comment on paper
zachzhang
AIEthicsApprovalofGovernmentUsagesofAIinCOVID-19context-SBE-KellyAshihara.pdf
1
2
3
4
Abstract 5
6
Based on respondents in AI superpowers China, Europe/US, Korea, our AI Ethics impact study 7
focuses on factors that may impact one’s approval of government usage of AI facial scanning for 8
police and criminal investigations and for minor law enforcement including street surveillance. We 9
seek to identify the degree to which respondents accept the ‘deep state’ governance scenario or 10
infringement upon their privacy and freedom. We tested fifteen factors and categorized them into a 11
model with three constructs: factors related to society, culture, government; factors related to one’s 12
personal values and ethics; and factors related to one’s views on technology. Based upon our two-part 13
COVID-19 survey in 2020-2021 (262) and 2022 (579), we analyzed data from 561 respondents from 14
Study 2 who self-identified as from China (176), Korea (259) and Europe / U.S. (76) and other (50). 15
Utilizing logistical regression five variables had significance for both criminal investigations and 16
enforcement of public law and ten variables had significance for AI usage for only criminal 17
investigations; no variables were significant for only minor law enforcement, which we believe 18
recognizes a degree of difference between the two AI usages. Implications include our effort to 19
expand the vernacular of AI ethics to include Government as users and citizens as subjects, inclusion 20
of political economy and philosophy which are often excluded in AI research studies, connection 21
between Hofstede cultural dimensions and individual ethics, and implications on the potential 22
influence of COVID-19 policies on AI usage and governance. 23
Keywords: COVID-19, tracking, privacy, AI, ethics, trust, Hofstede 24
25
26
27
28
29
30
31
32
33
34
2
1.1 Introduction 35
Artificial Intelligence is a 58 billion Dollar worldwide investment (MIT-BCG 2020 AI Research) at 36
the vertex of leading to national competitive advantage. While major superpowers including 37
US/Europe, China and Korea compete to be at the forefront - private enterprise giants, Microsoft, 38
Facebook, Amazon, Google as well as major semi-conductor manufacturers Nvidia, Intel, AMD 39
contend to be at the cutting edge of development and adoption of AI technology. Juxtaposed with this 40
phenomenon, demand for interconnectivity has led to our almost tacit consent to the terms and 41
conditions of usage for social media, search, and cloud. In the shadows looms the question of ethics, 42
responsible usage, and impact on the way we work and our rights to privacy. We adopt a ‘necessary 43
but not sufficient’ view that AI Ethics needs to address topics of transparent technologies with 44
inclusive algorithms and mechanisms to ensure that the harmful entities cannot hack and interject 45
antisemitic content, but that such is not sufficient. Rather, that there is a need to expand the scope of 46
AI ethics vernacular to consider the context of users and values – inclusive of national culture and 47
Government policy. 48
The cataclysmic occurrence of COVID-19 and its prolonged duration led to an unprecedented and 49
somewhat simultaneous disruption in our daily lives. While many were under lockdown for prolonged 50
periods, relegated to remote work and school, and were subject to intense COVID-19 social 51
distancing restrictions, Governments and private enterprise simultaneously increased investment in 52
technology and systems to monitor COVID-19 transmission, including tracking cases, tracing 53
contacts, border containment and quarantine (Kim, Ashihara, 2020). According to a recent MIT-BCG 54
2020 Artificial Intelligence Global Executive Study and Research Project, 55% of companies 55
accelerated their AI strategy and adoption because of COVID-19. At the national superpower level, 56
the pandemic seemingly empowered some national superpowers to seek unlimited term limits and 57
arguably even in democratic states political abuse of AI is connected to fake news, micro-targeting 58
3
propaganda and election fraud (Daly, 2019). The who in government and their own moral compass 59
has a daily impact on at least 1.87 billion citizens in these three AI superpowers alone. 60
Related to the concepts of democratic control, governance and political deliberations we evaluated 61
views of respondents related to two types of Government usage of AI, firstly for major criminal 62
investigations and by the police - inferring namely whether one is willing to approve of Government 63
usage of AI for preservation of justice and in doing so, willing to succeed rights of data privacy to the 64
Government. Secondly, Government usage of AI for street surveillance and transportation. In this we 65
ask respondents subjects whether they approve of AI for minor purposes and trust the government 66
and whether they are willing to forfeit their rights to this data. As a dependent variable, AI for minor 67
purposes is essentially asking if one is ok with the ‘deep state’ or ‘big brother is watching’ scenario. 68
We purport that these realities prompt the understated and too often seldomly asked question of how 69
COVID-19 governance and granting rights to the government impacts one’s views on their rights to 70
privacy and views towards Government approval of AI? This paper posits that in the wake of return 71
to normalcy, as a collective it is important to address the larger implications and links to political 72
economic philosophy. This research seeks to address the fundamental question of governance and the 73
social impact of COVID-19 policies. More specifically, we delve into aspects and ask how 74
government, culture, ethics and acceptance of technology impact approval of AI usage. In this pursuit, 75
we undertook a study to evaluate AI Ethics and factors that contribute to the approval of government 76
usage of AI . Our focus on AI Ethics is contributory in that we conduct an empirical study that is 77
intertwined with national superpowers investing in AI. On philosophical level, these countries and 78
their investment in and fundamental ties to ethics is simultaneously interconnected with political 79
economy with roots in philosophical ideology, culture and individual ethical values. 80
This paper is organized as follows. Section 1 looks at the construct model and links factors with 81
political philosophy and national culture. In this we seek to look at psychology and how political 82
4
economy impacts respondent views on AI and elaborate on the selection of questions and independent 83
variables. We also draw upon cultural factors integrated into our study including Hofstede’s cultural 84
dimensions as such highlights differences in national culture and values. Section 2 describes our study 85
participants and three-part construct model and overview of the questions integrated. Section 3 covers 86
methodology for data collection and research methodology. Section 4, includes data analysis and 87
outcomes with box plots, and logistic regression results. Lastly, section 5 concludes with broader 88
significance of the findings and limitations as well as implications for future research. 89
1.2 Literary Review 90
Ethics in artificial intelligence (AI) is an emerging field within applied ethics and the philosophy of 91
technology. AI or the theory that computer systems are able to perform tasks requiring human 92
intelligence is widely heralded as a “revolution” that has the potential to transform society, work and 93
science (Jobin., et.al., 2019, Harari, 2017, Appenzeller, 2017). Given the transformative force and 94
profound implications across societal domains, AI has sparked debate on the principles and values 95
that should guide its development and use (Müller, 2020, Ryan & Stahl, 2020). Common AI ethical 96
principles are for example transparency, justice and fairness, non-maleficence, responsibility, and 97
privacy (Jobin et al., 2019). These principles function as non-legislative policy instruments which are 98
non-binding. This prompts the question as to what extent ethical principles are actually implemented 99
and embedded in the development and application of AI, or whether merely good intentions are 100
deployed (Hagendorff, 2020). 101
AI ethics is an ongoing dialogue between the related roles of humans and machines (Friend, 2018, 102
Bakineer, 2022). The darkest speculation is some sort of Artificial General Intelligence that no longer 103
needs human oversight, another more collaborative vision is some sort of mutual coexistence in which 104
increasingly capable machines interact with humans in sophisticated ways (Stone., et al., 2016; 105
Bakineer, 2022). As leading technology companies increasingly take the lead in AI R&D, one 106
5
quintessential question is who is responsible for development and enforcement of ethical guidelines? 107
This same question applies to social media, data privacy and cloud storage. Is it ultimately up to the 108
developers of technology to self-police? Are private companies expected to self-govern – or are 109
national governments expected to pass legislation that impacts R&D for the largest most innovative 110
companies? And with China leading the way in development, how do cultural norms or government 111
surveillance impact views of the citizens under their dominion? 112
In line with principles of justice, fairness, responsibility and privacy, rather than evaluating the ethics 113
related to programming mechanisms (referred to as “machine ethics”), bias and AI agency, we look 114
at the subjects and objects of the equation. Within a larger framework where Government leaders and 115
industry developers or social media companies are actors and citizens as followers, we maintain that 116
in line with the enormous benefits, artificial intelligence and related technologies also pose enormous 117
risks to humanity (Bostrom, 2014). In other words, responsible innovation in AI calls for public 118
deliberation and a well-informed democratic debate with actors from the public, private, and civil 119
society to critically address the goals and facilitate well-informed judgments on opaque AI systems 120
to effectuate democratic governance (Buhmann and Fieseler, 2022). 121
Acknowledging this conundrum, Brad Smith President of Microsoft predicts that the kind of 122
controlled, mass surveillance society portrayed by George Orwell’s 1984 (Orwell, 1949) dystopian 123
novel 'could come to pass in 2024' if more isn't done to curb the spread of AI (Hsu, 2019). Related to 124
the intent to prevent abuse of mass surveillance, in his book Tools and Weapons, The Promise and the 125
Peril of the Digital Age (Smith et.al, 2019) Smith addressed issues of governmental abuse of mass 126
surveillance and proposed laws that only allow law enforcement to use facial recognition for ongoing 127
surveillance of specific individuals when police get a court order or during emergencies – efforts that 128
in part contributed to the a passing of a bill by Washington state senators. 129
6
In defining our scope, we undertook a semi-systematic overview of issues and normative stances in 130
the field, demonstrating how the details of AI ethics coalesce into a larger dialogue. We reviewed 131
comprehensive AI guidelines (The IEEE Global Initiative on Ethics of Autonomous and Intelligent 132
Systems 2019) and to understand existing government guidelines we reviewed the European 133
Commission’s “Ethics Guidelines for Trustworthy AI” (Pekka et.al, 2018); and the Obama 134
Administration’s “Report on the Future of Artificial Intelligence” (Holdren et al. 2016). We also took 135
special note that that the Beijing AI Principles (Beijing Academy of Artificial Intelligence 2019) both 136
embraces human values and is also removed from many official websites; a fact that coincided with 137
the removal of our survey on many websites and challenges due to the banning of google surveys in 138
China. Lastly, to understand the supranational perspective, we also reviewed the “OECD Principles 139
on AI” (Organisation for Economic Co-operation and Development 2019). 140
We took special note of what Hagendorff (2019) identified as “omissions” or underrepresentation of 141
topics of political abuse or social and ecological costs of AI systems and seek to broaden the 142
guidelines as many contributors in the AI field are from computer science rather than social sciences 143
and philosophy backgrounds. In fact, we could only find two documents – the Montreal Declaration 144
for Responsible Development of Artificial Intelligence (2018) and the AI Now 2019 Report that 145
explicitly address aspects of democratic control, governance and political deliberations of AI systems 146
(Hagendorff, 2019). AI Ethics refers to the entire field of discourses and practices emphasizing the 147
potentially negative impact of AI driven technologies which includes shared concepts and meanings, 148
philosophical underpinnings and citations (Bakineer, 2022). 149
During the pandemic – the disparity in COVID-19 policy and implementation highlighted the vast 150
differences in political ideology. Government’s role in the economy is debated with views ranging 151
from anarchists, who believe in no government to that of totalitarian socialists who believe in 152
complete control of the economy and society (Lipford and Slice, 2007). On one hand, China’s zero 153
7
covid and draconian containment policies highlighted the sheer power of Xi Jingping and extent of 154
the reign of his influence. In between, Korea adopted swift measures and quickly passed laws that 155
enabled and empowered near complete control with laws that mandated masks, mandatory testing, 156
GPS tracking and extreme social distancing measures (Kim, Ashihara 2020). Meanwhile, Western 157
democracies and leadership had much laxer and more inconsistent COVID-19 policies with stimulus 158
checks from the Government signed in the former U.S. President’s name and uprisings for Black 159
Lives Matter in the U.S.; meanwhile in France, citizens also protested related to the health vaccine 160
pass, which both exemplify direct impact of the values of individualism. 161
1.3 Study Background and Methodology 162
Our self-funded exploratory study seeks to identify the factors that may lead one to approve of AI 163
and the associated parameters of usage. This question comes in the wake of COVID-19, when in the 164
face of an urgent need to protect and ensure public health, government has had unprecedented power 165
over citizenry. While some parties promote the benefit of usage in criminal investigations to public 166
safety there is also a view that the intrusiveness may outweigh the benefits (Muthensentil and Kim, 167
2018). Such also raises the question of how privacy and usage of technologies such as AI infringes 168
upon ethical rights to data? We ask respondents to indicate their views on the Government’s right to 169
track, quarantine, lockdown and seek to understand how such correlates with their views towards 170
government approval of AI for criminal purposes by police and minor law enforcement in public 171
spaces for transportation and street surveillance. Our study is a limited attempt to measure how one’s 172
culture, covid-19 governance, personal ethics and values and views on technology influence tendency 173
to approve of AI. Through this we seek to expand the dialogue to assess how the usage impacts one’s 174
approval and factors that impact young citizens – and try to consider the context of culture and 175
personal ethics. 176
177
8
178
179
180
181
182
- 184
INDEPENDENT VARIABLE 1 GOVERNMENT POLICY AND CULTURE CONSTRUCT
[INDV-COL] In the country that I live in people are generally responsible for their own success and individualistic or concerned about others (collective).
[POW.D] In your society, are citizens expected to question or obey their leaders when in disagreement?
[HIER] 2-9.1 What is your opinion of your home country government and the individual? [My country is hierarchical and autocratic]
[QTNE] 4-4: (2022) Government Quarantine: Related to COVID - the government should quarantine those exposed to covid
[Freedom] Indicate your views on how the Government of the country you were living in in 2021 handled the crisis. [Individual freedom can be limited for the sake of the community]
[GOVFREE] What is your opinion of your home country government and the individual? [My government guarantees individual freedom]
INDEPENDENT VARIABLE INDIVIDUAL ETHICS AND VALUES CONSTRUCT
[ETHICS] If you find out in your company or organization that
there is an ethical problem (corruption, cheating, violation of
human rights) - what would you do?
[TRK-MIN] Do you believe your government should be able to
track individuals for the following purposes? Ensure minor
laws are followed (theft, vandalism, traffic)
[TRKCRIM] Do you believe your government should be able to
track individuals for the following purposes? [Assist with
criminal investigations (murder, sexual abuse, terrorism)
[TRST_PVT] Do you trust the government with responsible
use of your personal data? Level of trust in Private Companies
(cloud providers, social media platforms.
[TRSTGOV] Do you trust the government with responsible use
of your personal data? [Level of Trust in Gov’t with personal
data]
INDEPENDENT VARIABLE 3 TECHNOLOGY ADOPTION AND PERCEPTION CONSTRUCT
[TECHADPT] How do you classify yourself in terms of the technology acceptance? (Innovator, Early Adopter, Early
Majority, Later Majority, Laggard)
[QCOMP] If you know you were GPS tracked but could leave your device at home and believed you wouldn’t get
DEPENDENT VARIABLE GOVERNMENT USAGE OF AI
On one hand, AI can be used to help the blind
see, deaf hear and help us translate languages.
However, AI is also known to reinforce racial
bias and used for password phishing and
internet fraud. It is being used for spying, track
citizens, in schools and to process financial
transactions. Indicate your approval for AI
usage
[AI-CRIM] Indicate your approval for AI usage
below: Facial scanning and criminal
investigations by police
[AI-LAW] Indicate your approval for AI usage
below: Enforcement of law in public spaces
(transportation, streets)
7. Likert: 5 point scale: 1 - strongly disapprove, 2 somewhat disapprove, 3 – neutral, 4 – slightly approve, 5 – strongly approve
Table 2: Construct Model: Government Usage of AI
9
Freedom: Indicate your views on how Government in the country you were living in in 2021 185
handled the crisis [Individual freedom can be limited for the sake of community] 186
- GOVFREE: What is our opinion of your home country government and the individual? [My 187
government guarantees individual freedom] 188
- QTNE: Related to COVID - indicate to the extent that you agree. The government should 189
[quarantine those ‘exposed’ to COVID?] 190
- INDV_COL In the country that I live in people are generally responsible for their own 191
success and individualistic or concerned about others (collective). 192
- POW.D: In your society are citizens expected to question or obey their leaders when in 193
disagreement? 194
- HIER: What is your opinion of your home government and the individual? My country is 195
hierarchical and autocratic. 196
197
Three variables in our cultural construct incorporate Hofstede dimensions of Hierarchy – and the 198
degree of preference for authority and Individualism vs. Collectivism, Power Distance – the extent to 199
which the less powerful members of organization and institutions accept and expect that power is 200
distributed unequally. According to Hofstede, “Culture is a collective programing of the mind” and 201
maintains societal, national and gender cultures are deeply rooted and reside in values and broad 202
tendencies to prefer certain states of affairs over others (Hofstede, 2011, Hofstede 2001). In large 203
power distance countries, parents teach children obedience, older people are feared and education is 204
teacher-centered. Corruption is frequent, income distribution uneven – in the corporate world, 205
subordinates expected to obey and government is often autocratic (Hofstede, 2011). This is in contrary 206
to more democratic socialist ideas where power should be legitimate, parents treat children as equals, 207
education is student-centered, and subordinates expect to be consulted (ibid). 208
209
10
1.3.1 Construct 2: Individual Ethics and Values 210
Variables in construct 2 specifically relate to individual values and focus on the individual response 211
to situations. These variables incorporate government policy questions, however, instead of focusing 212
on the country and society, ask the respondent to identify their beliefs and ethical views. Regardless 213
of nationality, this construct takes into account one’s ethics with values with questions on one’s 214
actions “If you find out in your company or organization that there is an ethical problem (corruption, 215
cheating, violation of human rights) - what would you do?” Additional questions in this construct 216
focus on one’s beliefs related to the government’s ability to track individuals for major crimes and 217
minor laws are followed (theft, vandalism); and trust in government and private companies related to 218
data management. 219
- ETHICS: If you find out in your company or organization that there is an ethical problem 220
(corruption, cheating, violation of human rights) - what would you do? 221
- TRK_MIN: Do you believe your government should be able to track individuals for the 222
following purposes? [Ensure minor laws are followed (theft, vandalism, traffic)] 223
- TRKCRIM: Do you believe your government should be able to track individuals for the 224
following purposes? [Assist with criminal investigations (murder, sexual abuse, terrorism)] 225
- TRST PVT: Do you trust private companies (cloud providers, social media platforms) with 226
responsible use of your personal data? 227
- TRST GOV: Do you trust the government with responsible use of your personal data? 228
229
1.4 Construct 3: Technology Adoption 230
Construct 3 technology adoption looks at the adoption and perception on technology such as GPS 231
tracking and participant willingness to comply to government mandated quarantine using GPS 232
tracking. While this does not specify nationality, we hypothesize that respondents will differ according 233
to the government policy and their perception of government power to enforce policy and that 234
11
quarantine compliance. We believe that one’s relative views on technology adoption may influence 235
their receptiveness to AI in general and hence, also integrated three questions related to one’s 236
classification according to technology acceptance, one’s willingness to download apps that use GPS 237
tracking; and how their use of social media changed during the pandemic. 238
- TECHADPT: How do you classify yourself in terms of the technology 239
acceptance? (Innovator, Early Adopter, Early Majority, Later Majority, Laggard) 240
- QCOMP: If you know you were GPS tracked but could leave your device at home and 241
believed you wouldn’t get caught, how likely are you to comply? 242
- COV+TRK: Willingness to download a mobile app indicating if I was exposed to COVID+ 243
- SMEDIA: Indicate how your TIME spent changed during the pandemic? 244
1.5 Study Background and Participants 245
Our survey started in 2020-2021 when ten international exchange students were quarantined in their 246
University dorm after visiting Itaewon, a local night club region of the second super-spreader incident 247
in South Korea. In the Spring of 2020, we began a dialogue to understand COVID-19 policies; this 248
evolved to incorporate government policy and individual values related to the pandemic in late 2021, 249
prior to the Korean Government’s with COVID policy. In the Spring of 2022, the first author worked 250
with 13 teams across three sections of a business communications course focused on emotional 251
intelligence, ethics and negotiations. Under the umbrella of COVID-19 and social impact, team topics 252
spanned the spectrum of education, gender, privacy human rights and AI. Our survey included fifty-253
five questions organized in four sections with a mix of Likert scale, multiple choice and qualitative 254
questions. Individual questions that were revised and edited (see Table 1: Study Participants for 255
COVID-19 Impact Survey). 256
12
Table 1: Study Participants for COVID-19 Impact Survey 257
Data N
recruited
N
retained*
Gender Nationality Median
Age Male Female China EU/U
S Korea Other
Study
1* 262 262 - - 37 45 163 17 26
Study 2 579 561 229 328 176 76 259 50 26
*Study 1 data did not ask gender 258
259
As indicated in Table 1, our COVID-19 Impact Survey is separated into two studies. Study 1 was 260
conducted at the onset of COVID-19 in 2020 and continued for a second round of responses in late 261
2021 after vaccine rollout and restrictions were lifted. Study 2 was conducted in the late spring and 262
summer of 2022. Both surveys were part of three semester projects. 263
For Study 2, we had three separate survey questionnaires, two in Korean and English, and one in 264
English and Chinese. All three questionnaires were comprised of four sections. The first section about 265
demographics and views on the COVID pandemic (16 questions), second section about views on 266
government, values, and impact on education and society (14 questions). Third section on COVID 267
impact on human rights (22 questions), fourth section on data privacy, AI and the role of government 268
and private companies (14 questions). While we included 55 total questions including sub-questions, 269
our survey had 168 questions. Amongst these, we identified 15 questions that aligned with our study 270
of culture, government policy, ethics, views on data privacy and AI. 271
Our class teams recruited respondents for our main survey in both English and Korean with 267 272
responses. In the summer of 2022, we separately translated and recruited 117 mainland Chinese 273
respondents on social media and WeChat. After encountering barriers due to unavailability of Google 274
(google survey) and removal of our survey by Chinese bots that identified sensitive keywords. 275
Chinese respondents were compensated with milk tea or Americano coupons approximately $3.50 276
13
each. Based upon the original survey, one team independently recruited an additional 128 respondents 277
from Korea, China, and Japan. We included this data for the six independent variables for both AI 278
dependent variables. 279
The final sample included 561 participants including Koreans (259, 46.16%), Chinese (176, 31.37%), 280
76(17.55%); and European/U.S. (76, 13.54%) and Asia, Middle East, Africa Other (50, 8.91%). We 281
excluded 19 participants due to duplicate responses or incomplete surveys. Respondents were from 282
12 nationalities. Gender demographics include 58% female, 41% male, with 1% blank or non-283
determinate. The mode age of participants was 20-25, with median calculated for both at 26 years. 284
Informed consent was obtained from all participants. 285
To test our hypotheses, we developed a construct model as seen in Table 3: Construct Model: 286
Government Usage of AI with 15 factors that we believe may impact one’s approval of AI facial 287
scanning for police and criminal investigations and government usage of AI for more minor law 288
enforcement that includes transportation and street surveillance. In other words, we seek to ask what 289
factors affect one’s parameters for AI approval and posit that society, culture government are one 290
factor, one’s personal values and ethics are a second and views on technology adoption and 291
acceptance are a third. 292
Most questions were on a Likert 1-5 scale and directionally adjusted according to our hypotheses that 293
approval of AI would be higher for those who indicated they were more collective and more accepting 294
of government policies for quarantine (QTNE) and more receptive to technology. Questions scaled 295
on a Likert 1-7 scale were converted to a 1-5 scale with a few also containing yes, neutral and no 296
options. We also included a few qualitative short answer questions to understand their specific views 297
on COVID-19 impact and to gather specific insights. 298
299
14
1.6 Data Analysis Methodology 300
To assess correlations between construct variables and respondent perception on the use of AI for 301
facial scanning and criminal investigations by police (AI_CRIM); and AI facial scanning for 302
enforcement of law in public spaces (transportation and street surveillance) AI_LAW, we first created 303
box plots with means and standard deviation and distribution of responses. Our objective in this first 304
step was to revise our criteria and to eliminate potentially non-related variables and prioritize our 305
regressions as we utilized an independent Ordinal Logistic Regression model. After identifying more 306
significance than expected, we decided to run Logistic Regression for all 30 variables. 307
We ran boxplots for each independent and dependent variable (Figure 3,Table 3) (Figure 4, Table 4). 308
We expected to identify and significantly reduce our independent variables. In this step, we 309
established four criteria: 1) Comparing those who chose “1” (1,2) for the independent variable with 310
“5” (4,5), the difference in mean of the same values (1 and 5) for the dependent variable (AI_CRIM 311
or AI_LAW) should be greater than 0.5. 2) means of dependent variable should be successive 312
(increasing or decreasing’), which indicates consistency in responses and potential correlation; 3) In 313
the right graph, when the value of the independent variable increases from 1 to 5, the portion of the 314
dependent variable “1” (1, 2) should decrease while the portion of the dependent variable “5” (4, 5) 315
should increase. 316
15
Figure 2: Box Plot Variable Analysis - ETHICS for AI-CRIM 317
318
Box Plot Table 3 - Ethics 319
Ethics Independent Variable
Q: If you find out in your company or organization that there is an ethical problem (corruption, cheating, violation of human rights) - what would you do?
Evaluation
1. Comparing respondents who chose “1” (1,2) for ETHICS and “5” (4,5) the difference in mean of the dependent variable AI_CRIM for the same values is greater than 0.5. Since difference is .747>0.5.
True
2. Means of dependent variable should be successive (increasing or decreasing’), which indicates consistency in responses and potential correlation. AI.CRIM “1” < AI.CRIM “3”< AI.CRIM “5” hence, directional similar and likely correlative.
True
3. As the value of ETHICS increases, the portion of values of AI-CRIM “1” decreases while the portion of values of AI-CRIM “5” increases.
True
Decision: Inclusion in first run for regression and keep variable in Construct 2.
16
Figure 3: Box Plot Variable Analysis – INDV.COL and AI-CRIM 320
321
322
Box Plot Table 4 – INDV.COL 323
INDV.COL Variable Evaluation Criteria
Q: In the country that I live in people are generally responsible for their own success and individualistic or concerned about others (collective).
Evaluation
1. Comparing those who chose “1” (1,2) for INDV.COL and “5” (4,5) the difference in mean of the dependent variable AI_LAW for the same values is greater than 0.5. Since difference is .31< 0.5.
False
2. Means of dependent variable should be successive (increasing or decreasing’), which indicates consistency in responses and potential correlation. AI.LAW“1” < AI.LAW “5” <AI.LAW “3” hence, directional unsimilar and likely not correlative.
False
3. As the value of INDV.COL increases, the portion of values of AI-CRIM “1” does not decrease. Also, the portion of values of AI-CRIM “5” does not increase.
False
Decision: Eliminated variable from Construct 1.
324
4. Results and Discussion 325
By using boxplots analysis, we were able to assess the likely correlation between independent and 326
dependent variables. Through our three part criteria we identified and realized likely differences in 327
the variables that correlated with each of our AI variables AI.CRIM and AI.LAW, an aspect we had 328
not fully anticipated. 329
MedianAFt er
Maximum
Minimum
Interquartile range (IQR)
Q3 (75 th
percentile)
Q1 (25 th
percentile)
Mean
Outliers
Outliers
Maximum
17
For AI.CRIM, use of AI for facial scanning and criminal investigations by police variables: 330
INDV_COL, POW.D, HIER, QCOMP had two or three false and thus were prioritized for the 2nd 331
run of regression. 332
For AI.LAW, Enforcement of law in public spaces (transportation, streets), variables INDV-COL, 333
POW.D, Tech_ADP, SMEDIA had had two or three false and thus were prioritized for the 2nd run of 334
regression. 335
One key finding is that the variables that were likely significant differed for AI.CRIM and AI.LAW 336
indicating that our respondents distinguished between the factors that they deemed relevant to 337
evaluate approval of AI for different government purposes. We anticipated being able to significantly 338
reduce the number of variables and used this step to prioritize our regressions. After viewing the 339
results of our first regression variables, we decided to run all additional variables. This step also made 340
the need to reverse GOVFREE [My government guarantees individual freedom] apparent. Our 341
methodology to confirm statistical dependence of each of the independent variables is described 342
below. 343
1.7 Ordinal Logistic Regression 344
To confirm our hypotheses, we ran logistical regression for each of 15 independent variables against 345
our 2 dependent variables (total of 30). Regressions of each independent variable and each dependent 346
variable were run separately to avoid the effect of intercorrelation. 347
The scheme is useful when the objective is to identify contrasts in the dependent variable between 348
successive levels of the independent variable, or to identify critical threshold values of the 349
independent variables at which significant changes occur in the response (Walter et.al., 1987). 350
We grouped responses 1 and 2, responses that indicate a selection of values aligned with 351
individualism, freedom and indicating they were in quarantine were less likely to comply if they knew 352
18
they “were GPS tracked but could leave your device at home and believed you wouldn’t get caught.” 353
Similarly, we grouped responses for 4 and 5, which are responses that indicated more collective 354
values, more passive ethical values and a higher likelihood of compliance to quarantine. 355
1.8 Ordinal Logistical Regression Formula: 356
Let 𝒀 be an ordinal categorical dependent variable with 𝑱 categories. Then 𝑷(𝒀 ≤ 𝒋) represents the 357
cumulative probability of 𝒀 less than or equal to a specific category 𝒋 (𝑗 ∈ {1, … , 𝐽 − 1}). 358
The odds of being less than or equal to category 𝒋 can be defined as: 359
𝑃(𝑌 ≤ 𝑗)
𝑃(𝑌 > 𝑗) 360
The log odds is also known as the logit, so that: 361
𝑙𝑜𝑔 𝑃(𝑌 ≤ 𝑗)
𝑃(𝑌 > 𝑗) = 𝑙𝑜𝑔𝑖𝑡 (𝑃(𝑌 ≤ 𝑗)) = 𝛼𝑗 − 𝛽1𝑥1 − ⋯ − 𝛽𝑝𝑥𝑝 362
Note that 𝑙𝑜𝑔𝑖𝑡 (𝑃(𝑌 ≤ 𝑗)) = 𝛼𝑗 − 𝛽1𝑥1 − ⋯ − 𝛽𝑝𝑥𝑝 is the formula of Ordinal Logistic 363
Regression used in the polr function of package MASS in R software, which is the function used in 364
all the regressions of this study. 365
Based on this formula, if the coefficient of the independent variable is positive, the odds of being less 366
than or equal to category j will be decreased. In short, when the independent variable has a positive 367
coefficient, an increase in the independent variable tends to lead to an increase in the dependent 368
variable. 369
Encoding Independent Variables: 370
To simplify the encoding of variables, all variables in the analysis were converted to 3-scales. In 371
which 1 - Disagree (includes Strongly Disagree and Somewhat Disagree), 3 - Neutral, 5 - Agree 372
(includes Somewhat Agree and Strongly Agree). 373
19
The independent variables in the model are all ordinal categorical variables. To observe the difference 374
between the 3 groups of Disagree, Neutral and Agree, the independent variables are encoded as 375
follows: 376
Table 5: Ordinal Categorical Variables 377
INDEPENDENT VARIABLE X X.D1 X.D3
1, 2 – STRONGLY DISAGREE, DISAGREE 0 0
3 - NEUTRAL 1 0
4, 5 – STRONGLY AGREE, AGREE 1 1
378
To explain the encoding method, the variable coefficients can be interpreted as follows: 379
When a model uses D1 and D3 concurrently, then the coefficient of D1 represents the dependent 380
variable difference between a respondent who chooses X = 1 or 2 and a respondent who chooses X = 381
3. Simplified, the coefficient of D1 represents the difference in Y when either strongly 382
disagree/disagree or neutral is selected. Similarly, the coefficient of D3 represents the difference in Y 383
when either neutral or agree / strongly agree is selected for X. 384
When a model uses only the variable D1, the coefficient of D1 represents the difference in Y between 385
a respondent who chooses X = 1,2 strongly disagree/disagree and respondent who chooses either X 386
= 3 neutral or 4,5 agree/strongly agree. In this case, the coefficient of D1 represents the change in Y 387
when X changes between [strongly disagree/disagree] and [neutral or agree/strongly agree]. 388
When a model uses only the variable D3, the coefficient of D3 represents the difference in Y between 389
a respondent who chooses X = 4,5 agree/strongly agree and respondent who chooses either X = 3 390
neutral, or 1,2 disagree/strongly disagree. In this case, the coefficient of D3 represents the change in 391
Y when X changes between [agree/strongly agree] and [neutral or disagree/strongly disagree]. 392
20
When either D1 or D3 or both are statistically significant, we can conclude that the independent 393
variable influences the dependent variable. In the instances when the coefficients of D1 and D3 are 394
opposite, we looked at the coefficient and concluded that if the sum of the coefficients is positive then 395
the direction for the independent and dependent variables is similar and thus, that the independent 396
variable still has a positive effect on dependent variable. 397
Given our decision to run separate regressions to prevent intercorrelation between variables, for the 398
six independent variables: “INDV.COL, POW.D, GOVFREE, TRSTGOV, COV.TRK, SMEDIA” we 399
included 128 responses collected separately in 12 ordinal logistic regressions of six independent 400
variables vs AI.CRIM and AI.LAW (for details, see study background and participants section). 401
402
Results - AI Regression 403
The results of our Logistical Regression model are indicated below and categorized by construct. 404
Accordingly, our findings are as follows: 405
CONSTRUCT 1 GOVERNMENT POLICY AND CULTURE 406
1. INDV_COL: we were not able to find significance for either AI-CRIM and AI-LAW. 407
2. POW.D: significance for AI-CRIM, meaning that respondents who think there is a greater 408
distance between government and citizens or that citizens are expected to obey their leaders 409
when in disagreement are more likely to approve government usage of AI in enforcement of 410
law in public spaces (transportation, streets) [D3: Coefficient 0.554, T-stat 2.473, p < 0.05]. 411
3. HIER: we were not able to find significance for either AI-CRIM and AI-LAW. 412
4. QTNE, significance for both AI-CRIM and AI-LAW, meaning that people who are more 413
compliant for government regulations are more likely to approve government usage of AI in 414
21
both criminal investigations [D3: Coefficient 0.792, T-stat 3.967, p < 0.001] and enforcement 415
of law in public spaces (transportation, streets) [D3: Coefficient 0.444, T-stat 2.273, p < 0.05]. 416
5. FREEDOM: significance for both AI-CRIM and AI-LAW, meaning that people who think 417
individual freedom can be limited for the sake of the community are more likely to approve 418
government usage of AI in both criminal investigations [D1: Coefficient 0.833, T-stat 2.778, p 419
< 0.01] and enforcement of law in public spaces (transportation, streets) [D1: Coefficient 1.054, 420
T-stat 3.430, p < 0.001]. 421
6. GOVFREE, significance for AI-CRIM, meaning that people who think that their government 422
guarantees individual freedom are more likely to approve government usage of AI in criminal 423
investigations [D3: Coefficient 0.402, T-stat 2.327, p < 0.05]. 424
Based on this analysis, we were able to conclude that the approval of government usage of AI is 425
correlated with the views on government policy and culture. 426
CONSTRUCT 2: INDIVIDUAL ETHICS AND VALUE CONSTRUCT 427
1. ETHICS: significance for AI-CRIM, meaning that people who are more proactive and likely 428
to report unethical conduct are more likely to approve government usage of AI in criminal 429
investigations [D1: Coefficient 0.737, T-stat 2.919, p < 0.01]. 430
2. TRK_MIN: significance for both AI-CRIM and AI-LAW, meaning that people who believe 431
the government should be able to track individuals for ensuring minor laws are more likely to 432
approve government usage of AI in both criminal investigations [D3: Coefficient 0.827, T-stat 433
3.625, p < 0.001] and minor offenses [D3: Coefficient 0.888, T-stat 3.935, p < 0.001]. 434
3. TRK_CRIM: significance for both AI-CRIM and AI-LAW, meaning that people who believe 435
the government should be able to track individuals for assisting criminal investigations are 436
more likely to approve government usage of AI in both criminal investigations [D3: 437
22
Coefficient 1.261, T-stat 5.330, p < 0.001] and minor offenses [D1: Coefficient 0.724, T-stat 438
2.091, p < 0.05 / D3: Coefficient 0.750, T-stat 3.205, p < 0.01]. 439
4. TRST_PVT, we were not able to find significance for either AI-CRIM and AI-LAW. 440
5. TRST_GOV: significance both AI-CRIM and AI-LAW, meaning that people who trust the 441
government with responsible use of personal data are more likely to approve of government 442
usage of AI in both criminal investigations [D3: Coefficient 0.549, T-stat 2.679, p < 0.01] and 443
enforcement of law in public spaces [D1: Coefficient 0.500, T-stat 2.6922, p < 0.01. 444
Based on logistic regression results, we were able to conclude that with the exception of TRST_PVT, 445
which was not significant for either of our dependent variables related to the approval of government 446
usage of AI and Ethics, which was only correlated with AI_CRIM – the three other variables, namely 447
TRK_MIN, TRK_CRIM, TRST_GOV were correlated with both dependent variables. variables 448
related to individual ethics and values. 449
CONSTRUCT 3 TECHNOLOGY ADOPTION AND PERCEPTION CONSTRUCT 450
1. TECH_ADP was not significant for either AI-CRIM or AI.LAW. 451
2. QCOMP, we found significance for AI-CRIM for D1 and D3. The coefficient of D1 is negative 452
meaning that people who do not disagree but neutral to government quarantine mandates are 453
less likely to approve government usage of AI in criminal investigations. In contrast, the 454
coefficient of D3 is positive meaning that people who not neutral but agree to comply to 455
government quarantine mandates are more likely to approve government usage of AI in 456
criminal investigations. [D1: Coefficient -0.910, T-stat -2.489, p < 0.05 / D3: Coefficient 0.811, 457
T-stat 3.594, p < 0.001]. 458
3. COV+TRK was not significant for either AI-CRIM or AI.LAW. 459
23
4. SMEDIA, significance in AI-CRIM, meaning that people whose self-selected time on social 460
media during the COVID pandemic increased were more likely to approve government usage 461
of AI in criminal investigations [D1: Coefficient 0.697, T-stat 2.631, p < 0.01]. 462
Based on this analysis, we were able to conclude that the approval of government usage of AI is 463
correlated with quarantine compliance and social media and does not have a strong correlation with 464
technology adoption or one’s willingness to download a mobile app indicating whether they were 465
exposed to COVID+. 466
One interesting and unexpected conclusion is that while we identified five variables: POW.D, 467
SMEDIA, QCOMP, GOV.FREE and ETHICS that were all significant for AI_CRIM only, and five 468
variables: TRK_CRIM, QTNE, FREEDOM, TRK_MIN, TRST_GOV that were significant for both 469
dependent variables, AI.CRIM and AI.LAW, there were no variables that were significant for 470
AI.LAW only. 471
472
24
Table 6: Logistical Regression Results 473
APPROVAL OF GOVERNMENT USAGE OF AI
CRIMINAL INVESTIGATIONS BY POLICE ENFORCEMENT OF LAW IN PUBLIC SPACES (Transport and Street surveillance)
COEFFICIENT T-STAT P-VALUE COEFFICIENT T-STAT P-VALUE
I. GOV’T AND CULTURE
INDV_COL D1 -.054
-.168
-.246
-.778
.8057
.4363
-.314
-.057
-1.457
-.269
.1451
.7880 D3
POW.D D1 -.285
.554
-1.397
2.473
.1623
.0134*
-.179
.184
-.890
.844
.3730
.3987 D3
HIER D1 -.157
.176
-.673
.840
.5012
.4007
.221
.080
.967
.384
.3337
.7009 D3
QTNE D1 .393
.792
1.376
3.967
.1687
.0007***
.149
.444
.544
2.273
.5863
.0230* D3
FREEDOM D1 .833
.012
2.778
.057
.0055**
.9547
1.054
.112
3.430
.546
.0006***
.5852 D3
GOVFREE D1 -.245
.402
-.907
2.327
.3645
.0199*
.396
.226
1.480
1.343
.1388
.1789 D3
II. ETHICS AND VALUES
Ethics D1 .737
.226
2.919
.879
.0035**
.3794
.067
.213
.276
.860
.7824
.390 D3
TRK_MIN D1 -.355
.827
-1.523
3.625
.1278
.0002***
.112
.888
.476
3.935
.0633
.00008*** D3
TRK_CRIM D1 .207
1.261
.606
5.330
.5443
.0000***
.724
.750
2.091
3.205
.0365*
.0013** D3
TRST_PVT D1 .017
.497
.087
1.691
.9310
.0908
.294
.028
1.516
.099
.1295
.9209 D3
TRST_GOV D1 .012
.549
.064
2.679
.9488
.0073**
.500
.065
2.692
0.333
.0070**
.7390 D3
III. TECHNOLOGY ADOPTION
QCOMP D1 -.910
.811
-2.489
3.594
.013*
.0003***
.187
.257
.530
1.137
.5963
.2555 D3
TECH_ADP D1 .081
.186
.402
.734
.6878
.4628 -.258
.387
-1.278
1.566
.2010
.1173 D3
COV+TRK D1 .323
-.022
.643
-.044
.5197
.9645
.160
.187
.339
.389
.7343
.6973 D3
SMEDIA D1 .697
.349
2.631
1.916
.0085**
.0553
.287
-.065
1.072
-.368
.2835
.7127 D3
p<0.05 = *, p<0.01 = **, p<0.001 = *** 474
Variables underlined include 128 additional responses. 475
25
As seen in our Venn diagram –there was not a single variable for a single response (zero / 30) that 476
was significant for our AI.Law variable only. This, is an unexpected result and implies that our 477
respondents did recognize a difference related to the purpose of AI and that there were specific 478
variables that led to approval or disapproval of AI for law enforcement only. We do think if we looked 479
at nationality we could derive an explanation as Chinese may have different views on surveillance 480
than Europeans / Americans and we believe, Asians in general may be less sensitive to invasion of 481
their privacy and infringement upon inalienable rights such as freedom. 482
Figure 4: Government Approval of AI: Regression Results Venn Diagram 483
484
485
486
487
488
489
490
491
492
493
494
Based on the 30 logistical regression models, AI approval is correlated with four variables (POW.D, 495
QTNE, Freedom, GOVFREE) in construct 1: government policy and culture; four variables (Ethics, 496
TRK_MIN, TRK_CRIM and TRST_GOV) in construct 2: individual ethics and values; and two 497
variables (QCOMP, SMEDIA) in construct 3: Technology Adoption. There is an interesting finding 498
that people who think that their government guarantees individual freedom are more likely to approve 499
TRK_CRIM***/*/**
QTNE***/*
FREEDOM**/***
TRK_MIN***/***
TRST_GOV**/**
AI_CRIM AI_CRIM & AI LAW AI LAW
QCOMP*/***
POW.D*
GOV.FREE*
ETHICS**
SMEDIA**
COV+TRK
INDV_COL HIER TECH ADP TRST PVT
26
government usage of AI in criminal investigations (GOVFREE), which is contrary to other findings 500
in all constructs. 501
Another unexpected finding was that not a single variable for D1 or D3 (zero / 30) that was significant 502
for AI.Law variable only. We believe this reality implies that our respondents did delineate related to 503
the purpose of AI and that there is a part of our respondent population that has correlative responses 504
between variables AI.CRIM and AI.LAW. We might expect that TRK_CRIM – or – the ability for 505
government to track for criminal investigations may only correlate with AI_CRIM given similar 506
degree of infringement upon one’s privacy – but in fact TRK_CRIM was significant for both. 507
508
Discussion and Implications 509
In line with our view, this paper seeks to contribute an often omitted or at best underrepresented 510
question of the “hidden” social and ecological costs of artificial general intelligence and associated 511
dangers for humanity or existential threats (Mueller and Bostrom 2016; Tegmark 2017; Omohundro 512
2014, Hagendorff 2019). In line with Bostrom (2014), artificial intelligence and related technologies 513
pose risks to human civilization which will, if not properly managed, create catastrophic risks to 514
humanity. In tandem is also a question of how culture and allowing untampered surveillance and 515
overtly hostile leadership (Yun, Jung, Ashihara, 2020). 516
Our initial intuition of our construct model in which we selected 15 variables and categorized them 517
in three constructs – government and culture, individual ethics and values and technology adoption 518
proved to be statistically relevant with ten of the fifteen variables across all three constructs proving 519
to have significance for at least one variable. At a more granular level, there were details and trends 520
that we didn’t necessarily expect to find that shed light on the values of our population and how 521
27
consistently they view the role of government, ethics and technology with Government approval for 522
AI usage. 523
First, we expected to eliminate a number of variables based on initial trends and box plots. According 524
to our logistic regression model and decision to run individual regressions by variable – we predicted 525
that many variables – especially in technology would not be relevant. At the initial model stage, we 526
took great care and had many discussions on the selection of variables – as there were 55 questions 527
but a total of 128 including sub-questions. Contrary to our expectation although the box plot step was 528
useful to see trends and understand responses, we ended up running regression for all of our variables. 529
Second, our results imply that for our sample, for ten variables or two-thirds of our initial choices 530
there is definitely a high correlation between government and culture, individual ethics that are more 531
passive and technology adoption with similar results for AI for criminal purposes. In regards to 532
variables significant for both AI CRIM and AI LAW, one of the more interesting findings is that for 533
five variables respondents responded similarly. More specifically respondents answered similarly 534
between tracking for criminal purposes and minor law, belief in the governments right to track. 535
Third, contrary to our expectation Hofstede variables were not as significant as we expected, with 536
individualism - collective question related to culture and whether one’s country is hierarchal and 537
autocratic not finding significance for either AI CRIM or AI LAW. While we discussed that trust in 538
private companies, technology adoption and willingness to download a COVID+ mobile app may not 539
have relevance, we expected our sample to find significance along Hofstede’s values. Consistent with 540
these findings, Power Distance Pow.D only had minor significance related to AI.CRIM. 541
Fourth, as seen in our Venn diagram – there was not a single variable for a single response (zero / 30) 542
that was significant for our AI.Law variable only. This reality implies that our respondents did 543
28
delineate related to the purpose o,,, f AI and that there is a part of our respondent population that 544
has correlative responses between variables AI.CRIM and AI.LAW. 545
One question central to AI ethics is the role of the National Government related to values of freedom, 546
trust, ethics and privacy. We maintain that governance and sustainability require a more sophisticated 547
form of capitalism that embraces social purpose along with profit (Naz, 2014) and that a governing 548
body should transcend the individual and have the responsibility of upholding the common good. We 549
similarly hope that in line with Kant’s categorical imperative and Rawls’ conception of a democratic 550
society, that the principle of law and concept of justice is constituted based upon principles and not 551
by party lines and loyalty as prescribed by communist governments and more aligned with Confucian 552
values of piety, hierarchy and fidelity. Related to Kantian perspective, Smith and Dubbink (2011) 553
maintain, principles not only play an important role, principles are indispensable features that take on 554
a special importance in providing a basis for resolving new, complex moral problems without 555
enforcing an artificial uniformity across situations and contexts. Moral requirements are thereby 556
recognized as an expressing part of an individual’s identity as a rational, or autonomous, agent, and 557
not individual identity with varying aims, interests and aspirations (Smith and Dubbink, 2011). 558
1.9 Limitations and Future Direction 559
As with other experimental studies, this study is based on a limited sample. Our sample also focuses 560
on youth, which we acknowledge may be subjected to generational influences when we relate their 561
answers to questions of culture and also attempt to extrapolate how their values may affect 562
complicated issues related to AI usage for privacy and infringement upon their rights. We also 563
acknowledge that our China demographic data was slightly skewed in terms of gender with more 564
female respondents. While we did not incorporate gender into our study, we recognize the possibility 565
that Chinese females and males may have different views. While we were careful to have translators 566
who were well aware of the survey content, we acknowledge that there is a possibility of interpretation 567
29
based on language and slight aberrations based on the translation of terms in different languages. To 568
increase certainty, to the extent possible, we tried to clearly label each choice to enable the 569
respondents to select the answer they felt was most appropriate. 570
We believe future direction can include factor analysis of a wide range of questions to determine other 571
variables that may be correlated with AI law enforcement. Another possible direction would be to 572
include further analysis on other AI variables to determine if the disparity in AI for Criminal 573
investigations and AI for enforcement of public law (surveillance, transport) is similar. For example, 574
one possibility may be to look at the usage of AI for national security or use of AI in education. A 575
third potential avenue may be to look at nationality alone or in combination with other factors to 576
further views of the nation and culture from a categorical perspective. Lastly, we believe a larger 577
dataset that includes additional countries may illicit interesting insights into COVID impact and 578
approval of AI for law enforcement that is also connected with Hofstede values. A more diverse 579
dataset may shed insight related to government policy on cultural differences. 580
581
582
583
584
585
586
587
588
589
590
591
30
Conflict of Interest 592
The authors declare that the research was conducted in the absence of any commercial or financial 593
relationships that could be construed as a potential conflict of interest. 594
Author Contributions 595
KA responsible for survey questions and collection, KA and DG are responsible for the construct 596
model, DG and KA selected variables together, TD responsible for statistics and TD and DG 597
responsible for initial data analysis write-up; TD, DG, KA all reviewed entire document and worked 598
on data analysis results. KA responsible for writing literary review and discussion section; DG 599
responsible for Korean translation of survey; All authors edited and finalized the document. 600
601
Funding 602
The authors received no funding for this research. 603
Acknowledgments 604
This is a short text to acknowledge the contributions of specific colleagues, institutions, or agencies 605
that aided the efforts of the authors. 606
Reference styles 607
The following formatting styles are meant as a guide, as long as the full citation is complete and 608
clear, Frontiers referencing style will be applied during typesetting. 609
We provide permission to reuse and Copyright 610
All charts and tables are created for this article. 611
612
Data Availability Statement 613
The datasets [GENERATED/ANALYZED] for this study can be found in the [NAME OF 614
REPOSITORY] [LINK]. Please see the “Availability of data” section of Materials and data policies 615
in the Author guidelines for more details. 616
Data is available – will upload 617
618
619
620
621
31
References 622
AI now report 2018 M Whittaker, K Crawford, R Dobbe, G Fried, E Kaziunas, V Mathur, ... AI Now Institute at New York 623 University, 2018 624
Appenzeller, T. The AI revolution in science. Science https://doi.org/10.1126/ science.aan7064 (2017). 625
Bakineer, Onur. What do Academics say about artificial intelligence ethics? An overview of scholarship Spring, 2022. 626
Bakineer, Onur. Regulation an Artificial Intelligence Ethics: The State of Play. Seattle University, 2022. 627
Beijing Academy of Artificial Intelligence, The Beijing AI Principles (2019). 628
Bostrom, N. (2014). Superintelligence: Paths, dangers, strategies. Oxford University Press. 629
Buhmann, A. and Feiseler, C. (2022) Deep Learning Meets Deep Democracy: Deliberative Governance and 630 Responsible Innovation in Artificial Intelligence, Business ethics quarterly, Online First 24 January, p.1-34 631
Daly, A., Hagendorff, T., Hui, L., Mann, M., Marda, V., Wagner, B., ... & Witteborn, S. (2019). Artificial intelligence 632 governance and ethics: global perspectives. arXiv preprint arXiv:1907.03848. 633
Executive Office of the President National Science and Technology Council Committee on Technology, Preparing 634 for the Future of Artificial Intelligence. October 2016. Holdren et al. 2016 635
The Foundation for Constitutional Government, 2022. Great Thinkers, Immanuel Kant. 636 https://thegreatthinkers.org/kant/introduction/ 637
Friend, T. (2018). How frightened should we be of AI. The New Yorker, 14. 638
Hagendorff, Thilo. The Ethics of AI Ethics: An Evaluation of Guidelines. Minds and Machines (2020) 30:99-120. 639
Harari, Y. N. Reboot for the AI revolution. Nature 550, 324–327 (2017). 640
Hofstede, G. (2011). Dimensionalizing Cultures: The Hofstede Model in Context. Online Readings in Psychology and 641
Culture, Unit 2. Retrieved from http://scholarworks.gvsu.edu/orpc/vol2/iss1/8 642
Hofstede, G. (2001), Culture’s Consequences: Comparing Values, Behaviors, Institutions, and Organizations Across Nations, 643 2nd ed. Sage, Thousand Oaks, CA. 644
Hofstede Insights Website: https://www.hofstede-insights.com/fi/product/compare-countries/ 645
Hsu, J. Microsoft’s Brad Smith on How to Responsibly Deploy AI Microsoft’s president talks about the promise and 646 perils of artificial intelligence. IEEE Spectrum, April, 2019. 647
The IEEE Global Initiative on Ethics of Autonomous and Intelligent Systems 2019 648
Jobin., A., Ienca, M., Vayena, E. The global landscape of AI ethics guidelines. Nature of Machine Intelligence, Vol 1., 649 September 2019. 650
Kim, Junic, Ashihara, K. National Disaster Management System: COVID-19 Case in Korea 651
Lipford, J. and Slice, J (2007). Adam Smith's Roles for Government and Contemporary U.S. Government Roles: Is the 652 Welfare State Crowding Out Government's Basic Functions? Independent Institute Vol. 11, No. 4 (Spring 2007), pp. 653 485-501 (17 pages) 654
Montreal Declaration for Responsible Development of Artificial Intelligence (2018); montrealdeclaration – 655 responsibleai.com 656
MIT-BCG 2020 Artificial Intelligence Global Executive Study and Research Project 657
32
Müller, V. C. (2020). Ethics of artificial intelligence and robotics. 658
Müller, V. C., & Bostrom, N. (2016). Future progress in artificial intelligence: A survey of expert opinion. 659 In Fundamental issues of artificial intelligence (pp. 555-572). Springer, Cham. 660
Muthensenthil, B. Kim, HS. CCTV Surveillance System, attacks and design goals. International Journal of Electrical and 661 Computer Engineering (IJECE) Vol. 8, No. 4, August 2018, pp. 2072~2082 662
Naz, F. (2014). Adam Smith’s Model of Capitalism and Its Relevance Today. Filosofía de la Economía, 2014, Vol. 3, pp. 663 71-85 664
OECD 2019 Principles of AI https://www.oecd.org/digital/artificial-intelligence/ 665
Omohundro, S. (2014) Autonomous technology and the greater human good, Journal of Experimental & Theoretical 666 Artificial Intelligence, 26:3, 303-315 667
Ordinal Logistic Regression | R Data Analysis Examples. https://stats.oarc.ucla.edu/r/dae/ordinal-logistic-regression/ 668
Orwell, G. (2021). Nineteen Eighty-Four. Penguin Classics 669
Pekka, A. P., et al. "The European Commission’s high-level expert group on artificial intelligence: Ethics guidelines for 670 trustworthy ai." Working Document for stakeholders’ consultation. Brussels (2018): 1-37. 671
Ryan, M., & Stahl, B. C. (2020). Artificial intelligence ethics guidelines for developers and users: clarifying their content 672 and normative implications. Journal of Information, Communication and Ethics in Society. 673
Smith, B., Browne, CA, Gates, B. 2019. Tools and Weapons: The Promise and the Peril of a Digital Age. 674
Smith, J., & Dubbink, W. (2011). Understanding the Role of Moral Principles in Business Ethics: A Kantian 675 Perspective. Business Ethics Quarterly, 21(2), 205-231. 676
Stone, P., Brooks, R., Brynjolfsson, E., Calo, R., Etzioni, O., Hager, G., ... & Teller, A. (2022). Artificial intelligence and life 677 in 2030: the one hundred year study on artificial intelligence. 678
Tegmark, M. (2017). Life 3.0: being human in the age of artificial intelligence. First edition. New York, Alfred A. Knopf. 679
Walter, S. D., Feinstein, A. R., & Wells, C. K. (1987). Coding ordinal independent variables in multiple regression 680 analyses. American Journal of Epidemiology, 125(2), 319-323. https://doi.org/10.1093/oxfordjournals.aje.a114532 681
Yun, DW, Jung HJ, Ashihara, K. Dimensions of Leader Aner Expression Unveiled: How Anger Intensity and Gender of 682 Leader and Observer Affect Perceptions of Leadership Effectiveness and Status Conferral. Frontiers, Psychology, July, 683 2020. 684
685
Acknowledgements 686
We are grateful and wish to acknowledge contributions from Emeritus Professor Michael Ginzberg for his 687 input in the construct model and methodology input, editing, meeting face-to-face in Korea and taking many 688 calls. We wish to acknowledge the input and feedback from our advisory team including Professor Onur 689 Bakineer and EVP at Samsung SDS, Dan Paik. We appreciate the time and energy of Flavia Cortes-Carignan 690 for her contribution, inspiration and face-to-face meetings in Seoul. Thanks to Teddy Yu – who translated our 691 survey into Chinese and worked on attaining the majority of our survey responses from mainland China. 692 Thankful for Team leaders Amanda Kobayashi translation into Spanish, Emily Eruth for translation into 693 German. and recognize the participation of sections in 2020 and 2021 and 13 sections of Business 694 Communications at Konkuk University who assisted with brainstorming, survey development and initial data 695 collection – as well as genuine interest in bringing this research to fruition. 696
702
