD},
1441  year = {2022},
1442  title = {{IEEE Ethics in Action in Autonomous and Intelligent Systems}},
1443  mendeley-groups = {PhD/AI competency project},
1444  booktitle = {IEEE},
1448  year = {2021},
1449  url = {},
1450  title = {{A European approach to artificial intelligence - shaping Europe's digital future}},
1451  mendeley-groups = {PhD/AI competency project},
1452  booktitle = {Eur. Comm.},
1456  year = {2022},
1457  url = {},
1458  title = {{Responsible AI resources}},
1459  mendeley-groups = {PhD/AI competency project},
1460  booktitle = {Microsoft AI},
1464  year = {2022},
1465  url = {},
1466  title = {{AI Ethics Toolkit}},
1467  mendeley-groups = {PhD/AI competency project},
1468  booktitle = {IBM},
1472  year = {2022},
1473  url = {},
1474  title = {{Algorithmic Impact Assessment}},
1475  mendeley-groups = {PhD/AI competency project},
1476  booktitle = {Gov. Canada},
1480  year = {2016},
1481  title = {{The handbook of competency mapping}},
1482  publisher = {SAGE},
1483  mendeley-groups = {PhD/AI competency project},
1484  edition = {3e},
1485  author = {Sanghi, Seema},
1489  year = {1993},
1490  title = {{Competence at Work - Models for Superior Performance}},
1491  publisher = {John Wiley {\&} Sons, Inc.},
1492  mendeley-groups = {PhD/AI competency project},
1493  author = {Spencer, L.M. and Spencer, S.M.},
1494  address = {New York},
1498  year = {2022},
1499  url = {},
1500  title = {{The ESCO Classification}},
1501  mendeley-groups = {PhD/AI competency project},
1502  author = {{European Commission}},
1506  year = {2022},
1507  url = {},
1508  title = {{O*NET database Content Model}},
1509  mendeley-groups = {PhD/AI competency project},
1510  author = {Administration, Employment {\&} Training},
1514  keywords = {responsible AI, documentation, machine learning, datasets},
1515  address = {New York, NY, USA},
1516  year = {2022},
1517  month = {November},
1518  pages = {1--29},
1519  number = {CSCW2},
1520  volume = {6},
1521  publisher = {Association for Computing Machinery},
1522  journal = {Proc. ACM Hum.-Comput. Interact.},
1523  abstract = {Data is central to the development and evaluation of machine
1524learning (ML) models. However, the use of problematic or
1525inappropriate datasets can result in harms when the resulting
1526models are deployed. To encourage responsible AI practice
1527through more deliberate reflection on datasets and transparency
1528around the processes by which they are created, researchers and
1529practitioners have begun to advocate for increased data
1530documentation and have proposed several data documentation
1531frameworks. However, there is little research on whether these
1532data documentation frameworks meet the needs of ML
1533practitioners, who both create and consume datasets. To address
1534this gap, we set out to understand ML practitioners' data
1535documentation perceptions, needs, challenges, and desiderata,
1536with the ultimate goal of deriving design requirements that can
1537inform future data documentation frameworks. We conducted a
1538series of semi-structured interviews with 14 ML practitioners at
1539a single large, international technology company. We had them
1540answer a list of questions taken from datasheets for
1541datasets~\textbackslashcitegebru2018datasheets. Our findings
1542show that current approaches to data documentation are largely
1543ad hoc and myopic in nature. Participants expressed needs for
1544data documentation frameworks to be adaptable to their contexts,
1545integrated into their existing tools and workflows, and
1546automated wherever possible. Despite the fact that data
1547documentation frameworks are often motivated from the
1548perspective of responsible AI, participants did not make the
1549connection between the questions that they were asked to answer
1550and their responsible AI implications. In addition, participants
1551often had difficulties prioritizing the needs of dataset
1552consumers and providing information that someone unfamiliar with
1553their datasets might need to know. Based on these findings, we
1554derive seven design requirements for future data documentation
1555frameworks such as more actionable guidance on how the
1556characteristics of datasets might result in harms and how these
1557harms might be mitigated, more explicit prompts for reflection,
1558automated adaptation to different contexts, and integration into
1559ML practitioners' existing tools and workflows.},
1560  author = {Heger, Amy K and Marquis, Liz B and Vorvoreanu, Mihaela and
1561Wallach, Hanna and Wortman Vaughan, Jennifer},
1562  title = {Understanding Machine Learning Practitioners' Data Documentation
1563Perceptions, Needs, Challenges, and Desiderata},
1567  eprint = {2206.00335},
1568  primaryclass = {cs.AI},
1569  archiveprefix = {arXiv},
1570  year = {2022},
1571  month = {June},
1572  abstract = {The organizational use of artificial intelligence (AI) has
1573rapidly spread across various sectors. Alongside the
1574awareness of the benefits brought by AI, there is a growing
1575consensus on the necessity of tackling the risks and
1576potential harms, such as bias and discrimination, brought
1577about by advanced AI technologies. A multitude of AI ethics
1578principles have been proposed to tackle these risks, but the
1579outlines of organizational processes and practices for
1580ensuring socially responsible AI development are in a
1581nascent state. To address the paucity of comprehensive
1582governance models, we present an AI governance framework,
1583the hourglass model of organizational AI governance, which
1584targets organizations that develop and use AI systems. The
1585framework is designed to help organizations deploying AI
1586systems translate ethical AI principles into practice and
1587align their AI systems and processes with the forthcoming
1588European AI Act. The hourglass framework includes governance
1589requirements at the environmental, organizational, and AI
1590system levels. At the AI system level, we connect governance
1591requirements to AI system life cycles to ensure governance
1592throughout the system's life span. The governance model
1593highlights the systemic nature of AI governance and opens
1594new research avenues into its practical implementation, the
1595mechanisms that connect different AI governance layers, and
1596the dynamics between the AI governance actors. The model
1597also offers a starting point for organizational
1598decision-makers to consider the governance components needed
1599to ensure social acceptability, mitigate risks, and realize
1600the potential of AI.},
1601  author = {M{\"a}ntym{\"a}ki, Matti and Minkkinen, Matti and Birkstedt,
1602Teemu and Viljanen, Mika},
1603  title = {Putting {AI} Ethics into Practice: The Hourglass Model of
1604Organizational {AI} Governance},
1608  year = {2021},
1609  publisher = {},
1610  journal = {Behav. Res. Methods Instrum. Comput.},
1611  abstract = {… frameworks, we evaluated four commercial chatbots against four
1612responsible AI frameworks. We found that the ethical frameworks
1613produced quite di erent assessment scores. Many ethi…},
1614  author = {Atkins, Suzanne and Badrie, Ishwarradj and van Otterloo,
1616  title = {Applying Ethical {AI} Frameworks in practice: Evaluating
1617conversational {AI} chatbot solutions},
1621  year = {2023},
1622  publisher = {},
1623  abstract = {… ized in practice across sectors and organizations [61]. We use
1624the term `` responsible AI '' by its … primarily opted to defne
1625responsible AI in terms of the practices they used to enact it,
1626rather …},
1627  author = {Wang, Qiaosi and Madaio, Michael Adam and Kapania, Shivani and
1628Kane, Shaun and Terry, Michael and Wilcox, Lauren},
1629  title = {Designing responsible {AI}: Adaptations of {UX} practice to meet
1630responsible {AI} challenges},
1634  year = {2020},
1635  month = {February},
1636  author = {Pak-Hang Wong, Judith Simon},
1637  title = {Thinking About `Ethics' in the Ethics of {AI}},
1641  year = {2023},
1642  month = {January},
1643  pages = {20539517231177620},
1644  number = {1},
1645  volume = {10},
1646  publisher = {SAGE Publications Ltd},
1647  journal = {Big Data \& Society},
1648  abstract = {Responsible artificial intelligence guidelines ask engineers to
1649consider how their systems might harm. However, contemporary
1650artificial intelligence systems are built by composing many
1651preexisting software modules that pass through many hands before
1652becoming a finished product or service. How does this shape
1653responsible artificial intelligence practice? In interviews with
165427 artificial intelligence engineers across industry, open
1655source, and academia, our participants often did not see the
1656questions posed in responsible artificial intelligence
1657guidelines to be within their agency, capability, or
1658responsibility to address. We use Suchman's ?located
1659accountability? to show how responsible artificial intelligence
1660labor is currently organized and to explore how it could be done
1661differently. We identify cross-cutting social logics, like
1662modularizability, scale, reputation, and customer orientation,
1663that organize which responsible artificial intelligence actions
1664do take place and which are relegated to low status staff or
1665believed to be the work of the next or previous person in the
1666imagined ?supply chain.? We argue that current responsible
1667artificial intelligence interventions, like ethics checklists
1668and guidelines that assume panoptical knowledge and control over
1669systems, could be improved by taking a located accountability
1670approach, recognizing where relations and obligations might
1671intertwine inside and outside of this supply chain.},
1672  author = {Widder, David Gray and Nafus, Dawn},
1673  title = {Dislocated accountabilities in the {``AI} supply chain'':
1674Modularity and developers' notions of responsibility},
1678  year = {2022},
1679  pages = {6},
1680  number = {2},
1681  volume = {34},
1682  publisher = {},
1683  journal = {Scandinavian Journal of Information Systems},
1684  abstract = {The increasing use of Artificial Intelligence (AI) systems has
1685sparked discussions regarding developing ethically responsible
1686technology. Consequently, various organizations have released
1687high-level AI ethics frameworks to assist in AI design. However,
1688we still know too little about how AI ethics principles are
1689perceived and work in practice, especially in public
1690organizations. This study examines how AI practitioners perceive
1691ethical issues in their work concerning AI design and how they
1692interpret and put them into practice. We conducted an empirical
1693study consisting of semi-structured qualitative interviews with
1694AI practitioners working in or for public organizations. Taking
1695the lens provided by the In-Action Ethics framework and previous
1696studies on ethical tensions, we analyzed practitioners'
1697interpretations of AI ethics principles and their application in
1698practice. We found tensions between practitioners'
1699interpretation of ethical principles in their work and ethos
1700tensions. In this vein, we argue that understanding the
1701different tensions that can occur in practice and how they are
1702tackled is key to studying ethics in practice. Understanding how
1703AI practitioners perceive and apply ethical principles is
1704necessary for practical ethics to contribute toward an
1705empirically grounded, Responsible AI.},
1706  author = {Figueras, Cl{\`a}udia and Verhagen, Harko and Pargman, Teresa
1708  title = {Exploring tensions in Responsible {AI} in practice. An interview
1709study on {AI} practices in and for Swedish public organizations},
1713  language = {en},
1714  year = {2023},
1715  month = {March},
1716  pages = {1--9},
1717  number = {1},
1718  volume = {10},
1719  publisher = {Palgrave},
1720  journal = {Humanities and Social Sciences Communications},
1721  abstract = {There is a growing debate amongst academics and practitioners on
1722whether interventions made, thus far, towards Responsible AI
1723have been enough to engage with the root causes of AI problems.
1724Failure to effect meaningful changes in this system could see
1725these initiatives not reach their potential and lead to the
1726concept becoming another buzzword for companies to use in their
1727marketing campaigns. Systems thinking is often touted as a
1728methodology to manage and effect change; however, there is
1729little practical advice available for decision-makers to include
1730systems thinking insights to work towards Responsible AI. Using
1731the notion of `leverage zones' adapted from the systems thinking
1732literature, we suggest a novel approach to plan for and
1733experiment with potential initiatives and interventions. This
1734paper presents a conceptual framework called the Five Ps to help
1735practitioners construct and identify holistic interventions that
1736may work towards Responsible AI, from lower-order interventions
1737such as short-term fixes, tweaking algorithms and updating
1738parameters, through to higher-order interventions such as
1739redefining the system's foundational structures that govern
1740those parameters, or challenging the underlying purpose upon
1741which those structures are built and developed in the first
1742place. Finally, we reflect on the framework as a scaffold for
1743transdisciplinary question-asking to improve outcomes towards
1744Responsible AI.},
1745  author = {Nabavi, Ehsan and Browne, Chris},
1746  title = {Leverage zones in Responsible {AI}: towards a systems thinking
1751  eprint = {2006.04707},
1752  primaryclass = {cs.CY},
1753  archiveprefix = {arXiv},
1754  year = {2020},
1755  month = {June},
1756  abstract = {Companies have considered adoption of various high-level
1757artificial intelligence (AI) principles for responsible AI,
1758but there is less clarity on how to implement these
1759principles as organizational practices. This paper reviews
1760the principles-to-practices gap. We outline five
1761explanations for this gap ranging from a disciplinary divide
1762to an overabundance of tools. In turn, we argue that an
1763impact assessment framework which is broad,
1764operationalizable, flexible, iterative, guided, and
1765participatory is a promising approach to close the
1766principles-to-practices gap. Finally, to help practitioners
1767with applying these recommendations, we review a case study
1768of AI's use in forest ecosystem restoration, demonstrating
1769how an impact assessment framework can translate into
1770effective and responsible AI practices.},
1771  author = {Schiff, Daniel and Rakova, Bogdana and Ayesh, Aladdin and
1772Fanti, Anat and Lennon, Michael},
1773  title = {Principles to Practices for Responsible {AI}: Closing the
1778  location = {Oxford, United Kingdom},
1779  keywords = {ISO 26000:2010 guidance on social responsibility, AI system
1780builders, responsible AI systems, AI ethics, corporate social
1781responsibility, stakeholder identification},
1782  address = {New York, NY, USA},
1783  year = {2022},
1784  month = {July},
1785  series = {AIES '22},
1786  pages = {227--236},
1787  publisher = {Association for Computing Machinery},
1788  abstract = {As of 2021, there were more than 170 guidelines on AI ethics and
1789responsible, trustworthy AI in circulation according to the AI
1790Ethics Guidelines Global Inventory maintained by AlgorithmWatch,
1791an organisation which tracks the effects of increased
1792digitalisation on everyday lives. However, from the perspective
1793of day-to-day work, for those engaged in designing, developing,
1794and maintaining AI systems identifying relevant guidelines and
1795translating them into practice presents a challenge.The aim of
1796this paper is to help anyone engaged in building a responsible
1797AI system by identifying an indicative long-list of potential
1798stakeholders. This list of impacted stakeholders is intended to
1799enable such AI system builders to decide which guidelines are
1800most suited to their practice. The paper draws on a literature
1801review of articles short-listed based on searches conducted in
1802the ACM Digital Library and Google Scholar. The findings are
1803based on content analysis of the short-listed literature guided
1804by probes which draw on the ISO 26000:2010 Guidance on social
1805responsibility.The paper identifies three levels of potentially
1806relevant stakeholders when responsible AI systems are
1807considered: individual stakeholders (including users,
1808developers, and researchers), organisational stakeholders, and
1809national / international stakeholders engaged in making laws,
1810rules, and regulations. The main intended audience for this
1811paper is software, requirements, and product engineers engaged
1812in building AI systems. In addition, business executives, policy
1813makers, legal/regulatory experts, AI researchers, public,
1814private, and third sector organisations developing responsible
1815AI guidelines, and anyone interested in seeing functional
1816responsible AI systems are the other intended audience for this
1818  author = {Deshpande, Advait and Sharp, Helen},
1819  booktitle = {Proceedings of the 2022 {AAAI/ACM} Conference on {AI}, Ethics,
1820and Society},
1821  title = {Responsible {AI} Systems: Who are the Stakeholders?},
1825  note = {Accessed: 2023-3-11},
1826  howpublished = {\url{}},
1827  abstract = {As applications of artificial intelligence (AI) have
1828proliferated so too have ethical concerns regarding their
1829potential to cause harm to society. As a result, many
1830organizations that build or use AI systems have developed
1831frameworks or codes of conduct specifying ethical or
1832responsible AI principles they strive to follow (e.g.},
1833  author = {Heger, Amy and Passi, Samir and Vorvoreanu, Mihaela},
1834  title = {All the tools, none of the motivation: Organizational culture
1835and barriers to responsible {AI} work},
1839  keywords = {Economics;Social factors;Education;Medical services;Artificial
1840intelligence;Sustainable development;Autonomous vehicles},
1841  year = {2021},
1842  month = {June},
1843  pages = {81--94},
1844  number = {2},
1845  volume = {40},
1846  publisher = {},
1847  journal = {IEEE Technol. Soc. Mag.},
1848  abstract = {As artificial intelligence (AI) permeates across social and
1849economic life, its ethical and governance implications have come
1850to the forefront. Active debates surround AI's role in labor
1851displacement, autonomous vehicles, military, misinformation,
1852healthcare, education, and more. As societies collectively
1853grapple with these challenges, new opportunities for AI to
1854proactively contribute to social good (AI4SG) and equity (AI4Eq)
1855have also been proposed [1], [2], such as Microsoft's AI for
1856Earth program. These efforts highlight the potential of AI to
1857address global challenges and help achieve targets like the
1858United Nation's sustainable development goals (SDGs) [3]. Yet,
1859whether AI efforts are directed explicitly at social good and
1860equity or not, there are many barriers that stand between
1861aspirations to be responsible and the translation of these
1862aspirations into concrete practicalities.},
1863  author = {Schiff, Daniel and Rakova, Bogdana and Ayesh, Aladdin and Fanti,
1864Anat and Lennon, Michael},
1865  title = {Explaining the Principles to Practices Gap in {AI}},
1869  keywords = {responsible ai, organizational structure, industry practice},
1870  address = {New York, NY, USA},
1871  year = {2021},
1872  month = {April},
1873  pages = {1--23},
1874  number = {CSCW1},
1875  volume = {5},
1876  publisher = {Association for Computing Machinery},
1877  journal = {Proc. ACM Hum.-Comput. Interact.},
1878  abstract = {Large and ever-evolving technology companies continue to invest
1879more time and resources to incorporate responsible Artificial
1880Intelligence (AI) into production-ready systems to increase
1881algorithmic accountability. This paper examines and seeks to
1882offer a framework for analyzing how organizational culture and
1883structure impact the effectiveness of responsible AI initiatives
1884in practice. We present the results of semi-structured
1885qualitative interviews with practitioners working in industry,
1886investigating common challenges, ethical tensions, and effective
1887enablers for responsible AI initiatives. Focusing on major
1888companies developing or utilizing AI, we have mapped what
1889organizational structures currently support or hinder
1890responsible AI initiatives, what aspirational future processes
1891and structures would best enable effective initiatives, and what
1892key elements comprise the transition from current work practices
1893to the aspirational future.},
1894  author = {Rakova, Bogdana and Yang, Jingying and Cramer, Henriette and
1895Chowdhury, Rumman},
1896  title = {Where Responsible {AI} meets Reality: Practitioner Perspectives
1897on Enablers for Shifting Organizational Practices},
1901  language = {en},
1902  year = {2021},
1903  month = {January},
1904  number = {1},
1905  volume = {19},
1906  publisher = {UCL Press},
1907  journal = {Lond. Rev. Educ.},
1908  abstract = {Artificial intelligence (AI) is impacting education in many
1909different ways. From virtual assistants for personalized
1910education, to student or teacher tracking systems, the potential
1911benefits of AI for education often come with a discussion of its
1912impact on privacy and well-being. At the same time, the social
1913transformation brought about by AI requires reform of
1914traditional education systems. This article discusses what a
1915responsible, trustworthy vision for AI is and how this relates
1916to and affects education.},
1917  author = {Dignum, Virginia},
1918  title = {The role and challenges of education for responsible {AI}},
1922  year = {2021},
1923  month = {February},
1924  pages = {87--91},
1925  number = {1},
1926  volume = {1},
1927  journal = {AI and Ethics},
1928  abstract = {Despite there being a strong call for responsible technology, the
1929path towards putting ethics into action is still yet to be fully
1930understood. To help guide the implementation of ethics, we have
1931seen the rise of a new professional title; the AI Ethicist.
1932However, it is still unclear what the role and skill set of this
1933new profession must include. The purpose of this piece is to
1934offer a preliminary definition of what it means to be an AI
1935Ethicist by first examining the concept of an ethicist in the
1936context of artificial intelligence, followed by exploring what
1937responsibilities are added to the role in industry specifically,
1938and ending on the fundamental characteristic that underlies it
1939all: bravery.},
1940  author = {Gambelin, Olivia},
1941  title = {Brave: what it means to be an {AI} Ethicist},
1945  keywords = {Ethics;Computer
1948impact;emotional engagement;ethics},
1949  year = {2023},
1950  pages = {1--1},
1951  publisher = {},
1952  journal = {IEEE Transactions on Technology and Society},
1953  abstract = {As computing becomes more powerful and extends the reach of
1954those who wield it, the imperative grows for computing
1955professionals to make ethical decisions regarding the use of
1956that power. We propose the concept of abstracted power to help
1957computer science students understand how technology may distance
1958them perceptually from consequences of their actions.
1959Specifically, we identify technological intermediation and
1960computational thinking as two factors in computer science that
1961contribute to this distancing. To counter the abstraction of
1962power, we argue for increased emotional engagement in computer
1963science ethics education, to encourage students to feel as well
1964as think regarding the potential impacts of their power on
1965others. We suggest four concrete pedagogical approaches to
1966enable this emotional engagement in computer science ethics
1967curriculum, and we share highlights of student reactions to the
1969  author = {Peterson, Tina L and Ferreira, Rodrigo and Vardi, Moshe Y},
1970  title = {Abstracted Power and Responsibility in Computer Science Ethics
1975  language = {en},
1976  copyright = {},
1977  year = {2022},
1978  month = {March},
1979  publisher = {Springer Science and Business Media LLC},
1980  journal = {AI Soc.},
1981  abstract = {AbstractArtificial intelligence ethics requires a united
1982approach from policymakers, AI companies, and individuals, in
1983the development, deployment, and use of these technologies.
1984However, sometimes discussions can become fragmented because of
1985the different levels of governance (Schmitt in AI Ethics 1--12,
19862021) or because of different values, stakeholders, and actors
1987involved (Ryan and Stahl in J Inf Commun Ethics Soc 19:61--86,
19882021). Recently, these conflicts became very visible, with such
1989examples as the dismissal of AI ethics researcher Dr. Timnit
1990Gebru from Google and the resignation of whistle-blower Frances
1991Haugen from Facebook. Underpinning each debacle was a conflict
1992between the organisation's economic and business interests and
1993the morals of their employees. This paper will examine tensions
1994between the ethics of AI organisations and the values of their
1995employees, by providing an exploration of the AI ethics
1996literature in this area, and a qualitative analysis of three
1997workshops with AI developers and practitioners. Common ethical
1998and social tensions (such as power asymmetries, mistrust,
1999societal risks, harms, and lack of transparency) will be
2000discussed, along with proposals on how to avoid or reduce these
2001conflicts in practice (e.g., building trust, fair allocation of
2002responsibility, protecting employees' autonomy, and encouraging
2003ethical training and practice). Altogether, we suggest the
2004following steps to help reduce ethical issues within AI
2005organisations: improved and diverse ethics education and
2006training within businesses; internal and external ethics
2007auditing; the establishment of AI ethics ombudsmen, AI ethics
2008review committees and an AI ethics watchdog; as well as access
2009to trustworthy AI ethics whistle-blower organisations.},
2010  author = {Ryan, Mark and Christodoulou, Eleni and Antoniou, Josephina and
2011Iordanou, Kalypso},
2012  title = {An {AI} ethics `David and Goliath': value conflicts between
2013large tech companies and their employees},
2017  year = {2020},
2018  institution = {IEEE},
2019  pages = {945--947},
2020  publisher = {},
2021  abstract = {… the key ethical requirements for AI development and … in AI
2022environments are well aware of potential consequences of the
2023technologies they develop and deploy, sensitive to the ethical
2025  author = {Gorur, Radhika and Hoon, Leonard and Kowal, Emma},
2026  booktitle = {2020 {IEEE} International Conference on Teaching, Assessment,
2027and Learning for Engineering ({TALE})},
2028  title = {Computer Science Ethics Education in {Australia--A} Work in
2033  language = {en},
2034  year = {2020},
2035  month = {April},
2036  pages = {13428--13435},
2037  number = {09},
2038  volume = {34},
2039  publisher = {},
2040  journal = {AAAI},
2041  abstract = {We propose an experimental ethics-based curricular module for an
2042undergraduate course on Robot Ethics. The proposed module aims
2043to teach students how human subjects research methods can be
2044used to investigate potential ethical concerns arising in
2045human-robot interaction, by engaging those students in real
2046experimental ethics research. In this paper we describe the
2047proposed curricular module, describe our implementation of that
2048module within a Robot Ethics course offered at a medium-sized
2049engineering university, and statistically evaluate the
2050effectiveness of the proposed curricular module in achieving
2051desired learning objectives. While our results do not provide
2052clear evidence of a quantifiable benefit to undergraduate
2053achievement of the described learning objectives, we note that
2054the module did provide additional learning opportunities for
2055graduate students in the course, as they helped to supervise,
2056analyze, and write up the results of this
2057undergraduate-performed research experiment.},
2058  author = {Williams, Tom and Zhu, Qin and Grollman, Daniel},
2059  title = {An Experimental Ethics Approach to Robot Ethics Education},
2063  location = {Virtual Event, USA},
2064  keywords = {middle school, cs education, machine learning, ai literacy},
2065  address = {New York, NY, USA},
2066  year = {2021},
2067  month = {March},
2068  series = {SIGCSE '21},
2069  pages = {1382},
2070  publisher = {Association for Computing Machinery},
2071  abstract = {We developed the How to Train Your Robot curriculum to empower
2072middle school students to become conscientious users and
2073creators of Artificial Intelligence (AI). As AI becomes more
2074embedded in our daily lives, all members of society should have
2075the opportunity to become AI literate. Today, most deployed work
2076in K-12 AI education takes place at strong STEM schools or
2077during extracurricular clubs. But, to promote equity in the
2078field of AI, we must also design curricula for classroom use at
2079schools with limited resources. How to Train Your Robot
2080leverages a low-cost (\$40) robot, a block-based programming
2081platform, novice-friendly model creation tools, and hands-on
2082activities to introduce students to machine learning. During the
2083summer of 2020, we trained in-service teachers, primarily from
2084Title 1 public schools, to deliver a five-day, online version of
2085the curriculum to their students. In this work, we describe how
2086students' self-directed final projects demonstrate their
2087understanding of technical and ethical AI concepts. Students
2088successfully selected project ideas, taking the strengths and
2089weaknesses of machine learning into account, and implemented an
2090array of projects about everything from entertainment to
2091science. We saw that students had the most difficulty designing
2092mechanisms to respond to user feedback after deployment. We hope
2093this work inspires future AI curricula that can be used in
2094middle school classrooms.},
2095  author = {Williams, Randi},
2096  booktitle = {Proceedings of the 52nd {ACM} Technical Symposium on Computer
2097Science Education},
2098  title = {How to Train Your Robot: {Project-Based} {AI} and Ethics
2099Education for Middle School Classrooms},
2103  eprint = {2109.02866},
2104  primaryclass = {cs.AI},
2105  archiveprefix = {arXiv},
2106  year = {2021},
2107  month = {September},
2108  abstract = {Medical students will almost inevitably encounter powerful
2109medical AI systems early in their careers. Yet, contemporary
2110medical education does not adequately equip students with
2111the basic clinical proficiency in medical AI needed to use
2112these tools safely and effectively. Education reform is
2113urgently needed, but not easily implemented, largely due to
2114an already jam-packed medical curricula. In this article, we
2115propose an education reform framework as an effective and
2116efficient solution, which we call the Embedded AI Ethics
2117Education Framework. Unlike other calls for education reform
2118to accommodate AI teaching that are more radical in scope,
2119our framework is modest and incremental. It leverages
2120existing bioethics or medical ethics curricula to develop
2121and deliver content on the ethical issues associated with
2122medical AI, especially the harms of technology misuse,
2123disuse, and abuse that affect the risk-benefit analyses at
2124the heart of healthcare. In doing so, the framework provides
2125a simple tool for going beyond the ``What?'' and the
2126``Why?'' of medical AI ethics education, to answer the
2127``How?'', giving universities, course directors, and/or
2128professors a broad road-map for equipping their students
2129with the necessary clinical proficiency in medical AI.},
2130  author = {Quinn, Thomas P and Coghlan, Simon},
2131  title = {Readying Medical Students for Medical {AI}: The Need to
2132Embed {AI} Ethics Education},
2136  location = {New York NY USA},
2137  conference = {AIES '20: AAAI/ACM Conference on AI, Ethics, and Society},
2138  address = {New York, NY, USA},
2139  year = {2020},
2140  month = {February},
2141  publisher = {ACM},
2142  abstract = {… Therefore, to explore one component of the current state of
2143AI ethics education , we examine a sample of these classes for
2144patterns in what topics are currently being taught. …},
2145  author = {Garrett, Natalie and Beard, Nathan and Fiesler, Casey},
2146  booktitle = {Proceedings of the {AAAI/ACM} Conference on {AI}, Ethics, and
2148  title = {More than ``if time allows'': The Role of Ethics in {AI}
2153  location = {Virtual Event, Canada},
2154  address = {New York, NY, USA},
2155  year = {2021},
2156  month = {March},
2157  series = {FAccT '21},
2158  pages = {515--525},
2159  publisher = {Association for Computing Machinery},
2160  abstract = {Given a growing concern about the lack of ethical consideration
2161in the Artificial Intelligence (AI) field, many have begun to
2162question how dominant approaches to the disciplinary education
2163of computer science (CS)---and its implications for AI---has led
2164to the current ``ethics crisis''. However, we claim that the
2165current AI ethics education space relies on a form of
2166``exclusionary pedagogy,'' where ethics is distilled for
2167computational approaches, but there is no deeper epistemological
2168engagement with other ways of knowing that would benefit ethical
2169thinking or an acknowledgement of the limitations of uni-vocal
2170computational thinking. This results in indifference,
2171devaluation, and a lack of mutual support between CS and
2172humanistic social science (HSS), elevating the myth of
2173technologists as ``ethical unicorns'' that can do it all, though
2174their disciplinary tools are ultimately limited. Through an
2175analysis of computer science education literature and a review
2176of college-level course syllabi in AI ethics, we discuss the
2177limitations of the epistemological assumptions and hierarchies
2178of knowledge which dictate current attempts at including ethics
2179education in CS training and explore evidence for the practical
2180mechanisms through which this exclusion occurs. We then propose
2181a shift towards a substantively collaborative, holistic, and
2182ethically generative pedagogy in AI education.},
2183  author = {Raji, Inioluwa Deborah and Scheuerman, Morgan Klaus and
2184Amironesei, Razvan},
2185  booktitle = {Proceedings of the 2021 {ACM} Conference on Fairness,
2186Accountability, and Transparency},
2187  title = {You Can't Sit With Us: Exclusionary Pedagogy in {AI} Ethics
2192  address = {New York, NY, USA},
2193  year = {2019},
2194  month = {January},
2195  pages = {13--15},
2196  number = {4},
2197  volume = {4},
2198  publisher = {Association for Computing Machinery},
2199  journal = {AI Matters},
2200  abstract = {In this column, we introduce our Model AI Assignment, A Module
2201on Ethical Thinking about Autonomous Vehicles in an AI Course,
2202and more broadly introduce a conversation on ethics education in
2203AI education.},
2204  author = {Furey, Heidi and Martin, Fred},
2205  title = {{AI} education matters: a modular approach to {AI} ethics
2210  language = {en},
2211  year = {2021},
2212  month = {February},
2213  pages = {61--65},
2214  number = {1},
2215  volume = {1},
2216  publisher = {Springer Science and Business Media LLC},
2217  journal = {AI Ethics},
2218  abstract = {… approaches to AI ethics and offer a set of recommendations
2219related to AI ethics pedagogy. … Within this realm, we highlight
2220recent approaches to AI ethics education , especially as they …},
2221  author = {Borenstein, Jason and Howard, Ayanna},
2222  title = {Emerging challenges in {AI} and the need for {AI} ethics
2227  location = {Seoul, Republic of Korea},
2228  keywords = {algorithm audit, ethical AI, AI bias, AI audit, audit,
2229algorithmic accountability, AI policy, AI harm},
2230  address = {New York, NY, USA},
2231  year = {2022},
2232  month = {June},
2233  series = {FAccT '22},
2234  pages = {1571--1583},
2235  publisher = {Association for Computing Machinery},
2236  abstract = {Algorithmic audits (or `AI audits') are an increasingly popular
2237mechanism for algorithmic accountability; however, they remain
2238poorly defined. Without a clear understanding of audit
2239practices, let alone widely used standards or regulatory
2240guidance, claims that an AI product or system has been audited,
2241whether by first-, second-, or third-party auditors, are
2242difficult to verify and may potentially exacerbate, rather than
2243mitigate, bias and harm. To address this knowledge gap, we
2244provide the first comprehensive field scan of the AI audit
2245ecosystem. We share a catalog of individuals (N=438) and
2246organizations (N=189) who engage in algorithmic audits or whose
2247work is directly relevant to algorithmic audits; conduct an
2248anonymous survey of the group (N=152); and interview industry
2249leaders (N=10). We identify emerging best practices as well as
2250methods and tools that are becoming commonplace, and enumerate
2251common barriers to leveraging algorithmic audits as effective
2252accountability mechanisms. We outline policy recommendations to
2253improve the quality and impact of these audits, and highlight
2254proposals with wide support from algorithmic auditors as well as
2255areas of debate. Our recommendations have implications for
2256lawmakers, regulators, internal company policymakers, and
2257standards-setting bodies, as well as for auditors. They are: 1)
2258require the owners and operators of AI systems to engage in
2259independent algorithmic audits against clearly defined
2260standards; 2) notify individuals when they are subject to
2261algorithmic decision-making systems; 3) mandate disclosure of
2262key components of audit findings for peer review; 4) consider
2263real-world harm in the audit process, including through
2264standardized harm incident reporting and response mechanisms; 5)
2265directly involve the stakeholders most likely to be harmed by AI
2266systems in the algorithmic audit process; and 6) formalize
2267evaluation and, potentially, accreditation of algorithmic
2269  author = {Costanza-Chock, Sasha and Raji, Inioluwa Deborah and Buolamwini,
2271  booktitle = {2022 {ACM} Conference on Fairness, Accountability, and
2273  title = {Who Audits the Auditors? Recommendations from a field scan of
2274the algorithmic auditing ecosystem},
2278  location = {Seoul, Republic of Korea},
2279  keywords = {regulation, start-ups, fairness, social practice, transparency,
2280innovation, AI ethics, organizations, socio-cultural history,
2282  address = {New York, NY, USA},
2283  year = {2022},
2284  month = {June},
2285  series = {FAccT '22},
2286  pages = {935--947},
2287  publisher = {Association for Computing Machinery},
2288  abstract = {The current AI ethics discourse focuses on developing
2289computational interpretations of ethical concerns, normative
2290frameworks, and concepts for socio-technical innovation. There
2291is less emphasis on understanding how AI practitioners
2292themselves understand ethics and socially organize to
2293operationalize ethical concerns. This is particularly true for
2294AI start-ups, despite their significance as a conduit for the
2295cultural production of innovation and progress, especially in
2296the US and European context. This gap in empirical research
2297intensifies the risk of a disconnect between scholarly research,
2298innovation and application. This risk materializes acutely as
2299mounting pressures to identify and mitigate the potential harms
2300of AI systems have created an urgent need to rapidly assess and
2301implement socio-technical innovation focused on fairness,
2302accountability, and transparency. In this paper, we address this
2303need. Building on social practice theory, we propose a framework
2304that allows AI researchers, practitioners, and regulators to
2305systematically analyze existing cultural understandings,
2306histories, and social practices of ``ethical AI'' to define
2307appropriate strategies for effectively implementing
2308socio-technical innovations. We argue that this approach is
2309needed because socio-technical innovation ``sticks'' better if
2310it sustains the cultural meaning of socially shared (ethical) AI
2311practices, rather than breaking them. By doing so, it creates
2312pathways for technical and socio-technical innovations to be
2313integrated into already existing routines. Against that
2314backdrop, our contributions are threefold: (1) we introduce a
2315practice-based approach for understanding ``ethical AI''; (2) we
2316present empirical findings from our study on the
2317operationalization of ``ethics'' in German AI start-ups to
2318underline that AI ethics and social practices must be understood
2319in their specific cultural and historical contexts; and (3)
2320based on our empirical findings, suggest that ``ethical AI''
2321practices can be broken down into principles, needs, narratives,
2322materializations, and cultural genealogies to form a useful
2323backdrop for considering socio-technical innovations. We
2324conclude with critical reflections and practical implications of
2325our work, as well as recommendations for future research.},
2326  author = {Sloane, Mona and Zakrzewski, Janina},
2327  booktitle = {2022 {ACM} Conference on Fairness, Accountability, and
2329  title = {German {AI} {Start-Ups} and {``AI} Ethics'': Using A Social
2330Practice Lens for Assessing and Implementing {Socio-Technical}


arXiv:2205.03946v2 [cs.CY]
License: cc-by-4.0

Related Posts

Think About the Stakeholders First! Towards an Algorithmic Transparency Playbook for Regulatory Compliance

Think About the Stakeholders First! Towards an Algorithmic Transparency Playbook for Regulatory Compliance

Introduction In the past decade, there has been widespread proliferation of artificial intelligence (AI) systems into the private and public sectors.

How to design an AI ethics board

How to design an AI ethics board

Introduction It becomes increasingly clear that state-of-the-art artificial intelligence (AI) systems pose significant societal risks.

International Institutions for Advanced AI

International Institutions for Advanced AI

Introduction The capabilities of AI systems have grown quickly over the last decade.