This paper has been accepted for publication and is forthcoming in The Global and Digital

Governance Handbook.

Cite as: Choung, H., David, P., & Seberger, J.S. (2023). A multilevel framework for AI
governance. The Global and Digital Governance Handbook. Routledge, Taylor & Francis Group.
A Multilevel Framework for AI Governance 1

A Multilevel Framework for AI Governance

Hyesun Choung, Ph.D.

Michigan State University

Prabu David, Ph.D.

Michigan State University

John S. Seberger, Ph.D.

Drexel University

Abstract

To realize the potential benefits and mitigate potential risks of AI, it is necessary to develop a
framework of governance that conforms to ethics and fundamental human values. Although
several organizations have issued guidelines and ethical frameworks for trustworthy AI, without
a mediating governance structure, these ethical principles will not translate into practice. In this
paper, we propose a multilevel governance approach that involves three groups of interdependent
stakeholders: governments, corporations, and citizens. We examine their interrelationships
through dimensions of trust, such as competence, integrity, and benevolence. The levels of
governance combined with the dimensions of trust in AI provide practical insights that can be
used to further enhance user experiences and inform public policy related to AI.

Chapter Keywords

Artificial Intelligence, Governance, Trust, Ethics, Multilevel Framework

Key Discussion Areas

● Relevance and acceptance of AI by its key stakeholders

● Governance models for ensuring trustworthy AI
● Offering a multidimensional, multilevel model of trust in AI
● Corporate self-governance and government regulation of AI

1. Introduction

Artificial intelligence (AI) systems are becoming increasingly adept in their ability to learn,
reason, self-correct, and emulate human decisions in various domains (Russell et al. 2016; Turing
1950; Watson 2019). Modern AI systems are driven by machine learning (ML) techniques to
accomplish tasks such as recognizing faces and voices, reading X-rays, screening job applicants,
identifying credit card fraud, and enabling law enforcement. Alongside professional and
institutional use of ML-driven AI, consumer applications of AI are increasingly ubiquitous in
daily personal life—embedded in home devices, smart city management, and autonomous driving
systems (Gorwa, Binns, and Katzenbach 2020; Lockey et al. 2021). AI is also used to tackle
challenging social problems, including the maintenance of public health (Wahl et al. 2018), social
justice (Graham and Hopkins 2021), and public safety (Kankanhalli et al. 2019).
Even though governments, corporations, and individuals benefit from the successes of AI
technologies, the detrimental effects of AI are increasingly apparent, including algorithmic biases
A Multilevel Framework for AI Governance 2

and challenges to civil liberties, increased surveillance (Whittaker et al. 2018), and the diminution
of human agency (Seberger 2021). To realize the potential benefits of AI, it is necessary to
address the detrimental effects and risks of AI. Such a process is further complicated by AI’s
reliance on ML algorithms that have been characterized as a black box, because the inner
workings of these algorithms are opaque and not easily explainable (Barredo Arrieta et al. 2020).
Though some researchers have developed methods to translate the deep learning inside the black
box into formal rules, this area is still nascent (Samek et al. 2019)—hence the legitimately
pressing concerns about the lack of transparency and understanding in current AI technologies
(Shin and Park 2019). The pervasiveness of AI and the autonomous decisions it renders in many
areas of life, compounded by its black-box approach to sensitive data, have created an urgent
need for well-defined and human-centered principles of governance that conform to ethics and
fundamental human values (Jobin, Ienca, and Vayena 2019). Such principles of governance may
mitigate the risks of AI and make it possible to realize AI’s full potential.
In the past few years, influential groups and organizations have addressed the need for
governance by issuing guidelines and ethical frameworks for the development and deployment
of trustworthy AI. These include the Organization for Economic Co-operation and Development
(OECD 2019), the European Commission’s High Level Expert Group on Artificial Intelligence
(AI HLEG 2019), the Institute of Electrical and Electronics Engineers (IEEE), Microsoft,
DeepMind, and Google (Hagendorff 2020; Jobin, Ienca, and Vayena 2019), organizations driven
by different motives, ethical imperatives, and missions.
However, in the absence of a mediating governance structure, ethical principles do not
automatically translate into practice (Mittelstadt 2019). Given the wide reach of AI across social,
cultural, and national borders, governance structures for AI require guiding principles sensitive
to the contexts in which they will be deployed. For example, the concept of privacy has widely
variable definitions across cultures (Farrall 2008). A codified principle for ethical AI that involves
privacy would, therefore, need to be sensitive to such cultural variance. While individual rights
and autonomy are foundational in the AI HLEG (2019) ethical guidelines, such Western
constructs may not map cleanly or ethically to social contexts where individual rights are
subsidiary to the rights of the political party and the state. Despite the complexity of intercultural
differences, three groups of dynamically interdependent stakeholders are universally implicated
in the concern for ethical AI governance: governments, corporations, and citizens. Any baseline
for AI governance, then, necessarily involves these three groups and their interrelationships. We
approach such interrelationships through the lens of trust.

2. A Multilevel Approach to AI Governance for Trust-Enhancing Practices

In the broadest sense, trust is defined as a confident relationship with the unknown (Botsman
2017). As we—users, scholars, designers, and policy-makers—prepare for a future with
ubiquitous AI and its unknown dimensions, trust is essential to realizing a future in which the
risks associated with machine intelligence are mitigated and its benefits are nurtured. Effective
governance is a cornerstone of trust and is necessary to mediate the relationship between humans
and AI technologies. A framework for governance of AI can be achieved only by acknowledging
the agency of governments, organizations, and citizens, and their interdependencies.
In the following sections, we review governance principles from three perspectives: the
transnational/national (i.e., governmental), the organizational (i.e., corporate), and the individual.
We provide a detailed review of trust at the individual level from a psychological perspective.
This review is followed by an examination of ethical frameworks for governance that have been
A Multilevel Framework for AI Governance 3

adopted by the European Union (EU), IEEE, and Big Tech. We conclude by offering a multilevel
framework of AI governance that is built on trust and guided by ethics.

3. International and National Governance

Amid the fierce competition among countries for global leadership in AI (Castro 2019), some
efforts toward international consensus building have been undertaken. The International Panel on
Artificial Intelligence (IPAI) at the Group of Seven (G7) summit in 2019 aimed to follow the
example of the International Panel on Climate Change (IPCC) to support the responsible
development of AI “grounded in human rights, inclusion, diversity, and innovation”
(Government of Canada 2019). In 2020, the Group of Twenty (G20) countries committed to
advance the G20 AI Principles drawn from the OECD’s recommendations about AI. These
principles seek to foster public trust and confidence in AI by promoting values such as
inclusiveness, human-centricity, transparency, robustness, and accountability (Box 2020).
Given AI’s potential to improve health, education, agriculture, manufacturing, and energy,
countries and international bodies are unsurprisingly eager to develop and adopt AI. At the same
time, such eagerness is countered by the perceived threat of machine agency and automated
decision-making, as well as the displacement of humans by automation and robots. Such risks
associated with AI are perceived differently in different societies relative to dominant political
ideology, value systems, and culture norms. For example, citizens from traditional collectivist
cultures, such as China, may be more inclined than citizens from individualist cultures to sacrifice
personal data and choose security over privacy (Kostka, Steinacker, and Meckel 2021). Despite
the unavoidable cultural and contextual differences, a mix of international principles and policies
are urgently needed to build trustworthy and human-centered AI (OECD 2020).
Such efforts are underway. The Beijing Academy of Artificial Intelligence has released a set of
principles endorsed by leading Chinese universities and organizations (Beijing AI Principles
2019), which includes basic human values such as the good of humanity, diversity and inclusion,
and ethics as founding principles. Similar values were highlighted by the European Commission’s
AI HLEG, including human dignity and autonomy, prevention of harm, fairness, explicability,
accountability, privacy, and social and environmental well-being (AI HLEG 2019). Despite the
similarities, the differences in political systems in China and the EU will eventually shape the
implementation of these values.
In the United States, despite efforts to pass AI governance policies in state legislatures (National
Conference of State Legislatures 2021), there have been no major AI-specific legislative
successes at the federal or state level. One related legislation, however, is the California
Consumer Protection Act (CCPA), which focuses on data privacy and generally emulates the
General Data Protection Regulation (GDPR) enacted by the EU (Somers and Boghaert 2018).
Though not directly related to AI, the CCPA could have a significant effect on how the data
practices of AI technologies become normalized.
In summary, international bodies and nations have proposed key principles of trustworthy AI.
Many of these principles are founded on the values of human rights, such as dignity, freedom,
and autonomy. Such principles optimistically envisage a future in which human intelligence
coexists with artificial intelligence. However, enforcement or the real-world implications of these
policies on technology firms, software developers, and data brokers is not clear. As described in
the next section, technology firms and oversight bodies have taken it upon themselves to develop
individual policies in the tradition of corporate self-governance.
A Multilevel Framework for AI Governance 4

4. Corporate Self-Governance

AI is designed and deployed by corporations, who have their own role to play in governance. In
addition to implementing national policy, industries and corporations can complement
government regulations with self-governance. A notable advance is a call to action for ethically
aligned design for business created by the IEEE Global Initiative on Ethics of Autonomous and
Intelligent Systems (Chatila and Havens 2019). The IEEE’s policy calls for participatory design
and attention to ethics throughout the AI development cycle. Such guidance frames AI
governance as a sociotechnical necessity that could positively impact humanity by underscoring
the importance of trust, transparency, and accountability among corporate leaders.
Similar policies have been adopted by Big Tech companies. Corporations such as Google and
Microsoft have developed policies, which are posted on their websites. Google’s policy states
that the company is optimistic about AI and is sensitive to its potential harm (Google, n.d.). Key
values that inform Google’s position include empowering humanity, uplifting society, and
addressing fairness and bias. Concern for privacy and accountability are also mentioned: Google
states that it will not use surveillance that violates internationally accepted norms, nor will it
pursue weapons or systems intended principally to cause harm. Microsoft similarly emphasizes
responsible and trustworthy AI that empowers others and encourages the use of AI for socially
responsible outcomes (Microsoft, n.d.). The importance of trust is widely discussed in
Microsoft’s policy. Other companies like Apple, Amazon, and Facebook employ AI in their
consumer-facing products and demonstrate a sensitivity to trust in written statements, which is
integral to corporate reputation. It appears that Big Tech companies are keenly aware of the
importance of trust. Their approach to self-governance hinges on building trust through ethical
practices promoted in their AI policies. It remains to be seen, however, how accurately such
written statements reflect the real-world practices of such companies.
In the United States, tech corporations exercise significant influence on both technological
development and regulatory methods (Cihon, Schuett, and Baum 2021). Although companies
formulate their own ethical guidelines or adopt ethically motivated self-commitments, they
currently do not have clear standards for practice or internal assessors and are hesitant to create a
binding legal framework (Hagendorff 2020). One way to enhance corporate self-governance is
to assess and mitigate risk from design to implementation and to institute a certification or
accreditation program for AI developers who are trained to anticipate risks (Roski et al. 2021).
Certification would also include training in ethics, and some have recommended virtue ethics as
a promising approach (Hagendorff, 2020), which differs from traditional ethics approaches in
that it locates ethics in specific contexts dependent upon traits of individuals rather than universal
codes of conduct. Another suggestion is developing and initiating “ethics-based auditing” for
organizations and academic researchers who develop or deploy AI (Mökander and Axente 2021).
Through self-governance and ethics in training and evaluation processes, corporations have the
power to shape our shared computational futures. However, as in many areas, corporate self-
governance can be held more accountable through higher-level policy from governments that
emphasize common good over corporate gain. Much like the added accountability that accrues
from top-down governmental policies, bottom-up demand from citizens can drive accountability
in corporate governance.

5. AI Literacy and Governance by Citizens

Citizens have a right to trustworthy AI. To empower individuals to claim agency, enhancing
citizens’ competencies and literacy in AI is essential. User empowerment requires interventions
A Multilevel Framework for AI Governance 5

that offer an alternative to the digital resignation — a sociotechnical phenomenon resembling

learned helplessness in which users resign themselves to having little agency over how their data
is used — that some individuals have grown to accept (Seberger et al. 2021). It begins with
providing users a meaningful understanding of a broad set of sociotechnical factors that
contribute to and characterize their use of an AI technology. AI literacy has been defined as “a
set of competencies that enables individuals to critically evaluate AI technologies; communicate
and collaborate effectively with AI; and use AI as a tool online, at home, and in the workplace”
(Long and Magerko 2020, 598). Recent efforts to foster AI literacy in the United States include
an open learning opportunity for underrepresented students (AI4ALL 2019) and integration of
AI programs into K-12 curricula (Zimmerman 2018).
Given the rapid pace at which AI is evolving, however, it is unclear how prior AI literacy
campaigns remain relevant to the changing landscape of AI. For example, recently, ChatGPT
developed by OpenAI has taken the world by storm because of its natural language processing
ability and fluency in writing. While the responses provided by the ChatGPT may sound
convincing, until the technology matures, literacy interventions may be required to train citizens
to critically evaluate the quality and veracity of responses.
However, AI is a relatively new technology, and the majority of citizens hold optimistic
perspectives about AI without a sophisticated knowledge of potential risks (Carrasco et al. 2019).
Relying purely on citizen action for an emerging technology whose future is unknown is
unrealistic and even unfair. In the long term, such an approach may amount to negligence. While
government- and corporate-led efforts to inform and educate the public are essential, an integrated
approach that includes citizens, corporations, and government is likely the most responsible and
robust approach.

6. Psychology of Trust

To develop a model of governance for a new and potentially disruptive technology that touches
the lives of citizens, trust is essential (Shin 2021; Wu et al. 2011). But the beginning of the 21st
century has been marked by a significant erosion of trust in institutions and governments
(Edelman 2021). The logical response to this crisis of trust is to build systems that instill
confidence among citizens from different cultural backgrounds and values systems (Gillath et al.
2021; Thiebes, Lins, and Sunyaev 2020).
Although earlier we defined trust as “a confident relationship with the unknown” (Botsman
2017), here we examine trust in depth as a cornerstone of all relationships. Among individuals,
trust is defined as “a psychological state comprising the intention to accept vulnerability based
upon positive expectations of the intentions or behavior of another” (Rousseau et al. 1998, 395).
Besides vulnerability and positive expectations of one another, trust combines characteristics,
intentions, and behaviors (J. D. Lee and See 2004; Mayer, Davis, and Schoorman 1995) and is
required to build mutuality and interdependence between parties.
Trust in humans can be defined as an amalgam of one’s belief in another’s ability, benevolence,
and integrity (Mayer, Davis, and Schoorman, 1995). Ability refers to skills and competencies to
successfully complete a given task. Benevolence pertains to whether the trusted individual or
party has positive intentions that are not based purely on self-interest. And integrity points to the
sense of morality and justice of the trusted party and is related to attributes such as consistency,
predictability, and honesty. Absent such characteristics, meaningful interpersonal interactions
cannot be forged.
A Multilevel Framework for AI Governance 6

Though interpersonal trust has been applied to technology with human-like characteristics
(Calhoun et al., 2019; Gillath et al. 2021), some researchers caution that the principles of trust
among humans cannot be applied directly to human-to-machine trust (Madhavan and Wiegmann
2007). Researchers have noted that trust in technology is qualitatively different from trust in
people (McKnight et al. 2011). The key distinction is that a human is a moral agent, whereas
technology lacks volition and moral agency. However, given the extent to which AI technologies
manifest the value system and agency of the designer, this may no longer be a tenable distinction.
Beginning with the dimensions of trust in human interaction, McKnight and colleagues (2011)
revised the three dimensions of trust in technology as functionality, reliability, and helpfulness.
Functionality refers to the capability of the technology, which the authors likened to human
ability. Reliability is the consistency of operation, which is analogous to integrity. Helpfulness
indicates whether a specific technology is useful to users and maps onto the benevolence
dimension of human trust.
But AI is more than just a technology—it is an emergent sociotechnical system that strives for
autonomy by replacing tasks and decisions made by humans. In that sense, McKnight’s definition
and the dimensions of trust in technology need to be adapted to the AI context. Unlike prior
information technologies that relied on user input and the execution of rules programmed by
humans, AI-driven technologies are capable of learning on their own and exercising functional
autonomy, such as making decisions. AI’s ostensible “intelligence” can be traced to its contextual
awareness, data collection, processing, and decision-making abilities.
The autonomy and decision-making of AI in social domains, such as interpreting language,
recognizing faces, or screening job applications, have led experts to sound the alarm on potential
risks. Lankton, McKnight, and Tripp (2015) found that, when humanness is embedded or
perceived in the technology, a trust-in-humans scale works better than a trust-in-technology scale.
And when these technologies are depicted as capable of human qualities, including reasoning and
motivations, they can induce high expectations and initial trust (Glikson and Woolley 2020).
These findings suggest that trust in the human dimension of technology is a dynamic concept that
varies considerably based on the context and the characteristics of the trusted agent. In short, trust
in AI includes components of trust in people and components of trust in technology (Choung,
David, and Ross 2022b).
Trust in automation (J. D. Lee and See 2004) is another framework relevant to the
conceptualization of trust in AI and is anchored on performance, process, and purpose.
Performance in automation refers to operational characteristics, including its reliability and
ability. Process corresponds to the consistency of behaviors. Purpose describes why the
automation was developed and the designers’ intent. Table 6.1 summarizes the dimensions of
trust in people, technology, and automation.
A Multilevel Framework for AI Governance 7

Table 6.1 Dimensions of Trust

Basis of Trust in People Basis of Trust in Technology Basis of Trust in

Automation
Mayer, Davis, and Schoorman Lankton, McKnight, and Tripp
(1995) (2015) Lee and Moray (1992)

Competence/ability Functionality Performance

Integrity Reliability Process

Benevolence Helpfulness Purpose

Multiple approaches to framing trust in technologies have been developed because our propensity
to trust machines is interesting in and of itself. Researchers have identified contexts in which
humans are more willing to entrust personal information to machines than to other humans, which
has been explained as the machine heuristic (Sundar and Kim 2019). This is a robust effect that
can be elicited by adding simple interface cues that suggest that the user is interacting with a
machine. In turn, these cues are enough to prime machine characteristics such as accuracy,
objectivity, neutrality, and unbiasedness. While some of these characteristics may be well
deserved, some of this trust may be misplaced because machine learning algorithms that rely on
data that represent human behaviors and practices may be inadvertently perpetuating human
biases in AI.
Positive appraisals of machines also extend to algorithmic appreciation (Logg, Minson, and
Moore 2019), the belief that algorithms are smarter, more objective, better decision-makers than
humans. The countervailing perception to algorithmic appreciation is algorithmic aversion, the
tendency to be critical of algorithms for their reductionism, lack of humanity, and subjective
thinking (Dietvorst, Simmons, and Massey, 2015, 2018). The psychology of machine heuristics,
algorithmic appreciation, and algorithmic aversion justify the need for a better understanding of
the relational attributes of trust between humans and machines. As governance policies for AI are
designed, one must pay heed to warnings that the human tendency to treat computers as social
actors is changing rapidly given our growing experience with computers and AI (Gambino, Fox,
and Ratan 2020). We must develop a suitable understanding of trust that is sensitive to the
emerging relationships between humans and AI.

7. Propensity to Trust

In the previous section, we focused on the dimensions of trust. Now we return to the multilevel
conceptualization of governance with individuals, corporations, and governments as stakeholders
and the trust propensity at each level. The multilevel conceptualization combined with trust
propensities accommodates interdependencies and power differences among the three
stakeholder groups, resulting in a governance model that spans from the individual (i.e.,
intrapersonal and interpersonal) to the collective (i.e., institutions and society) (Fulmer and Dirks
2018). Such a multilevel framework of trust furthers our understanding of the process through
which trust evolves and is crystallized over time (Hoff and Bashir 2015).
Trust at the individual level begins as a disposition or a general tendency to trust another person
(Mayer, Davis, and Schoorman1995). This is known as propensity of trust and is predictive of
initial trustworthiness (Alarcon et al. 2018; Colquitt, Scott, and LePine 2007). Much like the
A Multilevel Framework for AI Governance 8

propensity to trust people, propensities to trust corporations, institutions, industries, and

governments must be considered when building AI systems that aspire to trustworthiness. For
example, an individual with a high trust propensity for other individuals may have a low trust in
Facebook and support social media regulation but support development of AI for social good. A
multilevel, multidimensional understanding of trust would allow for a governance design that can
be principled and yet flexible to accommodate such nuances and complexity.

8. Ethics and Trust Lenses in the Multilevel Framework

Corporations (e.g., Google, Microsoft) and national and international bodies (e.g., IEEE,
European Commission) have identified ethical values and requirements that include the following
themes: privacy protection, fairness, diversity, nondiscrimination and social justice,
accountability, robustness, safety, resilience, transparency, explainability, human autonomy, loss
of human jobs, need for human oversight, and limiting the use of AI as a weapon in wars
(Hagendorff 2020). While there is significant overlap in the values and themes emphasized by
different groups, the policy from the European Commission’s AI HLEG (2019) offers a
framework that is sufficiently nuanced and is an inspiring call to action. The AI HLEG framework
rests on four ethical values—human autonomy, prevention of harm, fairness, and explicability.
In addition, it underscores the rights of vulnerable members of society and the historically
marginalized. The authors recognize the potential benefits and risks of AI and the need for risk
mitigation. Drawing from these values and ethical principles, seven requirements are offered (see
table 8.1).

Table 8.1 Seven Ethics Requirements for Trustworthy AI Proposed by the European
Commission’s High Level Expert Group on Artificial Intelligence

Ethics Requirements for

Description
Trustworthy AI

Human agency and oversight AI systems should allow people to make informed
decisions. There should be a human oversight
mechanism through a “human-in-the-loop” approach.

Technical robustness and safety AI systems should be safe, reliable, and reproducible to
minimize unintended harm.

Privacy and data governance Ensure privacy and data protection, which requires
adequate data governance framework.

Transparency AI systems and business models should be transparent,

and the AI systems’ decisions should be explainable to
the stakeholders. People need to be informed about the
systems’ capabilities and limitations.

Diversity, nondiscrimination, AI systems should be accessible to all, and unfair biases

and fairness should be avoided. Minimizing algorithmic bias is also
important.
A Multilevel Framework for AI Governance 9

Societal and environmental AI systems should benefit human beings and they should
well-being take into account the social impact and environmental
consequences.

Accountability AI mechanisms should be put in place to ensure

responsibility and accountability for AI systems and
their outcomes.

These requirements contribute to the three trust dimensions of people: competence, integrity, and
benevolence. These requirements of trust can help to identify possible foundations of trustworthy
AI (Choung, David, and Ross, 2022a). For example, the competence dimension encompasses
aspects of how an AI system functions, taking into consideration such aspects as safety,
robustness, accountability, and explainability. The integrity dimension focuses on such human-
centered characteristics of trust as fairness, nondiscrimination, privacy, and transparency. In
concert, integrity and competence serve as guardrails for the sociotechnical configuration of AI.
While these two dimensions may be sufficient to build trustworthy AI in the short term, the future
of AI as a cohabitant of the human ecosystem requires enlightened approaches to computing in
which machines must be trained to be benevolent. The best part of our humanity is in our
willingness to look past individual self-interest and to make commitments and sacrifices for the
common good. Creative approaches to nurture such benevolent values like respect for human
autonomy, social justice, social and environmental well-being, and compassionate computing are
emerging areas that should be considered as an ethical lens for trustworthy AI of the future.
A Multilevel Framework for AI Governance 10

Figure 8.1 The Framework of Multilevel Governance and Trust in AI

In the multilevel framework we outline in figure 8.1, each level—governmental, corporate, and
individual—serves as a lens through which to interrogate the impacts of AI on the complex
assemblage of people, corporations, and governments. More specifically, through such lenses we
might systematically observe and test the effects of different attempts at trustworthiness in AI,
ultimately arriving at an understanding of an emergent complex sociotechnical system that has
trust as its core. While responsibility of governance will fall heavily on the corporations using
and deploying AI, what corporations do will depend on top-down policy-making by governments
and bottom-up demands for trustworthy AI by people. In this regard, AI appears as a call to
agency even as the functions of AI challenge the agency of end users. The enforcement of
governance at the corporate level can be achieved by ethics-based auditing (Mökander and
Axente 2021). Although audits create an added layer of bureaucracy, which could slow down
innovation, they can take on different forms based on the scope and context of the project. Indeed,
when the credo best defining contemporary techno-solutionism is, “move fast and break things,”
additional bureaucratic checks may well prove beneficial. Much like universities, which rely on
internal review boards to ensure the ethics and integrity of research, corporations could have their
internal review boards with different levels of scrutiny, from expedited to full review. Broader
scale review boards, such as those like the Food and Drug Administration (FDA), may also be
developed to provide external insurance.

9. Virtue Ethics

The ethics of trustworthy AI can be evaluated in relation to three groups of stakeholders:

governments, corporations, and people. The conceptual boundaries of trust for each of the three
groups of stakeholders and the interdependencies among them can be assessed using three
A Multilevel Framework for AI Governance 11

dimensions of trust: ability, integrity, benevolence. Such dimensions should be bolstered by

engagement with contemporary research on the nature of human trust and trust propensities for
machines, humans, and institutions. Trustworthy AI evolves in tandem with the ethics of AI. A
framework of ethics that draws from the three widely recognized ethical approaches –
deontological, utilitarian, and virtues – could further improve our trust in AI.
The deontological approach based on rules, and the utilitarian approach based on the cost-benefit
calculus of maximizing good and minimizing harm to the largest group of individuals can both
inform the development of governance principles. In addition, the virtue-ethics approach
(Hagendorff 2022; Vallor 2016) can be a promising addition to nurture the right values in
individual developers of AI.
While both the deontological and utilitarian approaches have an important role in macro-level
perspectives, virtue ethics, which focuses on the individual, is a burgeoning area that can have an
immediate and lasting impact. In other words, top-down ethical principles and actions can serve
as extrinsic motivation, and virtue ethics can serve as intrinsic motivation.
Virtue ethics aims to nurture in individuals their own internal compass of values and virtues along
with guides to implement them into practice. Virtue ethics inculcates sensibilities like fairness,
kindness, compassion, and generosity and examines these virtues in specific contexts relevant to
a domain, such as privacy or surveillance. Such investments in training in virtue ethics
complemented by ethics-based auditing or internal review processes appear to be promising
avenues to build trustworthy AI within the proposed multilevel framework.

10. Conclusion

AI is an exciting technology with enormous risks and benefits. Until recently, the technology has
been developed without sufficient consideration of its social implications. Given the pervasive
nature of AI and consequential decisions it makes, governance policies are needed. In this chapter
we have outlined a framework that emphasizes the interdependencies among three stakeholders
– governments, corporations, and citizens. Further, we offer competence, integrity and
benevolence dimensions of trust as lenses to examine the interrelationships among the
stakeholder groups. We concluded by offering AI ethics as a mechanism to build trust among
stakeholders and identify virtue ethics. Along with rule-based ethics, virtue ethics is a promising
approach to nurture the right sensibilities required for AI governance.
A Multilevel Framework for AI Governance 12

References

AI HLEG. 2019. “Ethics guidelines for trustworthy AI.” European Commission, November 8,
2019, https://data.europa.eu/doi/10.2759/346720.
AI4ALL. (2019). “AI4ALL.” https://ai-4-all.org.
Alarcon, G. M., J. B. Lyons, J. C. Christensen, S. L. Klosterman, M. A. Bowers, T. J. Ryan, S.
A. Jessup, and K. T. Wynne. 2018. “The Effect of Propensity to Trust and Perceptions of
Trustworthiness on Trust Behaviors in Dyads.” Behavior Research Methods 50(5): 1906–
20. doi: 10.3758/s13428-017-0959-6.
Barredo Arrieta, A., N. Díaz-Rodríguez, J. Del Ser, A. Bennetot, S. Tabik, A. Barbado, S. Garcia,
S. Gil-Lopez, D. Molina, R. Benjamins, R. Chatila, and F. Herrera. 2020. “Explainable
Artificial Intelligence (XAI): Concepts, Taxonomies, Opportunities, and Challenges
toward Responsible AI.” Information Fusion 58: 82–115. doi:
10.1016/j.inffus.2019.12.012.
Beijing AI Principles. 2019. BAAI, May 25, 2019, https://www.baai.ac.cn/news/beijing-ai-
principles-en.html.
Botsman, R. 2017. Who Can You Trust? How Technology Brought Us Together and Why It Might
Drive Us Apart, 1st ed. New York: Public Affairs.
Box, S. 2020. July 24). “How G20 Countries Are Working to Support Trustworthy AI.” OECD
Innovation Blog, July 24, 2020, https://oecd-innovation-blog.com/2020/07/24/g20-
artificial-intelligence-ai-principles-oecd-report.
Calhoun, C. S., P. Bobko, J. J. Gallimore, and J. B. Lyons. 2019. “Linking Precursors of
Interpersonal Trust to Human-Automation Trust: An Expanded Typology and
Exploratory Experiment.” Journal of Trust Research 9(1): 28–46. doi:
10.1080/21515581.2019.1579730.
Carrasco, M., S. Mills, A. Whybrew, and A. Jura. 2019. “The Citizen’s Perspective on the Use
of AI in Government.” BCG Digital Government Benchmarking, March 1, 2019,
https://www.bcg.com/publications/2019/citizen-perspective-use-artificial-
intelligence-government-digital-benchmarking.
Castro, D. 2019. “Who Is Winning the AI Race: China, the EU or the United States?” Center for
Data Innovation, August 19, 2019, https://datainnovation.org/2019/08/who-is-winning-
the-ai-race-china-the-eu-or-the-united-states.
Chatila, R., and J. C. Havens. 2019. “The IEEE Global Initiative on Ethics of Autonomous and
Intelligent Systems.” In Robotics and Well-Being, vol. 95, edited by M. I. Aldinhas
Ferreira, J. Silva Sequeira, G. Singh Virk, M. O. Tokhi, & E. E. Kadar. Basel,
Switzerland: Springer International. doi: 10.1007/978-3-030-12524-0_2.
Choung, H., David, P., and Ross, A. 2022a. “Trust and Ethics in AI.” AI & Society. doi:
10.1007/s00146-022-01473-4.
Choung, H., David, P., and Ross, A. 2022b. “Trust in AI and Its Role in the Acceptance of AI
Technologies.” International Journal of Human-Computer Interaction. doi:
10.1080/10447318.2022.2050543.
Cihon, P., J. Schuett, and S. D. Baum. 2021. “Corporate Governance of Artificial Intelligence in
the Public Interest.” Information 12(7): 275. doi: 10.3390/info12070275.
Colquitt, J. A., B. A. Scott, and J. A. LePine. 2007. “Trust, Trustworthiness, and Trust Propensity:
A Meta-analytic Test of Their Unique Relationships with Risk Taking and Job
Performance.” Journal of Applied Psychology 92(4): 909–27. doi: 10.1037/0021-
9010.92.4.909.
A Multilevel Framework for AI Governance 13

Dietvorst, B. J., J. P. Simmons, and C. Massey. 2015. “Algorithm Aversion: People Erroneously
Avoid Algorithms after Seeing Them Err.” Journal of Experimental Psychology: General
144(1): 114–26. doi: 10.1037/xge0000033.
———. 2018. “Overcoming Algorithm Aversion: People Will Use Imperfect Algorithms If They
Can (Even Slightly) Modify Them.” Management Science 64(3): 1155–70. doi:
10.1287/mnsc.2016.2643.
Edelman. 2021. “Edelman Trust Barometer 2021.” Annual Edelman Trust Barometer, p. 58.
https://www.edelman.com/sites/g/files/aatuss191/files/2021-
03/2021%20Edelman%20Trust%20Barometer.pdf.
Farrall, K. N. 2008. “Global Privacy in Flux: Illuminating Privacy across Cultures in China and
the U.S.” International Journal of Communication 2: 993–1030.
Fulmer, A., and K. Dirks. 2018. “Multilevel Trust: A Theoretical and Practical Imperative.”
Journal of Trust Research 8(2): 137–41. doi: 10.1080/21515581.2018.1531657.
Gambino, A., J. Fox, and R. Ratan. 2020. “Building a Stronger CASA: Extending the Computers
Are Social Actors Paradigm.” Human-Machine Communication 1: 71–86. doi:
10.30658/hmc.1.5.
Gillath, O., T. Ai, M. S. Branicky, S. Keshmiri, R. B. Davison, and R. Spaulding. 2021.
“Attachment and Trust in Artificial Intelligence.” Computers in Human Behavior
115(52): 106607.
Glikson, E., and A. W. Woolley. 2020. “Human Trust in Artificial Intelligence: Review of
Empirical Research.” Academy of Management Annals 14(2): 627–60. doi:
10.5465/annals.2018.0057.
Google. n.d. “AI at Google: Our Principles.” Google AI. https://ai.google/principles (retrieved on
October 20, 2021).
Gorwa, R., R. Binns, and C. Katzenbach. 2020. “Algorithmic Content Moderation: Technical and
Political Challenges in the Automation of Platform Governance.” Big Data & Society
7(1): 2053951719897945. doi: 10.1177/2053951719897945.
Government of Canada. 2019. “Declaration of the International Panel on Artificial Intelligence.”
Innovation, Science and Economic Development Canada [Backgrounders], May 16,
2019, https://www.canada.ca/en/innovation-science-economic-
development/news/2019/05/declaration-of-the-international-panel-on-artificial-
intelligence.html.
Graham, S. S., and H. R. Hopkins, 2021. “AI for Social Justice: New Methodological Horizons
in Technical Communication.” Technical Communication Quarterly 31(1): 89–102. doi:
10.1080/10572252.2021.1955151.
Hagendorff, T. 2020. “The Ethics of AI Ethics: An Evaluation of Guidelines.” Minds and
Machines 30(1): 99–120. doi: 10.1007/s11023-020-09517-8.
Hoff, K. A., and M. Bashir. 2015. “Trust in Automation: Integrating Empirical Evidence on
Factors That Influence Trust.” Human Factors: The Journal of the Human Factors and
Ergonomics Society 57(3): 407–34. doi: 10.1177/0018720814547570.
Jobin, A., M. Ienca, and E. Vayena. 2019. “The Global Landscape of AI Ethics Guidelines.”
Nature Machine Intelligence 1(9): 389–99. doi: 10.1038/s42256-019-0088-2.
Kankanhalli, A., Y. Charalabidis, and S. Mellouli. 2019. “IoT and AI for Smart Government: A
Research Agenda.” Government Information Quarterly 36(2): 304–09. doi:
10.1016/j.giq.2019.02.003.
Kostka, G., L. Steinacker, and M. Meckel. 2021. “Between Security and Convenience: Facial
Recognition Technology in the Eyes of Citizens in China, Germany, the United Kingdom,
and the United States.” Public Understanding of Science 30(6): 671–90. doi:
10.1177/09636625211001555.
A Multilevel Framework for AI Governance 14

Lankton, N., D. H. McKnight, and J. Tripp. 2015. “Technology, Humanness, and Trust:
Rethinking Trust in Technology.” Journal of the Association for Information Systems
16(10): 880–918. doi: 10.17705/1jais.00411.
Lee, J. D., and K. A. See. 2004. “Trust in Automation: Designing for Appropriate Reliance.”
Human Factors 46(1): 50–80. doi: 10.1518/hfes.46.1.50_30392.
Lee, J., and N. Moray. 1992. “Trust, Control Strategies and Allocation of Function in Human-
Machine Systems.” Ergonomics 35(10): 1243–70. doi: 10.1080/00140139208967392.
Lockey, S., N. Gillespie, D. Holm, and I. A. Someh. 2021. “A Review of Trust in Artificial
Intelligence: Challenges, Vulnerabilities and Future Directions.” Hawaii International
Conference on System Sciences, January 5, 2021. doi: 10.24251/HICSS.2021.664.
Logg, J. M., J. A. Minson, and D. A. Moore. 2019. “Algorithm Appreciation: People Prefer
Algorithmic to Human Judgment.” Organizational Behavior and Human Decision
Processes 151: 90–103. doi: 10.1016/j.obhdp.2018.12.005.
Long, D., and B. Magerko. 2020. “What Is AI Literacy? Competencies and Design
Considerations.” Proceedings of the 2020 CHI Conference on Human Factors in
Computing Systems, April 23, 2020. doi: 10.1145/3313831.3376727.
Madhavan, P., and D. A. Wiegmann. 2007. “Effects of Information Source, Pedigree, and
Reliability on Operator Interaction with Decision Support Systems.” Human Factors: The
Journal of the Human Factors and Ergonomics Society 49(5): 773–85. doi:
10.1518/001872007X230154.
Mayer, R. C., J. H. Davis, and F. D. Schoorman. 1995. “An Integrative Model of Organizational
Trust: Past, Present, and Future.” Academy of Management Review 20(3): 709–34.
Mcknight, D. H., M. Carter, J. B. Thatcher, and P. F. Clay. 2011. “Trust in a Specific Technology:
An Investigation of Its Components and Measures.” ACM Transactions on Management
Information Systems 2(2): 1–25. doi: 10.1145/1985347.1985353.
Microsoft. n.d. “Responsible AI Principles from Microsoft.” Microsoft.
https://www.microsoft.com/en-us/ai/responsible-ai (retrieved on October 20, 2021).
Mittelstadt, B. 2019. “Principles Alone Cannot Guarantee Ethical AI.” Nature Machine
Intelligence 1(11): 501–07. doi: 10.1038/s42256-019-0114-4.
Mökander, J., and M. Axente. 2021. Ethics-Based Auditing of Automated Decision-Making
Systems: Intervention Points and Policy Implications.” AI & SOCIETY, October 27, 2021.
doi: 10.1007/s00146-021-01286-x.
National Conference of State Legislatures (NCSL). 2021. “Legislation Related to Artificial
Intelligence.” NCSL, September 15, 2021,
https://www.ncsl.org/research/telecommunications-and-information-technology/2020-
legislation-related-to-artificial-intelligence.aspx.
Organization for Economic Co-operation and Development (OECD). 2019. Artificial Intelligence
in Society. Paris: OECD Publishing. doi: 10.1787/eedfee77-en
———. (2020). Examples of AI National Policies: Report for the G20 Digital Economy Task
Force. Saudi Arabia. https://www.oecd.org/sti/examples-of-ai-national-policies.pdf.
Roski, J., E. J. Maier, K. Vigilante, E. A. Kane, and M. E. Matheny. 2021. “Enhancing Trust in
AI through Industry Self-governance.” Journal of the American Medical Informatics
Association 28(7): 1582–90. doi: 10.1093/jamia/ocab065.
Rousseau, D. M., S. B. Sitkin, R. S. Burt, and C. Camerer. 1998. “Not So Different after All: A
Cross-discipline View of Trust.” Academy of Management Review 23(3): 393–404. doi:
10.5465/amr.1998.926617.
Russell, S. J., P. Norvig, E. Davis, and D. Edwards. 2016. Artificial Intelligence: A Modern
Approach, 3rd ed. London: Pearson.
A Multilevel Framework for AI Governance 15

Samek, W., G. Montavon, A. Vedaldi, L. K. Hansen, and K. R. Müller, eds. 2019. Explainable
AI: Interpreting, Explaining and Visualizing Deep Learning. Basel, Switzerland:
Springer. doi: 10.1007/978-3-030-28954-6.
Seberger, J. S. 2021. “Reconsidering the User in IoT: The Subjectivity of Things.” Personal and
Ubiquitous Computing 25(3): 525–33. doi: 10.1007/s00779-020-01513-0.
Seberger, J. S., M. Llavore, N. N. Wyant, I. Shklovski, and S. Patil. 2021. “Empowering
Resignation: There’s an App for That.” Proceedings of the 2021 CHI Conference on
Human Factors in Computing Systems, May 7, 2021. doi: 10.1145/3411764.3445293.
Shin, D. 2021. “The Effects of Explainability and Causability on Perception, Trust, and
Acceptance: Implications for Explainable AI.” International Journal of Human-
Computer Studies 146: 102551. doi: 10.1016/j.ijhcs.2020.102551.
Shin, D., and Y. J. Park. 2019. “Role of Fairness, Accountability, and Transparency in
Algorithmic Affordance.” Computers in Human Behavior 98: 277–84. doi:
10.1016/j.chb.2019.04.019.
Somers, G., and L. Boghaert. 2018. “The California Consumer Privacy Act and the GDPR: Two
of a Kind?” Financier Worldwide, November 2018,
https://www.financierworldwide.com/the-california-consumer-privacy-act-and-the-
gdpr-two-of-a-kind.
Sundar, S. S., and J. Kim. 2019. “Machine Heuristic: When We Trust Computers More Than
Humans with Our Personal Information.” CHI’19: Proceedings of the 2019 CHI
Conference on Human Factors in Computing Systems, May 2019. doi:
10.1145/3290605.3300768.
Thiebes, S., S. Lins, and A. Sunyaev. 2020. “Trustworthy Artificial Intelligence.” Electronic
Markets 31: 447–64. doi: 10.1007/s12525-020-00441-4.
Turing, A. M. 1950. “Computing Machinery and Intelligence.” Mind 59(236): 433–60. doi:
10.1093/mind/LIX.236.433.
Vallor, S. 2016. Technology and the Virtues: A Philosophical Guide to Future Worth Wanting.
Oxford University Press.
Wahl, B., A. Cossy-Gantner, S. Germann, and N. R. Schwalbe. 2018. “Artificial Intelligence (AI)
and Global Health: How Can AI Contribute to Health in Resource-Poor Settings?” BMJ
Global Health 3(4): e000798. doi: 10.1136/bmjgh-2018-000798.
Watson, D. 2019. “The Rhetoric and Reality of Anthropomorphism in Artificial Intelligence.”
Minds and Machines 29(3): 417–40. doi: 10.1007/s11023-019-09506-6.
Whittaker, M., K. Crawford, R. Dobbe, G. Fried, E. Kaziunas, V. Mathur, S. Myers West, R.
Richardson, J. Schultz and O. Schwartz. 2018. “AI Now Report 2018.” AI Now Institute,
New York University, December 2018,
https://kennisopenbaarbestuur.nl/media/257225/ai_now_2018_report.pdf.
Wu, K., Y. Zhao, Q. Zhu, X. Tan, and H. Zheng. 2011. “A Meta-analysis of the Impact of Trust
on Technology Acceptance Model: Investigation of Moderating Influence of Subject and
Context Type.” International Journal of Information Management 31(6): 572–81. doi:
10.1016/j.ijinfomgt.2011.03.004.
Zimmerman, M. R. 2018. Teaching AI: Exploring New Frontiers for Learning. Eugene, OR:
International Society for Technology in Education.