Social, Legal, and Technical Considerations for Machine Learning

and Artificial Intelligence Systems in Government

Richard Dreyling

, Eric Jackson

, Tanel Tammet

, Alena Labanava

and Ingrid Pappel

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.

https://orcid.org/0000-0001-7269-6236

https://orcid.org/0000-0003-3726-5814

https://orcid.org/0000-0003-4414-3874

https://orcid.org/0000-0001-6318-7704

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

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