Searching for inclusive artificial intelligence for social good: Participatory governance and policy recommendations for making AI more inclusive and benign for society

May combine two claims.

The proportion of women among AI researchers is less than 15% in major AI companies like Facebook and Google, and the percentage of female AI specialists is reported below 25%.

“The proportion of women among AI researchers is less than 15% in major AI companies like Facebook and Google (Nouri, 2021); that percentage of female AI specialists is reported below 25% (WEF, 2021).”

• cites Nouri, 2021Not checked yet

  Not checked in the free run.

• cites WEF, 2021Not checked yet

  Not checked in the free run.

May combine two claims.

Private business representatives participate most actively in discussions of AI ethics, while civil society and ethics experts participate least.

“Because of their business interests in AI systems, private business representatives most actively participate in discussions on AI's ethical dimensions; in contrast, among all actors, civil society and ethics experts participate the least in discussions (Moon et al., 2021).”

• cites Moon et al., 2021Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Many AI-related programs have often been introduced as pilot projects that are sporadic, temporary, and stretched thin and are often initiated by individual researchers, universities, or civil society organizations rather than by large-scale organizations with sustainable funding sources.

“However, many of these AI-related programs have often been introduced as pilot projects that are sporadic, temporary, and stretched thin and are often initiated by individual researchers, universities, or civil society organizations rather than by large-scale organizations with sustainable funding sources.”

May combine two claims.

Technical and institutional obstacles hinder AI's positive impacts on society and prevent those impacts from lasting and scaling up.

“there are still technical and institutional obstacles that not only hinder the positive impacts of AI on society but also prevent the lasting and scaling up of these impacts (Tomašev et al., 2020).”

• cites Tomašev et al., 2020Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

Governments consider the uptake of AI a key strategy to promote future competitiveness and resolve societal challenges concerning economic growth and noneconomic issues such as poverty, sustainability, health, and human wellbeing in both developed and developing countries.

“the uptake of AI has been considered one of the key strategies to promote future competitiveness and resolve future societal challenges concerning economic growth as well as noneconomic issues, such as poverty, sustainability, health, and human wellbeing in both developed and developing countries.”

May combine two claims.

AI's potential impact on the environment is both negative and positive.

“In fact, the potential impact of AI on the environment is both negative and positive.”

May combine two claims.

AI has not been fully addressed as a potential tool for social good, although public administration scholars have begun to pay attention to related aspects of AI.

“AI has not been fully addressed as a potential tool for social good, although scholars of public administration have begun to pay attention to robotic bureaucracy (Bozeman & Youtie, 2020), algorithmic bureaucracy (Kim et al., 2022), the threats to legitimacy caused by algorithmic decisions (Grimmelikhuijsen & Meijer, 2022), AI accountability (Busuioc, 2021), and public values and algorithms (Andre”

• cites Bozeman & Youtie, 2020Not checked yet

  Not checked in the free run.

• cites Kim et al., 2022Not checked yet

  Not checked in the free run.

• cites Grimmelikhuijsen & Meijer, 2022Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

Participatory governance is an appropriate institutional arrangement because stakeholders have asymmetric structures and different perceptions of AI's dangers and benefits.

“Participatory governance has also been presented as an appropriate institutional arrangement because of the asymmetric structure and different perceptions of the dangers and benefits of AI among the various stakeholders.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The World Economic Forum's Global Risks Perception Survey suggested AI could be the riskiest and most beneficial among 12 disruptive technologies.

“the Global Risks Perception Survey of the World Economic Forum (2017) suggested that AI could be the riskiest as well as the most beneficial technology among a selection of 12 other disruptive technologies”

• cites World Economic Forum (2017)Not checked yet

  Not checked in the free run.

May combine two claims.

The population working in AI is neither diverse nor inclusive.

“For example, the population working in AI is neither diverse nor inclusive.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

AI can replace many other machines to enhance energy efficiency and promote sustainability.

“On the other hand, AI can replace many other machines to enhance energy efficiency and promote sustainability.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

AI has a negative impact on the environment because as a machine it inevitably leads to pollution, for example through CO2 emissions.

“On the one hand, AI has a negative impact on the environment because, as a machine, it inevitably leads to pollution, for example, through CO 2 emissions.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

AI technologies have developed rapidly and scaled up dramatically, while preparations for negative consequences such as inequality, partiality, and bias have been insufficient.

“AI technologies have developed rapidly and have been scaled up dramatically, although preparations for the potentially negative consequences of AI, such as inequality, partiality, and bias, have been insufficient.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

Participatory governance is an appropriate governing mechanism for an inclusive AI that contributes to social good.

“it is posited that this governance is an appropriate governing mechanism for an inclusive AI that contributes to social good.”

Probably not a substantial claim: may not pass S4 (based).

AI's benefits have not necessarily been inclusive and equally distributed across social classes within a society and across countries of different income levels.

“the benefits of AI have not necessarily been inclusive and equally distributed across different social classes within a society as well as across countries of different income levels.”

May combine two claims.

The interest-based logic encourages different interest groups to voice and communicate in policy-making, but influence varies among groups.

“The interest-based logic of participatory governance encourages different interest groups to raise their voices and communicate with other interest groups in the policy-making process; however, the degree of influence might vary among different interest groups (Gustafson & Hertting, 2017).”

• cites Gustafson & Hertting, 2017Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Most discussion on AI has concerned economic impact, particularly labor savings and cost efficiency, while little attention has been given to how inclusive AI can be used for social good.

“While most of the discussion on AI has concerned its potential impact on the economy, particularly on labor savings and cost efficiency in the corporate world, little attention has been given to how inclusive AI can be used for social good.”

May combine two claims.

The design, development, and deployment of AI have often been affected exclusively by a small, select group of people in major digital enterprises and select countries, making participation in the AI life cycle extremely exclusive.

“the design, development, and deployment of AI have often been affected exclusively by a small, select group of people working in major digital enterprises and in select countries; thus, participation in the AI life cycle has been by nature extremely exclusive.”

May combine two claims.

By aligning with the SDGs, AI can contribute to social good by contributing to achieving the SDGs, particularly poverty eradication, health and safety, education, justice, and gender equality.

“By aligning with the SDGs, AI can contribute to social good by also contributing to achieving the SDGs, in particular by (1) eradicating poverty and reducing inequality; (2) enhancing health and safety, particularly for less advantaged populations of countries; (3) promoting education (quality education) for less advantaged populations of countries; and (4) promoting justice and gender equality.”

May combine two claims.

Data on underprivileged populations is rarely available or is socially biased.

“However, data on underprivileged populations is rarely available or is socially biased.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

There has been insufficient discussion of AI's role in contributing to social good.

“there has been a lack of discussion concerning its role in contributing to social good.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The affordability of AI-based systems is a constraining factor that limits inclusiveness.

“The affordability of AI-based systems is also a constraining factor that limits inclusiveness.”

Probably not a substantial claim: may not pass S4 (based).

Despite increased efforts, many AI-based solutions have been confined to limited domains where adequate relevant data are available and economies of scale are relatively guaranteed.

“Despite increased efforts to search for AI-based and AI-powered solutions to emerging societal problems, many of these efforts have been confined to limited domains for which adequate and relevant data are available and economies of scale are relatively guaranteed.”

Probably not a substantial claim: may not pass S4 (based).

The quality and utility of AI-based systems in environment, education, and public health are still questionable because the volume and quality of data are not guaranteed.

“Although AI-based systems have been developed in the fields of the environment, education, and public health, the quality and utility of these systems are still questionable because the volume and quality of the data are not guaranteed.”

Classified as probably not a substantial claim.

The three logics of participatory governance are critical procedural mechanisms for AI governance that facilitate communication, coordination, and collaboration among stakeholders, eventually leading to social good.

“The three logics of participatory governance constitute critical procedural mechanisms for AI governance through which a society orchestrates concerted efforts to facilitate communication, coordination, and collaboration among different stakeholders, which eventually leads to social good.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

Inequality caused by AI at national and global levels has been a compelling concern.

“the inequality caused by AI at both the national and global levels has also been a compelling concern.”

May combine two claims.

Certain criminal facial recognition systems, such as Amazon's Rekognition, were banned because of bias against minority populations, and the LAPD stopped using PredPol after controversy over bias against minority communities in 2021.

“For example, certain criminal facial recognition systems (i.e., Amazon's Rekognition system) were banned because of their bias against minority populations (O'Brian, 2019), and the Los Angeles Police Department (LAPD) also decided to stop using PredPol (a predictive policing algorithm system) after a longstanding controversy over bias against minority communities in 2021 (Bhuiyan, 2021).”

• cites O'Brian, 2019Not checked yet

  Not checked in the free run.

• cites Bhuiyan, 2021Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

The European Commission's Ethics Guideline for Trustworthy AI proposes three components: lawful AI, ethical AI, and robust AI.

“The Ethics Guideline for Trustworthy AI proposed by the European Commission (2019) suggests three key components: lawful AI, ethical AI, and robust AI.”

• cites European Commission, 2019Not checked yet

  Not checked in the free run.

May combine two claims.

When data or perceptions of data are biased or skewed, AI systems are likely to become biased and distorted, which reduces public trust in them.

“If the data are biased and skewed, or if the perceptions of the data are skewed, then the AI systems are likely to be biased and distorted accordingly, which will decrease people's trust in those systems.”

May combine two claims.

These figures raise the concern that a biased workforce can produce biased technologies, biased data interpretation, and biased algorithm design.

“These numbers raise concerns about the possibility that a biased workforce may lead to biased technologies, biased data interpretation, and the biased design of algorithms.”

Classified as probably not a substantial claim.

The complex and uncertain nature of AI policies creates governance challenges inherently associated with difficulties in discussing, deciding upon, and implementing AI-related policies.

“Because of the complex and uncertain nature of AI policies, there have been compelling governance challenges inherently associated with difficulties in discussing, deciding upon, and implementing any AI-related policies.”

Classified as probably not a substantial claim. May combine two claims.

The first governance challenge is the difficulty of preparing for potential risks in a timely manner, which has hindered progress in using AI for public good because AI technologies advance rapidly.

“The first challenge has been related to the difficulty of preparing for potential risks in a timely manner; this difficulty has hindered progress in the use of AI for the public good because AI technologies have advanced rapidly.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

Much of the risk associated with AI has emerged from an overemphasis on economic objectives such as business profits while social benefits have been overlooked, underlining the importance of inclusiveness in reaching AI objectives and benefits.

“Much of the risk associated with AI has often emerged as a result of an overemphasis on economic objectives, such as business profits, while social benefits have been overlooked, which underlines how important it is to pursue inclusiveness in reaching the objectives and benefits of AI technologies.”

Classified as probably not a substantial claim.

AI impact assessment is intended to embed social good into the AI lifecycle from design and development, helping to minimize negative risks and uphold ethical principles such as transparency, accountability, explainability, safety, and reliability.

“AI impact assessment is designed to incorporate social good into the lifecycle of AI from the design and developmental stages; this approach helps ensure that the negative risks associated with AI are minimal and comply with AI ethical principles, such as transparency, accountability, explainability, safety, and reliability.”

May combine two claims.

In many countries, the focus of AI policy has remained at the moral stage, where core values and principles are identified for AI systems, but these countries have slowly shifted to explicit policy making.

“Hence, in many countries, the focus of AI policy has remained at the moral stage, at which core values and principles are identified as they apply to the design, development, deployment, and uses of AI systems; however, these countries have slowly shifted from these moral considerations to explicit policy making (Moon & Park, 2020).”

• cites Moon & Park, 2020Not checked yet

  Not checked in the free run.

Classified as probably not a substantial claim.

Trustworthy AI rests on fundamental human rights, including human dignity, individual freedom, democracy, justice, the rule of law, equality, nondiscrimination, solidarity, and citizens' rights.

“The foundations of trustworthy AI are based on fundamental human rights, including (1) respect for human dignity, (2) freedom of individuals, (3) respect for democracy, justice, and the rule of law, (4) equality, nondiscrimination, and solidarity, and (5) citizens' rights;”

• cites European Commission, 2019Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S4 (based).

Because AI is still evolving, it is often very difficult to specify regulations for its development, deployment, and use.

“Considering the fact that AI is still evolving, it is often very difficult to specify regulations applied to the development, deployment, and utilization of AI.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Impact assessment also leads the AI industry and developers to consider how to build inclusivity into AI systems and embed social values from the start of the AI lifecycle.

“Impact assessment also makes the AI industry and AI developers consider possible ways in which to code inclusivity into AI systems and embed social values from the early stages of the AI lifecycle.”

Probably not a substantial claim: may not pass S2 (checkable).

Broad participation and active engagement of different stakeholders constitute a critical part of participatory governance leading to inclusive AI.

“The broad participation and active engagement of different stakeholders constitute a critical part of participatory governance leading to inclusive AI.”

Probably not a substantial claim: may not pass S4 (based).

The underrepresentation of various stakeholders in policy-making processes has been a critical challenge in AI governance.

“Fourth, the underrepresentation of various stakeholders in policy-making processes has also been a critical challenge in AI governance.”

May combine two claims.

The AI workforce lacks diversity, resembling the long-standing diversity problem in STEM fields.

“The AI workforce is far from diverse, which is akin to the diversity problem that has existed in STEM disciplines.”

• cites Nouri, 2021Not checked yet

  Not checked in the free run.

• cites AIAI, 2021Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

The lack of data on underprivileged populations and the danger of biased data remain crucial challenges in developing AI systems.

“While the lack of data on underprivileged populations and the danger of biased data remain crucial challenges in the development of AI systems, the voices of stakeholders have not often been equally heard, considered, and incorporated into AI policies.”

May combine two claims.

Women make up only 15% of AI researchers at Facebook and 10% at Google.

“For example, women researchers account for only 15% and 10% of AI researchers working at Facebook and Google, respectively (Nouri, 2021).”

• cites Nouri, 2021Not checked yet

  Not checked in the free run.

May combine two claims.

Because of the fluid and uncertain nature of AI technology, the scope and scale of AI policy have not been clearly defined, and the regulatory framework has not been easy to determine.

“Because of the fluid and uncertain nature of the technology, the scope and scale of AI policy have not been clearly defined, and the regulatory framework has not been easy to determine.”

May combine two claims.

AI policy is more complicated and complex to govern than many other policies, often causing problems in interagency policy coordination and conflicts among different policy goals and vested interests of policy clientele groups.

“Hence, AI policy is more complicated and complex to govern than many other policies, which often causes problems in interagency policy coordination and conflicts among different policy goals (i.e., data privacy protection versus business opportunities for customized services that use personal data) and the vested interests of particular policy clientele groups (i.e., industry, consumers, or resear”

Probably not a substantial claim: may not pass S3 (load-bearing).

There have been extensive discussions and initiatives on AI ethical guidelines, often called human-centered or trustworthy AI, as proposed by the European Commission.

“To maximize AI's potential benefits while minimizing risks, there have been extensive discussions and initiatives concerning AI ethical guidelines, often called human-centered or trustworthy AI, as proposed by the European Commission (2019).”

• cites European Commission, 2019Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S4 (based).

In developing a pan-governmental AI policy, related agencies can be easily overruled through bureaucratic politics dominated by powerful agencies backed by the private industrial sector and the economy.

“In developing a pan-governmental AI policy, governments must pay close attention to the voices of marginalized populations; moreover, related agencies can be easily overruled through bureaucratic politics occupied by powerful agencies that are supported by the private industrial sector and economy.”

• AssumptionRests on an assumption: AI researchers often influence the Ministry of Education or the Ministry of Science and Technology, embedding their interests in those agencies' policy positions, while the AI industry influences the Ministry of Industry and Trade.

Probably not a substantial claim: may not pass S4 (based).

Because of substantial information asymmetry and uncertainty, as well as predominance of economic objectives and national interests in national competitiveness, AI policy has often been addressed in an unstructured way and loosely handled at national and international levels.

“Because of the substantial information asymmetry and uncertainty, as well as the predominance of economic objectives and national interests in national competitiveness, AI policy has often been addressed in an unstructured way and loosely handled at both the national and international levels.”

Probably not a substantial claim: may not pass S2 (checkable).

Engaging diverse stakeholders, including designers, developers, researchers, and users, is critical to preventing and mitigating risks by incorporating different inputs and perspectives into AI design and development.

“Engaging diverse stakeholders, including not only designers and developers but also researchers and users, is critical to preventing and mitigating risks by incorporating different inputs and perspectives into AI design and development.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

AI policy covers a wide range of issues, including R&D, education, industry, and social welfare, and therefore needs long-term interagency collaboration.

“Because AI policy is related to a wide scope of issues, such as R&D, education, industry, and social welfare, it requires long-term interagency collaboration.”

Probably not a substantial claim: may not pass S4 (based).

AIAs are expected to help implement pan-government AI policy because they formally and informally push AI-related agencies to include basic assessment elements in their policy positions.

“AIAs are expected to be instrumental in the implementation of a pan-government AI policy because they formally and informally force various AI-related agencies to incorporate basic assessment elements into their AI policy positions.”

Probably not a substantial claim: may not pass S4 (based).

Ensuring diversity and inclusion in regulation is more difficult than reaching consensus on AI ethical principles because stakeholder interests and voices are often overrepresented across government agencies.

“In fact, ensuring diversity and inclusion while making regulations is even more challenging than building a consensus on the morality of AI ethical principles because the interests and voices of various stakeholders are often (over)represented across various government agencies.”

May combine two claims.

Globally, women are 50.85% of the population and minorities are 36.9%, but women hold only 17% of technology jobs and women and minorities together make up only 5% of the AI workforce.

“At the global level, women represent 50.85% of the world's population, and 36.9% of the global population are minorities who could potentially enter the workforce; however, only 17% of technology-related employees are women, and only 5% of the AI workforce consists of women and minorities (AIAI, 2021).”

• cites AIAI, 2021Not checked yet

  Not checked in the free run.