Social, Legal, and Technical Considerations for Machine Learning
and Artificial Intelligence Systems in Government
Richard Dreyling
a
, Eric Jackson
b
, Tanel Tammet
c
, Alena Labanava
d
and Ingrid Pappel
e
Department of Software Science, School of Information Technologies, Tallinn University of Technology,
Akadeemia Tee 15a, Tallinn, Estonia
Keywords: Artificial Intelligence, Machine Learning, Government Services, e-Government, e-Governance, AI Ethics,
ML Ethics, AI Bias, ML Bias, Estonia.
Abstract: Expansion of technology has led to governments increasingly reconciling with advanced technologies like
machine learning and artificial intelligence. Research has covered the ethical considerations of AI as well as
legal and technical aspects of the operation of these systems within the framework of government. This
research is an introduction to the topic in the Estonian context which uses a multidisciplinary inquiry based
in the theoretical framework of technology adoption and getting citizens to use these services for their benefit.
(Suggest that there are first results as well).
1 INTRODUCTION
The twenty-first century has brought with it the
expansion of digital transformation in the public and
private sectors. Information and communications
technologies have been used by the public and private
sectors to enhance efficiency and service delivery.
Since the introduction of the microchip in 1971, the
technological revolution has changed the way
businesses conduct affairs as well as the ways in
which governments handle governance tasks (Perez,
2002, 2010). The advent of the internet and the
information technology boom has changed not only
the ways that bureaucrats can govern, but also the
items which must be governed. Expansion of
technology provides new ways for businesses and
citizens to push against laws in ways that
governments could not have imagined at the advent
of the microchip.
Governments have adopted E-government
methodologies and platforms to be able to use
information and communications technologies to
streamline the business processes of government and
deliver services to citizens in a more efficient manner.
a
https://orcid.org/0000-0001-7269-6236
b
https://orcid.org/0000-0003-3726-5814
c
https://orcid.org/0000-0003-4414-3874
d
https://orcid.org/0000-0001-6318-7704
e
https://orcid.org/0000-0002-5148-9841
One country that has developed a reputation for the
use of ICTs in service provision is Estonia. The small
Baltic country has put a lot of effort into digitizing
many government services. They offer many services
online with the ability for citizens to accomplish the
majority of their interactions with the government
through authentication through various forms of
electronic ID. The country has worked to minimize its
digital divide, ranked as the twelfth most inclusive
country in the world in a recent index (Economist
Intelligence Unit, 2020). The combination of a tech
savvy populace that also trusts its government has
helped these efforts be successful. Since the 2000’s
Estonia has offered increasing government service
offerings online with electronic identification (eID)
and data exchange between government entities in a
secure and tracked manner. They have even been
successful in bringing e-Government to local
municipalities and attracting people to virtual
residency through their e-Residency program (Pappel
et. al., 2015) (Kimmo et. al., 2018).
The expansion of computing power since the
early 2010s, driven by graphics processing units has
allowed for the expansion of artificial intelligence and
Dreyling, R., Jackson, E., Tammet, T., Labanava, A. and Pappel, I.
Social, Legal, and Technical Considerations for Machine Learning and Artificial Intelligence Systems in Government.
DOI: 10.5220/0010452907010708
In Proceedings of the 23rd International Conference on Enter prise Information Systems (ICEIS 2021) - Volume 1, pages 701-708
ISBN: 978-989-758-509-8
Copyright
c
2021 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
701
machine learning research. Governments across the
world have begun to use AI and ML in the conduct of
government business and governance to try to better
deliver services to citizens and in some cases control
them.
However, Estonia would like to go further in
using technology to help make life better for its
citizens. In March of 2020, the Chief Technology
Officer of Estonia launched a Next Generation Digital
State Architecture Vision Paper. In this document the
CTO discusses the concept of AI enabled virtual
assistants to help achieve easier access to government
services (Vaher, 2020). In Estonia, the public sector
has a history of cooperating with academia in the
country to ensure that the public officials were
following the best available science at the time.
Because of this cooperation, the research began after
the release of the paper to investigate and support the
topics laid out by the CTO through academic
research.
This introductory research seeks to find the
answer to the main research question that asks, “How
can virtual assistant systems affect eGovernance
services in Estonia?” This multidisciplinary paper
will address the ways in which virtual assistant
systems can enable government services in Estonia,
what particular challenges are inherent to the general
practice of using AI and machine learning in
government and the specific case. This paper seeks to
introduce this research topic as well as formalize the
research gaps involved and lay out a roadmap and
preliminary results regarding automation of
government services and enablement through Next
Generation Digital Government Architecture
(NGDA) initiative. This paper will be an overview of
the challenges that AI and ML enabled programs in
government face from a legal, technical, and social
perspective and how stakeholders in current active
pilot programs in Estonia intend to contend with these
challenges
.
2 STATE OF THE ART
2.1 Introduction to Estonian
e-Government Systems
The Estonian government has used technology as a
way to ameliorate the issues caused by having a small
population from which they can hire government
employees. The Estonian government now has almost
all services able to be completed by eID validated
transactions online. The key building blocks
necessary for this infrastructure from a technological
perspective are the electronic ID and the Estonian
implementation of a data exchange layer they call “X-
Tee” or “X-Road” in English. All the official
identification cards have a cryptographic chip capable
of electronic authentication and giving signatures to
documents. This enables use of a public key
infrastructure (PKI) that enables encryption and
digital signing of documents and transactions that are
secure and legally binding. The X-Road acts as a data
exchange layer. Developed in the early 2000’s. X-
Road uses security servers to authorize service clients
and service providers. Any transaction, to include
making changes to data or accessing data, registers
with the time-stamping server and leaves a trace.
Through this architecture, they ensure authentication,
authorization, and accounting (Vaher, 2020). The
time stamping server leaves a time hack on any
transaction, which must be accompanied by an eID
signature. Estonia ensured at the time that these
innovations came into use that they included the
social aspects, legal framework, and technical aspects
of the solution all were primed in order to encourage
use of the solution. The state subsidized the purchase
of the ID cards containing the eID signing ability, as
well as partnered with banks to make the IDs useful
for logging into internet banking and completing
transactions. The country also chose the best
technical solution for eID, and has continued to
handle any technical or security issues that have
arisen from the non-compliance to best practices by
contractors (Lips et. al., 2018). This enhances trust
among the citizenry which is a likely factor in the
strong adoption of the Estonian population of e-
services.
Similar to other contexts, when a country is an
early adopter of new technologies, technical debt and
other phenomena can make further innovation a
difficult task. The vision paper released by the Chief
Technology Officer (CTO) of Estonia proposes
methods to continue the path of innovation in the area
of public sector service implementation. Some of
these initiatives primarily focus on updating the
technology currently in use in the Estonian
eGovernance architecture. These include moving
from monolithic applications toward an event driven
microservices architecture. More than simply
discussing some architectural changes, this paper
outlines a vision that would have Estonians
conducting government services through virtual
assistants.
As outlined in the NGDA paper the uses for
artificial intelligence and machine learning in
government are called “Kratt.” This name is based on
ICEIS 2021 - 23rd International Conference on Enterprise Information Systems
702
an entity from Estonian mythology (Scholl &
Velsberg, 2020). KrattAI “is first a vision of how
public services should digitally work in the age of
artificial intelligence” (Sikkut et. al., 2020). When
the Estonian Government refers to a “Kratt” this
specifies a use of AI or ML, whereas the specific
signifier “KrattAI” is the initiative that focuses on the
aforementioned provision of government services
that use the human computer interaction method of
virtual assistants or chatbots (Scholl & Velsberg,
2020).
2.2 Technology Adoption Theories
One area of research has tried to codify the factors
which can help to predict whether a citizen or
employee will adopt a piece of technology. The area
of technology adoption models began with the Theory
of Reasoned Action (TRA) in 1975, which focused
primarily on a social psychological explanation of
people’s perceptions and norms (Fishbein and Ajzen,
1975). Fishbein and Ajzen then expanded TRA into
the Theory of Planned Behavior (TPB). From these,
the research expanded into many different theories
related to the adoption of technology in different
contexts. Some of these include the Technology
Acceptance Model (TAM), the expansions of TAM,
including TAM2 and TAM3, as well as The Unified
Theory of Acceptance and Use of Technology
(UTAUT), and (UTAUT2). Each of these have
various identified ontologies of factors which the
researchers believed would affect technology
adoption. Some of these theories have similarities that
help to show the importance of factors that would
encourage successful execution of projects
containing machine learning and artificial
intelligence. For example, in the Technology
Acceptance Model’s third version (TAM3) some of
the determinants include the perceived ease of use of
a piece of technology. These factors are “computer
anxiety,” “perceived enjoyment,” “objective
usability,” as well as “perceived usefulness” from
earlier TAM models (Venkatesh and Bala). In the
Unified Theory of Acceptance and Use of
Technology (UTAUT) the determinants of “effort
expectancy,” and “performance expectancy” are
relevant to the specific challenges of AI and ML
based systems in government, even though this model
originally considered the corporate sphere
(Venkatesh et al., 2003). These factors from a
theoretical perspective can be considered proxies for
the general concepts of effectiveness, usefulness, and
usability. These concepts show the reasons that
practitioners in the government would want to ensure
that a tool that uses AI and ML are useful, effective,
and usable by everyday citizens. In further research
conducted on technology adoption shows trust to be
an important factor in the use of e-government
services (Grimsley & Meehan, 2007), (Colesca 2005,
pp.39), (Carter & Bélanger, 2005). In addition,
further research stated that trust is one of the most
important factors related to “behaviour intention”
(Alharbi et. al., 2016, pp. 1). For the solution to be
successfully adopted in the populace, trust could be a
key factor. The theories regarding technology
adoption also apply to adoption of artificial
intelligence and machine learning in government.
Specific factors in the areas social, technical, and
legal concerns will have an effect on the success of
the Estonian Next Generation Digital Government
Architecture (NGDGA) and its artificial intelligence
related proposals
2.3 Social Perspective
Specific social challenges exist related to the
effectiveness, usefulness, and usability of machine
learning and artificial intelligence initiatives in
government. One of the main challenges to AI and
ML initiatives is that these will end up enhancing
current disparities through the digital divide, and bias.
One issue that causes concerning social factors is
research related to bias in AI and ML. A report called
Government by Algorithm suggests that three
findings became apparent in their investigation of the
literature. They found that “the potential for machine
learning to encode bias is significant” (Freeman
Engstrom, et al., 2020). The researchers used the
example of criminal risk assessment scores in the
United States that have different rates of false
positives for those of different ethnic groups
(Freeman Engstrom, et al., 2020). The reasons for this
are that AI can become biased due to programming or
training, based on the data inputted to train the model,
which can have the effect of making bias integral to
the decision making of the AI (Mehr 2017)(Center for
Public Impact, 2017). In addition, proposed methods
of keeping machine learning fair can potentially not
co-exist if these methods must have more than one
definition of “fairness” (Freeman Engstrom, et al.,
2020). If considering multiple groups of people who
have multiple differences in race or gender it is
impossible to ensure that all possible key
performance metrics are equal across the groups
(Freeman Engstrom, et al., 2020). The report also
pointed out the necessity to consider how human and
AI-assisted decisions correlate with one another
because the bias in the AI and ML decisions comes
Social, Legal, and Technical Considerations for Machine Learning and Artificial Intelligence Systems in Government
703
from the human decision making (Freeman
Engstrom, et al., 2020).
The context of the above review of the literature
was the United States. However, the European
Parliamentary Research Service has also considered
bias in these issues. They explain a resolution adopted
by the European Parliament in 2019. The report
states, “'any AI model deployed should have ethics by
design'. The resolution specifically mentions four sets
of issues in relation to the ethical discussion: 1)
human-centric technology; 2) embedded values in
technology – ethical-by-design; 3) decision-making –
limits to the autonomy of artificial intelligence and
robotics and 4) transparency, bias and explainability
of algorithms (pp. 9). The European Parliament
guidance on these systems recommends that any AI
or ML based system does not perpetuate bias by
ensuring ethical behavior integration in systems.
When taken into account this in a practical sense puts
the responsibility of making sure that bias and lack of
ethics do not perpetuate current disparities.
2.4 Legal Considerations
Any Estonian implementation using AI for
government purposes should comply with Estonian
and European Law with regard to automated decision
making and data protection. In the European Union at
the moment there are competing existing frameworks
for adopting AI. One assessment suggested that, “a
common EU framework on ethics has the potential to
bring the European Union €294.9 billion in additional
GDP and 4.6 million additional jobs by 2030” (Evas,
2020 pp. 1). Beyond the general approach to data
protection brought by the GDPR, Europe does not
have specific legislation dictating how member states
can implement AI in their countries. However,
Estonia has a law that may impact the ability for AI
to achieve what could be considered its full potential.
The Personal Data Protection act passed in 2018
has provisions that give specific purposes and criteria
that need to be met for data processing which could
mean that organizations other than the one which
collected the data are unable to use AI or ML
applications to provide services (Personal Data
Protection Act, 2018). This law also provides specific
criteria that must be met for automated decision
making. According to some legal experts, one of
these criteria means that the only two state registers
which would qualify are the land register and
company register because they are “considered
having legal effect” (Kerikmäe & Pärn-Lee, 2020 pp.
6). In practice this means leads to the hypothesis that
that any automated capability would be used more as
a decision support system for a human decision
maker. This law also has ramifications for technical
best practices that will be discussed in the following
section. In addition, the cross-border aspect of the
data sovereignty requirements put in place by GDPR,
the US CLOUD Act and the Estonian PDPA may
make integration with the large virtual assistant
providers complicated (Varughese, 2020).
2.5 Technical Concerns
The vision for a next generation digital government
architecture must overcome technical challenges to
ensure success. Although chatbots originated in
private sector use cases, researchers have studied
chatbots as a method of allowing consumers to
directly speak through an AI mediated platform to
government entities to assist in completing tasks
(Akkaya & Krcmar, 2019) (Freeman Engstrom & Ho,
2020) (Androutsopoulou et. al., 2018) (Mehr, 2017).
A chatbot is a system that has to accomplish several
tasks. The chatbot must use natural language
processing be able to interpret intent of a customer or
citizen. After understanding intent, the bot should be
able to complete the required tasks or connect the
citizen with the relevant stakeholders to help assist
them in completing the task. A chatbot may use
supervised learning and when properly trained will
improve its ability to operate the more it is used.
Data is a key factor in the accuracy of machine
learning and artificial intelligence systems. Estonia
has had over twenty years of e-government service
experience. Because of this, they have accumulated
massive amounts of data and have done a better job
than some other countries of ensuring this data is
machine readable (Scholl & Velsberg, 2020). The
way the Estonian PDPA has been put into practice
makes one legal challenge into a technical challenge.
Estonia follows the “once only principle,” which
means that data is stored where it is collected and the
citizen should not have to provide it to other
government authorities. For example, if the police
would like to know a person’s address, they should
query the population registry database. This leaves a
signature through X-Road, the data exchange layer.
When discussing an AI system though, even though
the Estonian government may have more data
available it is in various databases around the country.
Researchers have attempted to ameliorate some of the
organizational issues related to data, quality, and
formatting in Estonia (Tepandi et. al., 2017). Because
of this, there is no massive data pool from which the
chatbots could be trained. This theoretically would
make it difficult for the chatbot and virtual assistant
ICEIS 2021 - 23rd International Conference on Enterprise Information Systems
704
programs to be able to gain the accuracy necessary to
achieve instant citizen uptake. Although, they could
get better as time continues if the proper training and
feedback mechanisms were implemented into the
workflows of the system.
The NGDGA document elaborates on a vision in
which chatbots would move beyond a single instance
on a website toward a virtual assistant model. One of
the options could be to integrate the Estonian
government’s hypothetical chatbot with the large
virtual assistant providers to provide a more robust
experience for the citizen (Vaher, 2020). This poses
an issue because the Estonian language does not have
support in the large virtual assistant providers or the
existing translation APIs are not sufficient in quality.
The language issue and the method of integration
with virtual assistant providers are issues that must be
solved.
A report regarding the United States Federal
Government’s adoption of AI and ML mentions the
concept of internal and external competencies
(Freeman Engstrom, et al., 2020). They found that
some of the most successful implementations were
created by employees of the government who were
hired in a capacity such as lawyers and then
developed their own machine learning and artificial
intelligence capacity on their own time. They
recommended to government procurement personnel
in the US context to not simply outsource the
development of AI and ML projects to private sector
contractors. They found that the in-house developed
solutions solved some of the issues with data access
and source code access that outsourced projects
experienced. In the United States the private sector
has the advantage when it comes to AI and ML
experience. However, Estonia has shown in recent
years a propensity to use public private partnerships
(PPP) to procure technological expertise that leads to
successful projects when the need arises (Paide et.
Al., 2018).
Harvard researchers identified five potential use
cases for chatbots in the public sector which included,
“(i) answering citizens' questions, com- plaints and
inquiries through automated AI-based customer
support systems, (ii) searching in documents
(including legal ones) and providing guidelines to
citizens on filling forms, (iii) getting citizens' input and
routing them to the responsible public administration
office, (iv) translating governmental information, and
(v) drafting documents with answers to citizens'
questions” (Mehr, 2017) (Androutsopoulou et. al.,
2018). The vision put forth by the Estonian
government goes further than this and calls for the
virtual assistant technology to be able to help the
citizen complete tasks (Vaher, 2020). The Mehr
report quotes, CEO of Synthesis Corp. Ari Wallach,
“’Imagine having direct and constant access to a high-
level government concierge that is constantly
learning and improving” (2017, pp. 10). This entails
having a system that can constantly learn through
supervised learning across data sets and stepping into
territory which governments have not tread before at
scale.
3 METHODOLOGY
To better investigate the current and future states of
eGovernance with AI and ML enabled virtual
assistants, qualitative methods were used. A review
of recent literature served to get preliminary
information. In addition, two workshops were
conducted to elicit feedback from groups of experts
who are stakeholders in the Estonian eGovernance
context. Qualitative research has the inherent issue of
bias. However, the workshop format and its semi-
structured nature gives the participants the ability to
express themselves freely and to communicate the
way they perceive the issues at hand (Yin, 2014). Due
to the early investigatory nature of the research at
hand, the qualitative methods have the largest amount
of flexibility to gather information to determine the
future path of research. This methodology allows for
the researcher to get the maximum amount of
information from the experts in the field rather than
have them conform to already existing theories and
phenomena (Gioia et. al., 2012). This represents the
best way to ensure that the researchers would not ask
leading questions that bias responses when discussing
the topics with experts and stakeholders in
workshops. The workshops included stakeholders
from the Nordic Institute for Interoperability
Solutions (NIIS), stakeholders from the Ministry of
Economic and Social Affairs of Estonia (MKM) as
well as the software development company that is
developing the KrattAI chatbot proof of concept
(POC).
4 DISCUSSION AND RESULTS
Artificial Intelligence use can be considered to be
controversial. Apart from the popular culture
depictions of artificial intelligence as an antagonist
force toward humanity, there exists a lot of literature
on the topic. In section two, a review of some of the
social, legal and technical concerns explored some of
Social, Legal, and Technical Considerations for Machine Learning and Artificial Intelligence Systems in Government
705
the issues that a government implementation of AI
and ML would have to avoid.
The workshop led to a discussion of these topics
and how the Estonian government plans to ameliorate
some of the issues presented in section two. The
Estonian vision of may be considered one of the more
recent developments in government services due to
the initiation of the chatbot proof of concept to
eventually directly provide services to citizens.
Estonia is working right now to traverse the
challenges and barriers which have been pointed out
above. From the workshops with stakeholders the
researcher gained insights into how the social. legal,
and technical challenges have shaped the pilot
programs in Estonia. Many of these are interrelated
and will be presented in a manner which
acknowledges this factor. These methods can inform
the ways that other governments may shape their
programs to help ameliorate some of the difficult
points concerning AI and ML based initiatives.
From a social perspective, getting feedback from
users both inside and outside of the government is
important for the stakeholders in the various AI and
ML programs. This concerns the theoretical
grounding of technology adoption in a practical
manner. One thing that a stakeholder observed was
that though the team tried their best to make the
instructions and all relevant materials in as clear
language as possible, they got the feedback that some
of the directions were too complex for those not
already embedded in the IT world. This allowed them
to ensure that by the time the services roll out to
citizens and ordinary government workers, the
likelihood of adoption will increase because they can
iterate until usability has increased. They look at
usability not only of the end user but of all the
stakeholders in the chain who will be using
During the discussions, stakeholders
acknowledged the potential for machine learning and
artificial intelligence derived bias. However, they
pointed out that the Estonian government has signed
onto and helped shape the European Parliament’s
suggestions relating to ethical AI and controls against
bias. And in the areas in which there are no standards
that are universally accepted, the people in the
Estonian government who manage AI suggest them
to governing bodies. This helped to shape the way the
Estonian government set up the chatbot POC that is
the initial step toward the KrattAI vision as well as
other Kratts. They decided from the beginning that
whenever an AI or ML enabled decision support
system would have a decision point that directly
affects a citizen’s service provision, in accordance
with the Estonian law on automation, that a human
decisionmaker would be there to make the final
decision in some cases. Kerikmäe & Pärn-Lee
summarized the guidelines dictating the law in
practice as follows, “Human interaction should take
place only if the algorithm result turns out negative or
if the subject of the administrative decision disputes”
(2020 pp. 6). This still does not completely solve the
issue of bias due to human decisionmakers over time
causing the bias, but it does take steps toward
preventing hardcoded bias. Deference of human
decisionmakers to automated decision systems is
another potential source of problems in this area
(Freeman Engstrom, et al., 2020). The stakeholders in
this situation use the predictive, prioritization, and
optimization abilities from AI to help in areas that the
citizen and the government benefit from, not as a
punitive function like using AI imagery analysis to
determine subsidy compliance based on whether
farmers have mowed their land or not. Instead of
fining a farmer based on the results, the government
would contact the farmer to ask the situation.
Sometimes the farmer would have mowed the farm
earlier in the year or be ready to do it. This saves
government resources from doing on the spot
investigation of each farm and farmers appreciate the
ability to discuss with officials (Scholl & Velsberg,
2020).
In addition, there are some useful capabilities
inside the government which can use AI and automate
items that have no decision impact on the citizen but
increase the ability for government responsiveness to
the citizen. An example of this is internal email
forwarding. The Estonian government had a massive
problem with citizens emailing officials, employees,
or department email addresses requesting information
on where to direct their inquiries. One stakeholder
mentioned specifically that in addition to normal
government duties, some employees had to handle
over 1500 emails a day. Some departments have been
able to institute decision engines that look for similar
inquiries and send responses automatically. This is an
example of a situation where the laws as currently
written allow for automated decision making. The
government also gets feedback from the citizen to see
if this forwarding solved their issue. However, it must
be mentioned that this process is done on a
department-by-department basis and has not been
implemented across the entire government.
The method of handling the chatbot inquiries in
the absence of a united data pool is novel and also
helps solve the issue of referring citizens to the right
authorities. The design of the KrattAI chatbot POC is
to have networks of many chatbots with their own
knowledge which can speak to each other. They do
ICEIS 2021 - 23rd International Conference on Enterprise Information Systems
706
not store the data from the transaction. This way,
when a citizen contacts the chatbot and asks a
question, the chatbots can refer the citizen to the
chatbot with the proper knowledge base. The KrattAI
chatbot POC is not yet to the point of executing
government transactions but the POC has proven that
a network of chatbots can allow for the proper
functioning to find the proper chatbot for a
transaction. This method maintains the legal
boundaries put into place by Estonia while effectively
handling the technical concerns from not having large
data pools with which they can train the NLP engines
of the chatbots.
According to the workshop attendees, in
agreement with the NGDA vision paper, there are
changes in the current E-governance architecture are
necessary to enable the vision of virtual assistant
enabled services. One change that still must be made
is moving X-Road from a synchronous
communication mode to an asynchronous version of
communication. This could include publish,
subscribe messaging patterns. The CTO has called
this change introducing X-Rooms. X-Rooms would
allow more than one verified entity to be party to the
communication being passed and not require that both
entities be connected at the exact same time. This is
key for the vision to be achieved with virtual assistant
driven services.
With a PPP the Estonian authorities have
managed design, code, and test a system that uses AI
and ML for the benefit of the citizen while attempting
manage the difficulty points of these types of projects.
Limitations of the research are that the number of
interactions with stakeholders were few. The projects
are also not that far along. The specific partnership
potential with public virtual assistant providers is not
able to be discussed and legally very complex.
Because of these legal complexities, the options for
integration to make the chatbot POC able to use
virtual assistant capabilities would be conjecture.
Future work will take a specific case for which the
virtual assistant capability could be used, and follow
the business processes as well as specific technical
processes through to the end of the transaction. If
possible, an artefact will be designed to help solve a
technical issue pertinent to initiatives of similar
purpose.
REFERENCES
Ajzen, I and Fishbein M., 1980. Understanding Attitudes
and Predicting Social Behavior Englewood Cliffs, NJ:
Prentice-Hall, Inc.
Ajzen, I., 1991 The Theory of Planned Behavior,
Organizational Behavior and Human Decision
Processes 50, pp. 179–211.
Ajzen, I., 2010. The Theory of Planned Behavior,
http://www.people.umass.edu/aizen/tpb.html.
Akkaya, C. and Krcmar, H. 2019. Potential Use of Digital
Assistants by Governments for Citizen Services: The
Case of Germany. In Proceedings of dg.o 2019: 20th
Annual International Conference on Digital
Government Research, dg.o 2019, June 18, 2019,
Dubai, United Arab Emirates. ACM, New York, NY,
USA, 10 pages. https://doi.org/10.1145/3325112.3325
241
Alharbi, N., Papadaki, M., & Dowland, P., 2016. The
impact of security and its antecedents in behaviour
intention of using E-Government services. Behaviour &
Information Technology, 36(6), 620–636.
doi:10.1080/0144929x.2016.1269198
Androutsopoulou, A., Karacapilidis, N., Loukis, E., &
Charalabidis, Y. 2019. Transforming the
communication between citizens and government
through AI-guided chatbots. Gov. Inf. Q., 36, 358-367.
Carter, L., & Bélanger, F., 2005. The utilization of E-
Government services: citizen trust, innovation and
acceptance factors. Information Systems Journal, 15(1),
5–25. doi:10.1111/j.1365-2575.2005.00183.x
Centre for Public Impact, 2017. Destination unknown:
Exploring the impact of Artificial Intelligence on
Government. Retrieved from https://publicimpact.blo
b.core. windows.net/production/2017/09/Destination-
Unknown-AI-and-government.pdf.
Colesca, S.,. 2009. Increasing e-trust: A solution to
minimize risk in E-Government adoption. Journal of
Applied Quantitative Methods. 4.
Economist Intelligence Unit. 2020. The inclusive Internet
Index 2020. https://theinclusiveinternet.eiu.com/
explore/countries/performance
Engstrom, D.F., Ho, D.E., Sharkey, C.M., Mariano-
Florentino, C., 2020. Government by Algorithm:
Artificial Intelligence in Federal Administrative
Agencies: Report Submitted to the Administrative
Conference of the United States.
Engstrom, David Freeman and Ho, Daniel E., 2020.
Artificially Intelligent Government: A Review and
Agenda (March 9, 2020). Big Data Law (Roland Vogl,
ed., 2020, Forthcoming), Available at
SSRN: https://ssrn.com/abstract=3551549
European Parliament resolution of 12 February 2019 on a
comprehensive European industrial policy on artificial
intelligence and robotics (2018/2088(INI)).
Evas, T., 2020. European framework on ethical aspects of
artificial intelligence, robotics and related technologies.
European added value assessment. European
Parliamentary Research Service, European Value
Added Unit.
Gioia, D. A., Corley, K. G., & Hamilton, A. L., 2012.
Seeking Qualitative Rigor in Inductive Research.
Organizational Research Methods, 16(1), 15-31.
doi:10.1177/1094428112452151
Grimsley, M., & Meehan, A., 2007. E-Government
Social, Legal, and Technical Considerations for Machine Learning and Artificial Intelligence Systems in Government
707
information systems: Evaluation-led design for public
value and client trust. European Journal of Information
Systems,16(2),134–148. doi:10.1057/palgrave.ejis.30
00674
Kerikmäe, T., Pärn-Lee, E., 2020. Legal dilemmas of
Estonian artificial intelligence strategy: in between of
e-society and global race. AI & Soc.
https://doi.org/10.1007/s00146-020-01009-8
Kimmo, M.; Pappel, I; Draheim, D., 2018. E-residency as a
nation branding case. Proceedings of the 11th
International Conference on Theory and Practice of
Electronic Governance, ICEGOV2018: 4 - 6 April
2018, Galway, Ireland. Ed. Kankanhalli, Atreyi; Ojo,
Adegboyega; Soares, Delfina. New York, NY: ACM
Press, 419−428. (ACM International Conference
Proceedings Series).10.1145/3209415.3209447.
Lips, S., Pappel, I., Tsap, V., Draheim, D., 2018. Key
factors in coping with large-scale security
vulnerabilities in the eID field. Electronic Government
and the Information Systems Perspective: 7th
International Conference, EGOVIS 2018, Regensburg,
Germany, September 3-5, 2018, Proceedings. Ed. Kö,
Andrea; Francesconi, Enrico. Cham: Springer Verlag,
60−70. (Lecture Notes in Computer Science;
11032).10.1007/978-3-319-98349-3_5.
Mehr, H. 2017. Artificial intelligence for citizen services
and government. Cambridge, MA: Harvard Kennedy
School, Ash Center for Democratic Governance And
Innovation. Retrieved January 10, 2021, from
https://ash.harvard.edu/files/ash/files/artificial_intelligi
nte_for_citizen_services.pdf
Paide, K., Pappel, I., Vainsalu, H., & Draheim, D. (2018).
On the Systematic Exploitation of the Estonian Data
Exchange Layer X-Road for Strengthening Public-
Private Partnerships. Proceedings of the 11th
International Conference on Theory and Practice of
Electronic Governance - ICEGOV ’18.
doi:10.1145/3209415.3209441.
Pappel, I., Tsap, V., Draheim, D., 2020. The e-LocGov
Model for Introducing e-Governance into Local
Governments: an Estonian Case Study. IEEE
Transactions on Emerging Topics in Computing. DOI:
10.1109/TETC.2019.2910199 [ilmumas].
Perez, C., 2002. Technological Revolutions and Financial
Capital, Books, Edward Elgar Publishing, number
2640.
Perez, C., 2010 Technological Revolutions and Techno-
economic paradigms. Cambridge Journal of
Economics, Vol. 34, No.1, pp. 185-202
Personal Data Protection Act, (AKI). (Est). 2018.
Scholl, R., Velsberg, O. 2020. AI FOR GOOD LIVE: How
Estonia builds the next generation e-government with
AI use cases. International Telecommunications Union.
https://www.youtube.com/watch?v=O3jiLbYVLQU
(Nov. 10, 2020)
Sikkut, S., Velsberg, O., Vaher, K. 2020. #KrattAi: The
Next Stage of Digital Public Services in #Estonia,
Vision and Concept.
Tepandi, J., Lauk, M., Linros, J., Raspel, P., Piho, G.,
Pappel, I., Draheim, D., 2017. The data quality
framework for the Estonian public sector and its
evaluation: establishing a systematic process-oriented
viewpoint on cross-organizational data quality. In:
Hameurlain, A.; et al. (Ed.). Transactions on Large-
Scale Data- and Knowledge-Centered Systems XXXV
(1−26). Berlin: Springer. (Lecture Notes in Computer
Science; 10680).10.1007/978-3-662-56121-8_1.
Thong, James Y. L.; and Xu, X., 2016. Unified Theory of
Acceptance and Use of Technology: A Synthesis and
the Road Ahead, Journal of the Association for
Information Systems: Vol. 17: Iss. 5, Article 1. DOI:
10.17705/1jais.00428
Vaher, K. 2020. Next Generation Digital Government
Architecture.
Varughese, B., 2020. Cross- border data transfer in the
context of the GDPR and CLOUD Act. Tilburg Institute
for Law, Technology, and Society (TILT) LLM. Law
and Technology.
Venkatesh, V. & Bala, H. 2008. Technology Acceptance
Model 3 and a Research Agenda on Interventions.
Decision Sciences - DECISION SCI. 39. 273-315.
10.1111/j.1540-5915.2008.00192.x.
Venkatesh, V., Morris, M., Davis, G., & Davis. F., 2003.
User Acceptance of Information Technology: Toward a
Unified View. MIS Quarterly, 27(3), 425.
doi:10.2307/30036540
Venkatesh, V., Thong, J. Y. L.; and Xu, X. 2016. "Unified
Theory of Acceptance and Use of Technology: A
Synthesis and the Road Ahead," Journal of the
Association for Information Systems: Vol. 17: Iss. 5,
Article 1. DOI: 10.17705/1jais.00428
Yin, R., 2014. Case Study Research: Design and Methods
(5th ed.). Thousand Oaks, CA: Sage Publications, Inc.
ICEIS 2021 - 23rd International Conference on Enterprise Information Systems
708