You are on page 1of 20

Information Systems Frontiers

https://doi.org/10.1007/s10796-022-10293-2

How Do Users Feel When They Use Artificial Intelligence for Decision


Making? A Framework for Assessing Users’ Perception
Amit Kumar Kushwaha1 · Ruchika Pharswan2 · Prashant Kumar1 · Arpan Kumar Kar1 

Accepted: 13 May 2022


© The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2022

Abstract
Artificial intelligence (AI) transits from merely adopted technology to fueling everyday decision-making systems from
medication to navigation. With this combination of AI in decision-making systems (ADMS), the present study explores
how text-based users' data from social media helps organize the users' perspectives of ADMS? To investigate our research
questions, we used a framework consisting of three phases, exploratory, confirmatory, and validatory. We applied hierarchy
clustering and topic modeling in the exploratory study, hypothesis building, and empirical analysis during the confirmatory
study and support vector machine (SVM) in the validatory study. Our findings suggest that users are primarily concerned
about the risk involved in using ADMS. Factors like accountability, self-efficacy, knowledge of ADMS individuals' attitudes
towards ADMS impact the perception of ADMS among individuals. This study's theoretical and practical implications have
great scope as ADMS is still in its elementary stage.

Keywords  Artificial Intelligence · Perceived risk · Accountability · Fairness · Trust · Machine learning

1 Introduction nature never guarantees the correctness of a particular deci-


sion (Ameen et al., 2021; Bankins et al., 2022). Today, AI is
While the inception of AI technologies accelerates, its' used to assist human specialists in making decisions under
involvement and acceptance have increased substantially. AI high-stakes situations and assumed as an undeniably sig-
applications are associated with work viability, high-valued nificant part of dynamics. However, despite its advantages,
work, and firms' decision-making dynamics, promoting better AI-based automatic decision-making is not often desirable
performance (Kushwaha & Kar, 2021; Silver, 2020). While due to significant concerns like privacy, fairness, knowledge,
algorithms can perform impressively, entire delegation to ML trust in technology, and perceived values (Behera et al., 2021;
models is not desired in many situations as their probabilistic Chatterjee et al., 2020). The decision-makers or stakeholders
must realize when to trust or distrust the AI decision-making
tendency. There is a need to identify the factors influencing
Statement of contribution: All authors are equal contributors stakeholders' ability to foster AI capabilities.
Recently, a few businesses have advanced from just tak-
* Arpan Kumar Kar
arpan.kumar.kar@gmail.com ing on AI as a cutting-edge method to even more firmly
coordinating something similar to dynamic frameworks to
Amit Kumar Kushwaha
kushwaha.amitkumar@gmail.com create profits from the speculation made while building the
in-house AI practice (Kumar et al., 2021; Li & Kettinger,
Ruchika Pharswan
ruchi1966pharswan@gmail.com 2021). With AI in the Decision-Making System (ADMS),
stakeholders' trust in the choices has been seen as risky.
Prashant Kumar
prashant@dms.iitd.ac.in AI vests higher confidence in the results of ADMS; past
progress with gut-based decision making empowers more
1
Department of Management Studies, Indian Institute faith in manual decision making. Characterizing ADMS
of Technology Delhi, New Delhi, India deeply to break down the above upheavals, the literature nar-
2
Bharti School of Telecommunications Technology & rowly describes this as a "technology-driven decision with
Management, Indian Institute of Technology Delhi, no humans' involvement." This assertion further mitigates
New Delhi, India

13
Vol.:(0123456789)
Information Systems Frontiers

the convulsions. Nevertheless, expanding the definition by RQ1: How do we collect intelligence from users’ text?
expressing: a robust framework prepared on increasing vol- RQ2: What are factors extracted from collective intel-
ume, variety, veracity, and real-time data fueled by the most ligence, and how do they influence users' perspectives
recent algorithms, with the adaptability of over-judge the of ADMS?
decisions whenever required, might give some trust back
to ADMS (Kushwaha et al., 2021a; Nicolaou & McKnight, For this study, we borrow literature from social sciences
2006). and theories to understand how different factors pertinent to
ADMS is currently used in various services provided the influence the confidence in adopting ADMS. We have
to customers, for instance: overall advertising, recommen- segmented the entire study into three phases to achieve the
dations-based search, and contextual responses (Dwivedi study objectives. First is the exploratory phase; we per-
et al., 2021). Also, ADMS delivers a cultural and public formed various exploratory analyses on data extracted from
effect by energizing virtual wellbeing reaction frameworks, the social media platform to identify stakeholders' perspec-
automated decision-making frameworks to build application tives, confidence levels, and other views around the ADMS.
filling, detecting misinformation, support to e-administra- The second phase consists of statistical analysis. We col-
tion, and the medical care area is interminable (Balakrishnan lected data from stakeholders via survey, then determined
et al., 1999; Liu et al., 2021a, b). The presence of ADMS factors impacting the users' perception of ADMS using sta-
in the legal executive space likewise makes up for lost time tistical methods such as multivariate regression. The third
gradually, however continuously, with law and requirements phase is our last phase. We validate our theoretical model
(Balmer et al., 2020). Therefore, decision-support applica- and hypothesis using a support vector machine (SVM) and
tions should be developed not only for high performance, conclude the most significant factors that influence user per-
safety, and fairness but also for the decision-maker to under- spectives of ADMS (Alahakoon et al., 2020; Bankins et al.,
stand the predictions made by the model (Enholm et al., 2022).
2021; Shet et al., 2021). The rest of the article is structured as follows: Section 2
A few research likewise attempts to uncover the reper- presents the literature review on ADMS, followed by Sec-
cussion of ADMS results, alongside the risk-related spe- tion 3 indicates the research methodology and framework
cific domain and constraints (Dwivedi et al., 2021; Meske used for this study. Section 4 shows the result of this study,
& Bunde, 2022). This study additionally reveals the rate followed by Section 5 explains a discussion consisting of
of human involvement at each level when ADMS provides theoretical, managerial implications, and limitations of this
outcomes per domain, and afterward, it assesses the worth study. Lastly, the conclusion is presented in Section 6.
it can acquire. This may also facilitate building stakehold-
ers' confidence and develop some feeling of easiness toward
ADMS. Having expressed the abovementioned, much is yet 2 Literature
to be investigated as far as associating stakeholders' perspec-
tives of ADMS on the constructs of perceived usefulness, AI is no different from other technologies attempting to
risk, fairness, trust, knowledge, privacy, accountability, and develop automated systems, robust frameworks, and digi-
AI readiness. talized versions of nearly everything (Mikalef et al., 2021;
We also attempt to identify the problems in using local Pillai et al., 2020). AI has brought new characteristics like
explanations for AI-based decision-making scenarios and smartness, cognitive, and intelligence from clinical diagno-
invite the research community to explore new approaches sis to personal support and decision making within organiza-
to explainability, privacy, fairness, accountability, trust, tions (Liu et al., 2021a, b; Votto et al., 2021). The ramifica-
and knowledge for analyzing the level of stakeholder's con- tions of AI have impacted stakeholders who have created or
fidence in ADMS. To explore how stakeholders can become used its applications for economic and social benefits. Still,
more capable of deploying AI technologies and empower- assuming that AI and analytics are used ethically as ethics
ing them, they must adequately understand the deployment's pervades the whole process, from data selection to present-
main drivers to support these processes. Talking about stake- ing and exploiting information (Calzarossa et al., 2010). On
holders, social media allows users to express their opinion, top of that, algorithms running behind AI-based systems cre-
facilitates indivisible techno-social cooperative interaction, ate accountability-related issues during the decision-making
and encourages communal communication and social behav- process. Besides, AI algorithms generate early solutions;
ior (Oh et al., 2015; Tiwary et al., 2021). In this study, we they require human interpretation, reasoning, and action to
consider exploring how text-based users' data from social provide accurate, valuable outputs. So, before accessing the
media helps organize the users' perspectives of ADMS? ADMS, first, we need to consider asking what we know so
Therefore, we put forth the following two research questions: far on individual perceptions and decision-making systems.
Secondly, how ADMS differs from any previous techniques

13
Information Systems Frontiers

employed for supporting decision-making. In the subsequent think that, while the first two stages are mechanical in terms
section, we argue these questions and map literature with our of data, the second half is self-learning from the feedback
research questions. loop where the socio-technical application of the outputs
is crucial. While the model approach is stored as a set of
2.1 User Perception of Other Decision Support mathematical principles that lead to the formulation of a
Systems mathematical equation, it continually changes the weights
in these equations, which must be driven by socio-technical
Research suggests that personal decisions making character- implications for ADMS to be successful (Baabdullah et al.,
istics influence perceptions about outcomes and a person's 2021; Silver, 2020). Overall, ADMS is made up of three
value of a specific result. For example, system usage, deci- components: data (which feeds the model), the model's
sion performance, and user satisfaction (Riedel et al., 2012; algorithm (which is at the heart of ADMS), and the con-
Yuthas & Young, 1998). Compatibility may never emerge sequences of the outputs generated (self-learning). ADMS
as a determining factor in the decision-making of less inven- can now be classified into three types: high manual interven-
tive, those who do not believe in the value and complexity of tion at the start of data collection and decision-making at
the invention. Adoption of IT for expert system applications the end of output generation, medium manual intervention
creates awareness and information accessibility through dif- required to ensure ADMS does not produce extreme outputs
ferent channels. Media as determinants of user perception that disrupt socioeconomic or law and order in any way, and
while personal innovativeness as a moderator between user no or minimal manual intervention required at any stage
perception and decision-making changes based on user (Alsheibani et al., 2018; Shet et al., 2021).
belief of the benefits, they gain from using the decision- The level of human connection or engagement and the
making system (Agarwal & Prasad, 1998) like managerial, domain influence the "gear and hype cycle" stage of the cur-
operation, and IT infrastructure benefits significantly change rent ADMS for any field. For example, the healthcare appli-
user perception of ERP system implementation (Spathis & cation of ADMS, which involves no human intervention in
Ananiadis, 2005). necessary procedures, may put stakeholders in a dread cycle.
Sometimes, a lack of understanding about the decision Highly tailored adverts, suggestions, and search results, on
support system makes users unwilling to take risks with the other hand, may cause the same ADMS to enter a hype
insight that they gain from the system (Foshay et al., 2007; cycle (Ridhwan & Hargreaves, 2021). As a result, gov-
Gupta et al., 2022). Like business intelligence gained by ernmental and law enforcement uses of AMDS, including
transforming raw data to provide decision support, data ana- minimal human participation, may place stakeholders in
lytics affects the user perception, transforming further into the center of the awe and hype cycles. Knowing the present
satisfaction and user retention (Wang, 2016). The organiza- state, drivers of the fear and hype cycles is critical for any
tion interface, where trade-off qualities categorize outcomes, technology to succeed. Based on the above discussion, we
builds trustworthiness more than the traditional recom- analyzed that people continuously take about the ADMS
mendation methods. In addition, forecasting is critical for on different channels, but the study to access those talks
organizational decision-making (Önkal et al., 2008; Wang is minimal. In essence, organizations cautiously adopt the
et al., 2021). So, user perception is more inclined to return AI system for their decision-making, but their performance
to the interface and predictive format; it can help them make assessment is less, leaving a gap or need for the study. Based
choices contingent on the forecast. Also, perceptions of ease on the gap, we have formulated our first aim to access the
of use and usefulness of self-service analytics are related to collective intelligence from the user surrounding the ser-
task-technology fit, compatibility, and user empowerment vice provided by ADMS. In the next section, we will look
(Pu & Chen, 2007). for evidence of the theory's support in developing the same
viewpoint in the literature.
2.2 AI in Decision Making
2.3 Personal Characteristics that Influence
In this section, we discuss what we know about ADMS the Creation of an ADMS Viewpoint
and how this is different from other systems? ADMS are
automated frameworks that 1) gather data, 2) de-noise it for We employ some of the unique pivots of literature compo-
usage, 3) organize it for a model, 4) pick the best model, 5) nents to go deeper into the causes for shaping the mentality.
construct the outputs, 6) penalize if the result is incorrect, The research shows that the first aspect is "understanding"
6) reward if it is correct, and 7) learn from the mistakes to the method and domain. Suppose a government official has
improve the next iteration of results. In the last few feedback specialized domain expertise and some understanding of
rounds, the whole ADMS are understood and improvised how the ADMS self-trains; they should be better positioned
(Kushwaha et al., 2021b). As a result, the ADMS creators to make educated judgments on ADM's outcomes in any

13
Information Systems Frontiers

crisis. A doctor must know about the medical domain, simi- like perceived simplicity of use, perceived benefit, minimal
lar to how an ADMS anticipating Diabetic Millitus might danger, and confidence in the technology, might increase
utilize the assessment of margin of error they can deal with users' willingness to perceive any technology {Formatting
before prescribing treatment for the patient (Li et al., 2020). Citation}.
Similarly, stakeholders' perspectives on the algorithms'
heuristic nature encourage perspective building. This is
partly influenced by the individual aspect of safety regarding 3 Research Methodology
the quality of the output (Aribarg & Schwartz, 2020). Aside
from comprehension and quality, another vital element that The first objective aims to gather the intelligence from the
emerges from the literature analysis is demography, namely users' discussion about ADMS usage experience surround-
age. Age, as a factor, has played a significant part in deter- ing decision-making services. So, to extract the user discus-
mining the creation of perspective toward new technology sion, we have used social media data since it gives a high
acceptability. Therefore, the AMDS cannot be immune to volume of user discussion on each specific subject. Con-
the effects of the same. Why do humans accept or reject the sidering that our first and second objectives are to qualita-
use of artificial intelligence in decision-making? A previous tively analyze the users' conversations and link them with
study has found that when humans utilize AI as a decision- the factors influencing user perception of ADMS, followed
making tool, they may adopt various attitudes and behaviors by the third objective of establishing the relationship of
to carry out the AI system's recommendations (Luo et al., extracted factors with user perception, we have observed
2019). People's attitudes towards it will influence the recep- a mixed method-based research design (Venkatesh et al.,
tion and impression of ADMS. 2013). Following the primary design strategies suggested by
If consumers' understanding of how they utilize tech- Venkatesh et al. (2016), we formulate the methodology into
nology is jeopardized, their fears or anxieties may prevent three phases (Fig. 1.). We discussed each step separately:
them from using and embracing any comparable technol-
ogy. The ADMS user's risk in purchasing the technology 3.1 Phase 1. Exploratory Phase
may impact the user's perspective, either directly or indi-
rectly. Because ADM combines fully or partially automated In the first phase, we elaborate exploratory stage, where we
decision-making processes in a wide range of settings, there extract perspective around ADMS acknowledgment and
have been serious concerns that automatic choices like those adaption from social media platforms (SMP). An explora-
from recommender engines may be skewed (Sharma et al., tory study does not give a robust solution yet provides the
2021). Several previous research has found unacceptability underlying foundation to understand the problem to some
and unhappiness with the employment of AI algorithms for better extent. We employ exploratory analysis to identify
decision-making, making the fairness component of ADMS the potential constructs. User-generated content (UGC) sur-
even more critical to explore (Subramaniam et al., 2000). rounding a product or service is typically collected in the
AI approaches include some opacity, which causes fear of voluble interaction on SMPs like Twitter. Hence mining the
failure and a sense of bad decision-making on the part of data from Twitter helps us analyze the opinions and under-
users. When users cannot evaluate outcomes, the issue of stand the signals that could lead to theoretical constructs that
responsibility becomes more serious. It is critical to consider can be tested later. Its large user base and active interaction
the accountability component of ADMS to make it a depend- within and between the groups help us build a proper context
able source and acquire users' trust. for the study undertaken in the current analysis (Ridhwan
The danger of utilizing the ADMS is linked to trust, either & Hargreaves, 2021). With UGC from Twitter as an input,
directly or indirectly. The acceptance of any technology we employed content analysis of Social Media Analytics
requires trust, according to various studies. The higher the (SMA).
level of faith, the more positive the user's impressions of the
technology are (Kumar et al., 2021). As a result, this sense 3.1.1 Data Collection and Processing
of trust can significantly impact ADMS users' perceptions.
Because ideas like ADMS combine cutting-edge technology The data collection framework consists of two modules. In
like AI, users must at least be conversant with these tech- the first module, we collect the data related to ADMS from
nologies' technicalities to use them (Elbanna et al., 2020; SMP. In the second module, we collect the data using a ques-
Wendt et al., 2021). Although ADMS can be a sound system, tionnaire-based survey (detailed in Sect. 3.3) which we pre-
there is overwhelming evidence of its failure to gain accept- pare using the data from literature and exploratory studies.
ance and adoption owing to its complex design. Research The first data extraction module is built on the python toolkit
suggests that user willingness plays a critical part in form- utilizing the Twitter streaming API with the results stored in
ing a positive picture of technology. Several characteristics, a database. The hashtags and keywords that were utilized to

13
Information Systems Frontiers

Fig. 1  The framework used to


analyze research questions

gather data are as follows: ''#ArtificialIntelligence,’ ''#AI,’ and groupings of the themes arising from the content, we
''#DesicionMaking,’ ''#Trust,’ ''#ADMS,’ ''#Applications,’ use hierarchical clustering, topic modeling, sentiment analy-
''#AutomedDecisionMaking,’,’ ''Automatic''. We downloaded sis, and summarize the material. By applying hierarchical
over one million tweets from January 2021 to May 2021. crusting, we extract a cluster of words that could be in the
We found many noises (e.g., non-English, hyperlinks) and same category without any need for a predefined cluster.
uncertainty during the initial screening of collected data, However, the problem lies in the word cluster, which some-
which we removed by preprocessing (Kar, 2020). times does not make sense. So, for a more refined theme
We discovered that few users utilized hashtags midway selection, we employed a Latent Dirichlet Allocation (LDA)
through sentences, which totaled sense to the sentences. (Kushwaha et al., 2021b) established topic modeling tech-
Thus, we have preserved the hashtags' word terms and nique to identify users' unspoken themes or subjects in their
deleted the "#" character. Furthermore, a few tweets had a tweets regarding ADMS.
lower word estimate and a higher quantity of hashtags in the LDA is also a generative statistical method consisting of
content, and the words did not offer the crucial information. a bag of words generated by choosing a topic mixture from
Therefore, we have eliminated them. We employed Natu- the text sample. However, the mix of words selected from
ral Language Processing (NLP) which utilizes segmenta- topics does not consider the order in the sentence. LDA-
tion, tokenization, voice tagging, stop word detection and based topic modeling is a machine learning (ML) approach
exile, reliance parsing, and named entity identification. It that is frequently utilized by academics for marketing, client
also includes removing duplicate texts, denoising, and spell- satisfaction, and segmentation purposes. The LDA model
checks. Lastly, text exploration techniques (EDA) through has been used to derive market segmentation in various
topic modeling help in this scenario, so; we use several studies. Customers' feedback has been collected using the
topic-modeling methods to understand the crucial topics LDA model by others. They also highlighted that collecting
from the tweet texts. We analyzed the whole process using information from reviews using an LDA-based model is a
the NLTK and genism packages, which provide a variety of different but effective way to know the client better (Shet
characteristics, including updated stop words. et al., 2021). So, the LDA provides results in the form of the
topic with words associated with them. Each of these topics
3.1.2 Content Analysis represents a separate meaning which could be mapped with
the help of literature. Lastly, in the first phase, we assess
Researchers have proposed different data analytics ways, consumer sentiment and prime; intention regarding tech-
such as hashtag, word, and sentiment analysis and word nology, goods, brands, or services; sentiment analysis is
association, and inferential model validation employing extensively used in technology adoption, decision-making,
topic modeling, topic association, and polarity analysis for and customer-related literature (Kar & Kushwaha, 2021).
theory building computationally (Kar & Dwivedi, 2020; Research has treated sentiment as a proxy of various firm
Ridhwan & Hargreaves, 2021). To explore the associations performance parameters.

13
Information Systems Frontiers

3.2 Phase 2: Confirmatory Phase this will help us determine the context in which the users are
talking. The number of topics discovered applying the LDA
Secondly, we apply the confirmatory analysis, which estab- technique for multiple regression analysis is significant, and
lishes the relationship between factors and users' percep- not all topics significantly influence the dependent variable.
tion of ADMS. Besides, corroborative studies are used when This is a common challenge with extensive data, where pre-
educators solve issues and test hypotheses to develop ideas dicted multi-scale problems occur. Therefore, we went back
founded on facts. So, the topics from the topic modeling to the literature to see if there was any connection between
of exploratory analysis need to be mapped with its close the selected variable and how the end-user may perceive it.
related variable (Büschken & Allenby, 2016). For this pur- Usually, hypotheses are formulated before any study design
pose, studies have suggested two methods. First, incorporate in the early stages of building the framework. However, in
intercoder reliability, where theme to variable mapping is social media research, the context in which users are talking
done by one author, verified by others, and tested reliability. is not fixed. Data varsity is very high, so extracting the infor-
In the second method, a survey is designed based on the mation text content to set the variable is essential for this
topic and its associated word by mapping with its closest research design. This phase's last step ends with establishing
proximity variables and selecting based on the data theme. the hypothesis related to user perception. Following the sub-
In this study, we employed the second method. sequent section, we present the validation of the hypothesis.
We have collected data via surveys during June 2021.
Profile selection and shortlisting through professional net- 3.3 Phase 3: Validatory Analysis
working were used to choose participants in this research.
Professional SMPs such as LinkedIn were also used to select On the other hand, multiple regression analysis is typically
professionals who were asked to participate in the poll. Lis- appropriate and free of multicollinearity effects for empiri-
teners of AI podcast platforms such as "AI Galore," which cal research in content analysis methodologies (Kushwaha
provides technical mentoring and direction to stakeholders et al., 2021b). In our instance, though, a few subjects had
interested in pursuing or establishing AI as a specialized little influence. A linear regression study investigates the
area, were another significant source used for the study. link between the predictor factors and the predicted vari-
Sending one email every day also provides mentorship to able. The purpose is to reduce the loss function, such as
AI system users. The listeners' samples were randomly the sum of square residual, to the smallest possible value.
selected, and survey questionnaires were given out as part When the connection between predictors and predicted vari-
of the regular mailing. Participants were informed about ables is non-linear, the non-linear technique outperforms the
data utilization for academic research in AI. Following the linear method on more extensive datasets. When multivari-
reported constraint, the participants are asked questions to ate linear regression algorithms were developed, only valid
assess their degree of expertise in mathematics, AI, domain, components in every reliant variable were deemed momen-
social implications, and programming language knowledge. tous. Because the current study is based on ADMS-based
Following that, the participants were given reading material decision-making, we implemented a system to rigorously
on ADMS and how the results' application might be inves- exclude any replies that blared doubtfully or did not connect
tigated in other fields. After then, these professionals were to decision-making (Shet et al., 2021).
questioned about their thoughts on ADMS. Lastly, we conducted an SVM employed corroborative
Following their responses to the first questions, the par- analysis based on our grasping from the preceding and
ticipants were randomly allocated to one of four scenarios: exploratory investigations. In the confirmatory phase, we use
corporate use of ADMS, governmental or society use of the permutation testing approach to identify the significant
ADMS, legal and enforcement use of ADMS, and demand- constructs that impact the user's ADMS application experi-
ing health care use of ADMS. These situations have impli- ence, for which we use an SVM classifier. As a result of
cations for business, organization, law, enforcement, and the permutation testing approach, we now have a set of sig-
health. The approaches used by the participants were based nificant factors, and now we validate the same by using the
on three levels of vignette-experiment ideology: ADMS ver- stepwise regression analysis. Since we have high dimension-
sus human decision X self vs. others adopting the product X alities in our data and many potential constructs, prune the
high vs. low impact with the domains indicated. The partici- list. We conducted the iterative regression analysis and the
pants said the ADMS judgments were solely based on data SVM to scrutinize all the possible sets of most significant
and algorithms, with no human input. All the participants factors that impact the stakeholder's ADMS usage experi-
were given checkpoint questions to cross-verify the inputs ence (Balmer et al., 2020). To validate the proposed hypoth-
entered via them, the outcomes, and their robustness. esis in the current studies, we have analyzed the potential
So far, we have used the qualitative research approach to relationships among the elements and their inferential influ-
extract the variables from the text information of users' data; ence on the ADMS usage experience (dependent variable).

13
Information Systems Frontiers

To obtain the most significant set of constructs, we have or cluster of similar terms. The shorter the branch, the more
used the permutation testing approach. In this technique, similar the words will be. For instance, terms like "leak" and
random labels are assigned to the constructs, and then we "identity theft" (in red color) are closely connected in the
use an SVM classifier to train the model. After that, we test nearest branch, which gives a similar sense, such as secu-
the significance of each construct by using the "one left out rity-related concerns. Likewise, the terms "responsible" and
"cross-validation strategy. We used this procedure in itera- "authorization" (in green color) are closely connected in the
tion until we separated the most significant constructs. nearest branch, which gives a similar sense such as account-
ability concern. Another example is the terms "women" and
"race" closely placed in the nearest branch, which offers a
4 Findings sense of bias or targets the fairness aspect of ADMS.
Moreover, there are some constraints to using hierarchical
4.1 Insights from the Exploratory Phase clustering analysis. For example, hidden structures are not
promptly considered. We have utilized the topic modeling
4.1.1 Insights from Exploratory Analysis approach for extracting the themes. To derive the meaning
from the text data, topic modeling utilizes natural language
For the phase one study, data was collected from Twitter to processing techniques. For our analysis, we used the LDA
gain insights into the users' discussion surrounding ADMS. method to extract more refined topics and associated words
The keywords that repeatedly appeared in the conversations. (see Appendix 1 for the topic and words associated with
Analyzing these keywords closely, we learned that these key- them). Further, following the method suggested by Büschken
words illustrate sentiments of users' towards ADMS. These and Allenby (2016), we mapped each topic with their close
sentiments show various users' positive and negative per- related constructs. The topic to construct assignment needs
ceptions of the ADMS. For example, keywords like "leak," to check for reliability (Berente et al., 2019), so we verified
"identity theft," and "cyber" indicate the security concerns the mapping using a cross sectional survey to establish relia-
of users. Moreover, keywords like "prejudicial," "bureau- bility and validity of the cluster of words with the constructs
cracy," and "race" show the users' concern about the fairness which will subsequently be used for natural language pro-
of the decision taken by ADMS. Whereas keywords like cessing based model validation. We presented the question-
"gender," "women," and "Asians" depict the demographic naire used for the survey in Appendix 2, in which the first
elements. "Responsible," "principle," and "authorizations" item for each variable is assigned based on the results found
these keywords show the accountability concerns among from topic modeling. The result of the survey indicates reli-
stakeholders. ability of 0.89.The objective of the survey is to establish
Likewise, keywords like "skill," "interview," "capabil- inter-coder reliability and face-validity to the next steps of
ity," and "expert" show some knowledge concerns of ADMS natural language processing for theory building using big
users. Terms like "loss," "belief," and "misunderstood" show data analytics (Berente et al., 2019; Kar & Dwivedi, 2020).
some trust-related concerns of ADMS users. These key- After determining themes, we have mapped these themes to
words result from users' positive or negative perceptions of our exploratory analysis and clustering results to assess the
ADMS. Keywords like fear, loss, breach, and misunderstood constructs from the prior literature and test the correspond-
show some users' negative sentiments towards the ADMS; ing hypothesis.
however, expert, power, and growth kind of keywords show
some positive feelings of users. These keywords will be used 4.2 Insight from Confirmatory Analysis
as input for developing a theoretical framework constructed
for analyzing users' perception of ADMS and hypothesis From the signal captured in phase one and analyzing
development that will be tested in later phases. those using social media Analytics, we tracked down
some emerging keywords from the conversation around
4.1.2 Insights from Hierarchical Clustering the ADMS from various stakeholders' lenses. Contem-
plating these keywords, first, we head towards hypoth-
We have utilized the agglomerative hierarchy cluster- esis development. In this study dependent variable is the
ing approach to determine the latent structure within the stakeholder's ADMS perspective that directly and indi-
extracted data. The output of the hierarchical clustering is rectly depends upon individuals' ADMS usage experi-
represented in Fig. 2. In this approach, stakeholders use key- ence. Thus, in this research study, we have considered
words in general discussion over social media to make sense stakeholders' ADMS perspectives and emotions as a
of it. Terms that have a similar meaning or make sense are proxy for further ADMS acceptance. We have developed
hinged on the nearest branch. This process of placing terms our hypotheses as follows:
of similar meaning near each other sooner will form a group

13
Information Systems Frontiers

Fig. 2  The outcome of hierarchical clustering

4.2.1 Privacy/Security 4.2.2 Perceived Risk

Security becomes a primary concern when using ADMS If technology is incapable of delivering satisfactory out-
or any technology dealing with user data extensively. Also, comes, factors like risk and uncertainty might be incited,
with technology like AI, where the atomic work unit is and these factors influence the user's confidence in utilizing
data, privacy or security must especially be taken care of that technology. If there is a risk to users' comprehension
for a better user experience (Loh et al., 2021). ADMS must of their utilization of the technology, their concerns or fear
include sufficient protection to be a dependable asset as might keep them from using and accepting any comparative
data security, privacy and risk becomes critical concerns in technology (Jöhnk et al., 2021). The risk experienced by the
digital platforms (Kar & Aswani, 2021). Since any frame- ADMS user in acquiring the technology may directly or indi-
work involving AI captures enormous amounts of data, it rectly influence its user's perspective (Balakrishnan et al.,
is evident that users are uncertain regarding security as the 2021). Also, ADMS needs to ensure a deep level of integrity
vast majority of the information comprises the user's details and reliability to utilize the application and subsequently
(Chatterjee & Kar, 2018). The higher the technology ensures accept it effortlessly. Reducing the risk experienced by the
security and privacy, the more the user utilizes it and has an user when using ADMS ingrains high trust and certainty
optimistic attitude (Chatterjee et al., 2021a). We broaden this to accept it (Im et al., 2008). Broadening the argument that
contention and propose the hypothesis as follows: perceived risk may unfavorably influence the user's intention
to utilize the ADMS, we present our hypothesis as follows:
H1: There is a positive relationship between privacy/
security and users' perspective of ADMS. H2: There is a negative relationship between perceived
risk and users' perspective of ADMS.

13
Information Systems Frontiers

4.2.3 Fairness technologies like AI, they will initially have an unreliable


outlook on the results of its utilization. Users would ana-
Since ADM incorporates complete or semi-automated lyze how modern systems like ADMS would be dependable
decision-making processes in various environments, criti- and trustworthy (Gide, 1967). Trust is directly or indirectly
cal concerns have been raised that automated decisions may associated with the risk arising from using the ADMS. More
be biased (Baur, 2017). Several prior studies indicate unac- confidence naturally improves users' perception of ADMS
ceptability and dissatisfaction regarding the use of AI algo- and its recommendations. ADMS inclines toward the most
rithms for decision-making, making it even more essential recent technologies; lower transparency and understandabil-
to investigate the fairness aspect of ADMS. While utilizing ity lead to an impression of vulnerability among the ADMS
ADMS, standards of fairness should be preserved (Gupta users (Nicolaou & McKnight, 2006). Various literature dem-
et al., 2021a). By developing an ADMS that assists users in onstrates that trust is fundamental for the acceptance of any
their decision-making process, the fair use of data assures technology. The higher the confidence, the better the users'
that it is adequate for the purpose and respects the principle perception of the technology. Therefore, this sense of trust
of individual autonomy. If ADMS users feel that the deci- can highly influence the perception of ADMS users (Araujo
sions are fair, reasonable, and dependable, they are more et al., 2020; Chatterjee et al., 2020). So, the hypothesis sur-
likely to accept the ADMS results. Therefore, it becomes rounding the perceived trust is as per the following:
crucial to understand how AMDS influences user perception
regarding fairness (Wang et al., 2021). Here we propose our H5: Trust in ADMS positively impacts an individual's
following hypothesis around fairness: perspective of ADMS.

H3: Fairness is positively associated with users' perspec- 4.2.6 Attitude


tives of ADMS.
Mapping with the three main determinants out of five main
4.2.4 Accountability determinants mentioned by Robey (1979) regarding volun-
tary system use is user attitude and perceptions, technical
Another primary concern when using an automated system quality of the system, and situational and personal factors.
like ADMS is how the machine arrives at the particular deci- Based on his analogy, we relate user attitude with individual
sion and who is liable for the wrong decision. Unlike conven- attitude towards AI technology, technical quality of system
tional technologies, AI model algorithms used for decision- with technical complexity as AI system relying heavily
making are highly complicated and opaque in most cases on backend algorithms and situation and personal factor-
(Mikalef & Gupta, 2021; Trocin et al., 2021). Additionally, like user awareness with AI readiness. Various literature
advanced technology like AI operates on users' data to provide on stakeholders' attitudes implies that each individual has
better customization and personalization services. Few prin- a positive or negative attitude towards technology (Baur,
ciples and methods enable users to comprehend these models' 2017). Prior research also highlighted that technologies
essential workings; some of them are incomprehensible even concerned with internet usage and users' attitude play a
to developers. This opacity of AI methods sometimes creates crucial role in its reception (Nicolaou & McKnight, 2006).
fear of failure and perception of users' poor decision-making Users foster a particular attitude towards technology due to
(El-Haddadeh et al., 2021). When results are not interpretable many reasons. For instance, users will develop an affirma-
by users, and accountability aspect runs deeper. Considering tive attitude towards its reception if ADMS is easy to utilize
the accountability aspect of ADMS becomes crucial to make and understand, ensures high privacy, involves low risks, is
ADMS a reliable source and gain users' vote of confidence. dependable, and provides appropriate solutions (Chatterjee
More, the technology is accountable for its deliverables; it et al., 2021a, b). Users' attitudes about utilizing any particu-
seems more trustworthy to users and can positively influence lar technology, application, product, or brand are an equally
the ADMS perception of the user (Gupta et al., 2021b). significant factor in deciding if an

H4: Accountability will have a driving relationship with H6: Individual's attitude will have a driving relationship
an individual's perspective of ADMS. with an individual's perspective of ADMS.

4.2.5 Trust 4.2.7 Technological Complexity

The use of AI-based decision-making is implicated with the There is supporting research indicating that new technolo-
help of the internet. When users are acquainted with a new gies effortlessly interlaced with the existing ones have more
technology or framework that uses the internet and advanced prospects of success in acceptance and adoption (Kai-ming

13
Information Systems Frontiers

au & Enderwick, 2000). However, advanced technologies penalty and reward of the results turned out to be the central
are personalized to serve specific domains (Wendt et al., theme of ADMS (Chatterjee & Kar, 2018). Users of ADMS,
2021). Complexity varies depending upon the environment the data creators, unknowingly become its stakeholders and
they are applied in, making it challenging to gel up with might share more of their data without acknowledging it
current frameworks, jitters users' comfort, and minimizes (Ruggieri et al., 2021). If humans handle the data collec-
engagement, hence fostering a dismissive perception towards tion and usage, they might decide to overlook these to keep
that technology. Since concepts like ADMS incorporate the an individual's privacy by interfering with the processing
latest technology like AI, users need to become familiar pipeline. However, with no human intrusion, the information
with the technicality of these technologies to some extent, construction of ADMS can be sufficiently wise to isolate
basically to utilize them. Even though ADMS can be a pro- individual data from actual information for preparing AI
ductive system, there is compelling evidence of the failure models. Prior research has shown how an individual's view
of its acceptance and reception due to its intricate nature of the unique ability to secure personal data (self-efficacy
(Awa et al., 2017). Consequently, while examining the users' in online mode) can influence their perception of ADMS
perception of ADMS, technical complexity turns out to be (Wang et  al., 2021), especially AI in that structure. We
relevant to be hypothesized and tested. Here, we propose our develop another hypothesis underneath:
following hypothesis:
H9: The user's self-efficacy will have a driving relation-
H7: Technical complexity of ADMS influences an indi- ship with an individual's perspective of ADMS.
vidual's perspective of ADMS.
4.2.10 Demographic
4.2.8 AI Readiness
Prior literature on IT and technology adoption demonstrates
Various research demonstrates that users' behavior and will- that the demographics of stakeholders who contribute, own,
ingness to utilize any technology can be positive or negative, utilize, or consume, play an essential role in acquiring new
affecting the reception of technology. Individuals' enthusi- technology (Wendt et al., 2021). We suggest that reception
asm toward a new technology influences their readiness to of any technology fosters a vote of confidence when it is
perceive the technology (Wijnhoven, 2021). Some literature utilized over time. Research has featured that the reception
indicates the vital role user's willingness plays in building an of the technology is driven by users' perceived benefits, per-
impression of the technology. User readiness for perceiving ceived usability, and fairness. Some literature suggests that
any technology can be expanded by perceived ease of use, factors, for example, age and gender that come under social
value, low risk, and confidence in the technology (Manfreda, acceptability likewise impact the adoption of technology
2020). However, considering the organizational scenario, AI (Sun & Jeyaraj, 2013). Although gender does not appear to
readiness could be associated with AI awareness, technical- be more important in some research, age comes out to be a
ity, and AI infrastructure (Baabdullah et al., 2021). Even significant factor in the adoption literature (Chatterjee et al.,
if organizations can procure infrastructure and technicality, 2021a, b). Subsequently, we hypothesize that as technology
user awareness remains another aspect. So, the willingness is being adopted (in our study in the context of ADMS),
of individuals to participate in the organization development the confidence in adoption gradually increases with time,
process using a new decision support system needs to be wherein demographic factors influence users' perception of
considered (Alsheibani et al., 2018). Since ADMS com- ADMS for adoption. Along with the same lines, we propose
prises AI, users must initially be familiar with AI and accept our following hypothesis as follows:
it as an asset. Users' AI readiness is undoubtedly an essential
factor for identifying whether an individual is inclined to H10: Demographic factors will have a driving relation-
accept and use ADMS technology and what perception it ship with an individual's perspective of ADMS.
upholds towards it (Hong et al., 2013). With this, we arrive
at our following hypothesis: 4.2.11 Knowledge of ADMS

H8: AI readiness of users positively influences an indi- Previous research has demonstrated that understanding is a
vidual's perspective of ADMS. fundamental factor in how stakeholders can comprehend the
value of new technology. The individual's knowledge some-
4.2.9 Self‑Efficacy how influences how they perceive that technology, and it can
vary from user to user. Systems based on AI are generally
Designed on the entire structure of extensive automated data digital, web, or mobile-based services, which require some
collection, fabricating model and self-learning loop from the degree of knowledge and experience (Li & Kettinger, 2021).

13
Information Systems Frontiers

Subsequently, borrowing from prior research, it becomes better insights around ADMS, we propose the following
crucial to count on understanding as one of the significant hypothesis:
attributes in framing stakeholders' perspectives. However,
the understanding of individuals may vary and have varied H11: Technical and domain knowledge will have a
implications. Some earlier literature features an understand- driving relationship with an individual's perspective of
ing of mathematics as critical for developing a positive atti- ADMS.
tude toward AI algorithms.
In contrast, many propose domain understanding is Now try to determine the factors influencing the users'
essential to recognize the significance and Trust (Wijn- perspective regarding ADMS and correlate them to the theo-
hoven, 2021) in AI powering ADMS. Nevertheless, some retical constructs identified from apprehending literature and
literature also implies that the initial level of education for analyzing SMPs data in the phase represented in Fig. 3.
end-use ADMS and the stakeholders are somehow associ-
ated with the understanding of AI. Considering the prior 4.3 Empirical Validation (Insights from Phase 3)
literature, a good knowledge of mathematics and coding is
required when developing a perspective on algorithmic fair- For empirical validation, we performed statistical analysis to
ness while making decisions (Lin & Lee, 2005). So, So, determine the factors impacting users' perception of ADMS.
given the mixed signals from previous literature to fabricate Firstly, we estimate the convergent validity by calculating
the average variance extracts (AVE) needs to be greater than
0.5. Table 1 indicates the mean, standard deviation, and cor-
relation between each construct. The diagonal element (in
bold) of Table1 presents the AVE value for constructs, indi-
cating all values greater than 0.5.
Secondly, we utilized multivariate linear regression to
assess the users’ perception of ADMS, and the regression
output is presented in Table 2. These factors are later used
to validate the theoretical framework, and corresponding
hypotheses are tested. From Table 2, we can analyze that
factors like privacy/security, AI readiness, and technologi-
cal complexity were insignificant as their p-value is more
than 0.05. We have identified 11 constructs and formulated
the corresponding hypothesis, out of which three were
significant.
Besides, perceived risk has significant negative influences
on users' perception with a lesser p-value, representing that
Fig. 3  The theoretical framework of factors influencing the users' per-
ception of ADMS
the higher the perceived risk decreases the users' perception
of ADMS. Similarly, except for fairness, all other variables

Table 1  Descriptive statistics and correlation among constructs

Variables Mean Stan. Dev 1 2 3 4 5 6 7 8 9 10 11

Privacy/Security (PRIV) 3.66 1.15 0.89


Perceived Risk (PER) 3.12 1.03 0.32 0.86
Fairness (FAIR) 3.54 1.04 0.29 0.21 0.87
Accountability (ACC) 3.87 1.08 0.44 0.48 0.35 0.88
Trust (TRT) 3.21 1.16 0.29 0.63 0.55 0.48 0.86
Technological complexity (TCO) 3.65 1.09 0.26 0.33 0.56 0.37 0.46 0.85
Attitude (ATT) 3.74 1.12 0.28 0.48 0.64 0.35 0.51 0.51 0.82
AI readiness (AIR) 3.24 1.08 0.27 0.57 0.36 0.51 0.65 0.48 0.25 0.87
Self-efficacy (SEF) 3.11 1.04 0.37 0.47 0.54 0.45 0.35 0.51 0.45 0.36 0.89
Demographic (DEM) 3.18 1.15 0.49 0.28 0.22 0.52 0.46 0.33 0.65 0.74 0.54 0.89
Knowledge of ADMS (KADMS) 3.40 1.16 0.65 0.61 0.52 0.61 0.45 0.25 0.36 0.48 0.46 0.74 0.85

13
Information Systems Frontiers

Table 2  Output of inferential analysis a stepwise regression analysis to determine significant fac-


Variables Estimate Std. error p-value Result
tors out of all the factors identified in earlier phases. The
data collected had high dimensionalities and had many con-
Privacy/Security 0.0007 0.002 0.662 Insignificant structs that required pruning of data. So, to evaluate all the
Perceived risk -0.1487 0.023 0 Significant possible sets of most impacting factors, we conducted an
Fairness 0.0059 0.004 0.007 Significant iterative regression analysis and the SVM. Here, the permu-
Accountability -0.2795 0.024 0 Significant tation testing approach identifies the most valuable collec-
Trust -0.0564 0.035 0.036 Significant tion of constructs. Wherein labels are randomly assigned to
Attitude -6.6991 1.347 0 Significant the constructs, and the SVM classifier is used. After that, the
Technological complex- -0.0784 0.062 0.208 Insignificant "one left out " cross-validation strategy is used to determine
ity
the relevance of every construct in the building perspective
AI readiness 0.0004 0.002 0.795 Insignificant
of stakeholders regarding ADMS. The entire procedure is
User-specific variables
repeated until the most significant constructs are drawn out.
Self-efficacy -0.0367 0.006 0 Significant
The results of the permutation testing technique are repre-
Demographics 0.0737 0.035 0.036 Significant
sented through a box and whisker plot in Fig. 5.
Knowledge of ADMS -0.0552 0.01 0 Significant
Constant 0.1709 0.001 0
R-square 0.17
5 Discussion
Adjusted R-square 0.167

In this paper, we started by exploring how text-based users'


data from social media helps organize the users' perspec-
tives of ADMS? Now. This high-level question consists of
two objectives: first, finding the method to analyze the text-
based social media data to extract collective intelligence
from users or stakeholders of the understudied domain; and
second, finding the users' perspective on ADMS. The cur-
rent study explores two research questions, starting with
determining the most significant factors impacting users'
perspectives towards ADMS by exploring prior literature
and analyzing stakeholders' perceptions from real-time
data. We developed a framework of multiple analyses in
different phases to investigate these questions. Firstly, we
explained the exploratory phase, wherein we collected SMP
data and analyzed it using various SMA techniques to get
a hunch about the users' sentiments towards ADMS. So,
Fig. 4  A finalized framework of factors influencing the users' percep- using the insights from exploratory study and literature,
tion of ADMS we uncovered discussion surrounding ADMS and variables
associated with users' perception of ADMS. Secondly, we
collected data via surveys and conducted inferential statisti-
negatively influence user perception of ADMS. So, users' cal analysis to get a little clear view of users' perspectives
understanding of AI fairness influences the perception of its regarding ADMS and developed a theoretical framework
implementation in decision-making systems. On the other consisting of 11 constructs and formulated their corre-
hand, under user-specific variables, except demographic, sponding hypotheses.
both self-efficacy and knowledge of ADMS have a negative Moreover, the confirmatory phase helped us analyze
and significant relationship with users' perception of ADMS. our second and third aim of exploring the nature between
Further, we present the final developed framework in Fig. 4. extracted variables and users' perception of ADMS. Lastly,
we tested our hypothesis using linear regression and robust-
4.3.1 Robustness Checks ness using stepwise linear regression and the SVM tech-
nique. Our study verifies existing research while uncovering
We tested our formulated hypothesis in the third phase and some additional findings.
validated the constructed theoretical framework. We perform

13
Information Systems Frontiers

Fig. 5  Constructs' significance
testing using SVM

We first found that fairness directly influences users' per- 5.1 Theoretical Implications
ception of ADMS, even the regulatory and legislative bodies
are increasing the demand for fairness and transparency in The current study has two folds theoretical implications.
the algorithm (Dwivedi et al., 2021). So, due to the heavy First, this study makes a theoretical contribution to the nas-
reliance on the black-box algorithm, the system's fairness cent initiative of ADMS, exploring the factors influencing
and accountability of the overall design also come into the the perception and reception of ADMS. We have also aug-
picture, which could be enhanced by providing human-inter- mented the theoretical understanding of the emerging factors
pretative and meaningful arguments (Kim et al., 2020). Con- like fairness, accountability, and perceived risk in the recep-
sequently, our result demonstrates the influence of account- tion of ADMS. The current study determined the factors
ability on users' perception of ADMS. Father, perceived building and breaking users' perception of ADMS. Much
risk and trust hurt the user's perception of ADMS, which existing literature on technology adoption highlighted that
supports the literature (Vimalkumar et al., 2021), indicating factors like risk, security, expertise, trust, and demograph-
the user who considers risk and skepticism of accepting AI ics are important factors influencing the adoption of any
shows a lower level of perception in accepting the AI-based technology (Dwivedi et al., 2021; Vimalkumar et al., 2021).
decision in their business functionality. However, very little literature featured factors like fairness,
Surprisingly, despite ADMS as a recent technology, AI readiness, and accountability in ADMS. Since the ADMS
technological complexity and AI readiness significantly initiative is still at its nascent stage, various factors are yet to
influence users' perceptions. Instead of how complex and be explored for the successful reception of ADMS.
handling the system is, decision-makers are more interested Also, this study makes a valuable methodological con-
in the accountability and fairness of results provided by tribution by providing a framework (see Fig. 1.) consist-
ADMS. Moving to decision makers' specific parameters, ing of an analysis approach using survey and SMP data, as
self-efficacy; and prior knowledge of AMDS affect the the idea is to provide the solution to deal with unstructured
acceptance of AI, as solely relying on decisions made by text data. The framework employs hierarchical clustering
technology contradicts their intent and tendency to acquire and an LDA-based topic modeling approach to extract top-
knowledge, which is also similar to Erskine et al. (2019) ics, sentiment analysis, and statistical analysis to evalu-
indication of the insignificant effect of users with high self- ate the impact-assessment relationship between exacted
efficacy on task technology fit. and dependent variables. However, past studies have used

13
Information Systems Frontiers

a similar framework (Kushwaha et al., 2021b). Still, their influencing the users' perception of ADMS, but ADMS is
topic validation approach is tested using intercoder reliabil- still in its developing stage. There can emerge many more
ity, which leaves the topic to variables assignment sole in the factors gradually that can be addressed in future research.
hand of investigating author. So, extracting topics from SMP While moving beyond the descriptive analytics process,
data and validating the topics to variables relationship using very few approaches are available that could use social
surveys can help scholars understand the ADMS better, and media text-based data for empirical validation purposes. In
its subsequent outcomes are reliable. this context, very few have indicated the process of map-
ping the variable relation with the dependent variable in a
5.2 Practical Implications structural form (Liu et al., 2021a, b). Due to this limitation,
we currently only looked for the direct effect of all those
This study has significant practical implications since extracted variables from the descriptive and topic modeling
ADMS is still nascent (Enholm et al., 2021). The results analysis for empirical validation purposes. We have not con-
and findings exhibit the critical factors impacting the users' sidered this paper's mediation or indirect effect and kept it
perspective regarding the ADMS, affecting its receptions. as a future research direction. We have updated the limita-
Further it provides ADMS developers insights to develop tion of the report by incorporating this shortcoming, which
better AI-based automated decision-making systems by could be potential research for future researchers on how
looking into the factors that hampers the users' percep- we could move beyond direct effect to overall, indirect, or
tion of ADMS, which is a significant challenge (Meske & mediation effect.
Bunde, 2022; Motorny et al., 2021). The current research
can also provide some inferences to the firms or industries
trying to automate their decision-making system regarding 6 Conclusion
what capabilities they require and what factors they should
focus on. Providers or firms and ADMS users can reference The current study tries to discover the factors impacting
this study to perceive the notion of ADMS. Our study's users' perspectives based on social media and survey data
findings depict that factors like accountability, perceived of ADMS users on Twitter. A robust research model is
risk, knowledge of ADMS, individuals' attitude, self-effi- proposed, where the entire model is segmented into three
cacy, and demographics impact the individuals' perspective phases. The first phase performs an exploratory analysis of
of ADMS the most. Therefore, the current study can be a social media data users and discusses ADMS using SMA.
good reference source when determining the factors influ- In the second phase, the statistical model analyzes the data
encing individuals' perception of ADMS, hence its overall around users' perspectives regarding the AMDS collected
reception among users. via surveys. A theoretical framework is then developed by
exploring the prior literature and using exploratory and
5.3 Limitation and Future Scope statistical analysis results. Factors like privacy/security,
perceived risk, knowledge of ADMS, users' attitudes, fair-
Since ADMS employed in many sectors and gradually ness, Accountability, Trust, self-efficacy, AI readiness, and
becoming part of what businesses perceive and reflect. This demographic factors are highlighted. In the last phase, after
study tries to determine the factors, including the users' validating the framework and testing the hypothesis around
perceptions regarding the ADMS. To achieve this, we per- these factors via a machine learning classifier, factors like
formed various analyses; in one, we collected data from perceived risk, users' attitude, accountability, self-efficacy,
SMP and analyzed it using SMA techniques. On the other, and demographic factors were the most impacting factors in
we collected data from a survey and statistically analyzed deriving the users' perception of ADMS. Since the ADMS
it. We tested our hypothesis and validated our developed is an emerging concept, this study is more relevant to rec-
framework using a combination of stepwise regression and ognizing the factors influencing users' perception of ADMS
machine learning classifier SVM. However, this study has and resolving the critical issues of ADMS reception among
some limitations. We have collected data either from the stakeholders.
SMP or from participants familiar with the ADMS; in both
cases, users who are not tech-savvy and are not familiar with
AI or ADMS are left out. Therefore, this study is based on
users with the technology and leaves a scope of research.
In the current study, we have identified only 11 factors

13
Information Systems Frontiers

Appendix 2 Questionnaire used

consciousness
for construct measures

AIemotion
caucasian
Theme11

synthetic
reactive

AIsoul
aware
know
black
asian
In this questionnaire, ADMS is the acronym for Artificial
Intelligence in decision-making systems. Kindly provide
your rating between 1 and 5, where 1 represents least sig-

installation

prejudicial

penalising
frequency
Theme10 nificant and 5 is highly significant.

version

women
gender

loan

men
race
Constructs Description Items Rate
between

development
1–5

algorithms
computing
statistics

backend
Theme9

teachers

Privacy/security Privacy/safety On what scale do

set-up
intent
Table 3  Table presenting the topic with their associated words, which is used to map with variables and circulated among respondents for their response

large
is associated you think words

UI
with users' like "safe-guard,
concern about identity, access,
representative

their personal, unauthor-


performing

historical inputs ized, privacy,


affectivity
capacity
Theme8

or privacy being confidential,


breach
strong

robots
social
belief

exposed while non-disclosure,


task

ML models unprotected,
are built, and threat, spams"
transparency

the individual are in users'


dimensions
judgments

reliability

working on the text/comments


diagnosis
Theme7

actions

worthy

model could associated with


safety
peers
harm

use it privacy/security
of using ADMS
Perceived risk Perceived risk On what scale do
mistreatment

stakeholders
justification

indicates the you think words


responsible

intentions

user's concern like "volatility,


Theme6

explain
control
agreed

wrong
biases

of risk associ- loss, touch, vio-


ated with the lations, repeti-
use of ADMS tive, superintel-
in their plan- ligence, safety,
un-restricted

ning. Taking unwittingly,


principles
persistent
blogspot
Theme5

help from the surefire, estab-


concept
generic
Cloud

salary

verify

model owner lish" is in users'


eval

makes the user text/comments


confident about associated with
the risk associ- perceived risk
bureaucracy

substitution

ated with using of using ADMS


Theme4

delivery
service

ADMS in their
hours
SLA

slow
time
rate

decision-making
iot

Fairness Fairness repre- On what scale do


sents the pro- you think words
machine-vision

cess for making like "simulation,


specialization
intelligence

fair decisions intelligence,


simulation

interviews
AI_salary
processes
confused
Theme3

and indicates processes,


speech

that users’ cer- confused, AI_


fear

tainty about the salary, speech,


use of ADMS machine-vision,
has no bias interviews, fear,
super intelligence

over quarterly/ specialization"


annual business is in users' text/
unwittingly
violations
repetative

plans, and it comments asso-


volatility

establish
Theme2

surefire

may impact the ciated with the


safety
touch
loss

business fairness of using


ADMS
Appendix 1

non-disclosure
unauthorized

confidential

unprotected
safe-guard
Table 3

Theme1

identity

privacy
access

spams
threat

13
Information Systems Frontiers

Constructs Description Items Rate Constructs Description Items Rate


between between
1–5 1–5
Accountability Accountability On what scale do Attitude Attitude indicates On what scale do
indicates the you think words users' felling you think words
ability to take like "biases, or opinion of like "caucasian,
responsibility, mistreatment, using ADMS black, asian,
decide an action agreed, respon- for business reactive, aware,
plan, and be sible, control, functionalities. know, con-
accountable for explain, wrong, Like it is a wise sciousness, syn-
action. Using justification, and good idea to thetic, AIsoul,
ADMS, users stakeholders, use ADMS AIemotion" are
need a consider- intentions" are in users' text/
able opportunity in users' text/ comments asso-
or independ- comments asso- ciated with the
ence in making ciated with the attitude toward
quarterly/annual accountability using ADMS
business plans of using ADMS AI readiness AI readiness rep- On what scale do
and have signifi- resents the steps you think words
cant autonomy or principles to like "Cloud,
in determining collect informa- blogspot, sal-
Trust Trust indicates On what scale do tion about ary, generic,
users' trust in you think words the system, concept, un-
the decision like "actions, procedure, and restricted, eval,
made by AI. judgments, equipment so verify, persis-
Like, using diagnosis, that it can be tent, principles"
ADMS makes peers, harm, accessible in are in users'
it easier to safety, worthy, the future. Like, text/comments
have faith in reliability, Users intend to associated with
quarterly/annual transparency, use ADMS for AI readiness to
business plans dimensions" are future business use ADMS
in users' text/ plan activities
comments asso- Self-efficacy Self-efficacy indi- On what scale
ciated with the cates individual do you think
trust of using belief to execute words like
ADMS the task con- "representative,
Technological Technological On what scale do fidently. Like, breach, strong,
complexity complexity you think words users’ belief belief, capacity,
assesses the like "bureau- in the require- performing,
required tech- cracy, service, ment of strong task, affectivity,
nological level delivery, rate, confidence social, robots"
for the design, SLA, time, and behavioral are in users'
manufacturing, substitution, intention to run text/comments
and operation hours, slow, and use ADMS associated with
of the ADMS. iot" are in users' self-efficacy of
This construct text/comments using ADMS
indicates users’ associated with Demographic Demographic On what scale do
indication of the technologi- represents users you think words
ease of use cal complexity experience and like "installa-
and skillful- of using ADMS opportunities tion, version,
ness in system till time they frequency,
operation during get, in decision gender, preju-
decision making making dicial, race,
loan, penalizing,
women, men"
are in users'
text/comments
associated with
the demo-
graphic of using
ADMS

13
Information Systems Frontiers

Baabdullah, A. M., Alalwan, A. A., Slade, E. L., Raman, R., & Kha-
Constructs Description Items Rate
tatneh, K. F. (2021). SMEs and artificial intelligence (AI): Ante-
between
cedents and consequences of AI-based B2B practices. Industrial
1–5
Marketing Management, 98, 255–270. https://​doi.​org/​10.​1016/j.​
Knowledge of Knowledge On what scale do indma​rman.​2021.​09.​003
ADMS about ADMS you think words Balakrishnan, A., Kumara, S. R. T., & Sundaresan, S. (1999). Manufac-
represents the like "teachers, turing in the digital age: Exploiting information technologies for
users knowledge statistics, intent, product realization. Information Systems Frontiers, 1(1), 25–50.
surrounding large, comput- https://​doi.​org/​10.​1023/A:​10100​12712​144
ADMS internal ing, algorithms, Balakrishnan, J., Dwivedi, Y. K., Hughes, L., & Boy, F. (2021). Ena-
algorithm and development, blers and inhibitors of AI-Powered voice assistants: A dual-factor
operational UI, backend, approach by integrating the status quo bias and technology accept-
procedure set-up" are ance model. Information Systems Frontiers. https://​doi.​org/​10.​
in users' text/ 1007/​s10796-​021-​10203-y
comments Balmer, R. E., Levin, S. L., & Schmidt, S. (2020). Artificial intelligence
associated with applications in telecommunications and other network industries.
knowledge of Telecommunications Policy, 44(6), 101977. https://​doi.​org/​10.​
ADMS 1016/j.​telpol.​2020.​101977
Bankins, S., Formosa, P., Griep, Y., & Richards, D. (2022). AI decision
Other items except these captured the user’s information including making with dignity? Contrasting workers’ justice perceptions of
their profession details mentioned on their profile human and AI decision making in a human resource management
context. Information Systems Frontiers. https://​doi.​org/​10.​1007/​
s10796-​021-​10223-8
Baur, A. W. (2017). Harnessing the social web to enhance insights into
Author Contributions  All authors have contributed equally, and all people’s opinions in business, government and public administra-
contributors have been included as authors. tion. Information Systems Frontiers, 19(2), 231–251. https://​doi.​
org/​10.​1007/​s10796-​016-​9681-7
Behera, R. K., Bala, P. K., & Ray, A. (2021). Cognitive chatbot for
Declarations  personalised contextual customer service: Behind the scene and
beyond the hype. Information Systems Frontiers, 2. https://d​ oi.o​ rg/​
Conflict of Interests  The authors have no conflict of interests to de- 10.​1007/​s10796-​021-​10168-y
clare. Berente, N., Seidel, S., & Safadi, H. (2019). Research commentary—
data-driven computationally intensive theory development. Infor-
mation Systems Research, 30(1), 50–64. https://​doi.​org/​10.​1287/​
isre.​2018.​0774
References Büschken, J., & Allenby, G. M. (2016). Sentence-based text analysis
for customer reviews. Marketing Science, 35(6), 953–975. https://​
Agarwal, R., & Prasad, J. (1998). The antecedents and consequents of doi.​org/​10.​1287/​mksc.​2016.​0993
user perceptions in information technology adoption. Decision Calzarossa, M. C., De Lotto, I., & Rogerson, S. (2010). Ethics and
Support Systems, 22(1), 15–29. https://​doi.​org/​10.​1016/​S0167-​ information systems — Guest editors’ introduction. Informa-
9236(97)​00006-7 tion Systems Frontiers, 12(4), 357–359. https://​doi.​org/​10.​1007/​
Alahakoon, D., Nawaratne, R., Xu, Y., De Silva, D., Sivarajah, U., & s10796-​009-​9198-4
Gupta, B. (2020). Self-building artificial intelligence and machine Chatterjee, S., & Kar, A. K. (2018). Effects of successful adoption of
learning to empower big data analytics in smart cities. Information information technology enabled services in proposed smart cities
Systems Frontiers. https://​doi.​org/​10.​1007/​s10796-​020-​10056-x of India: From user experience perspective. Journal of Science
Alsheibani, S., Cheung, Y., & Messom, C. (2018). Artificial Intel- and Technology Policy Management, 9(2), 189–209. https://​doi.​
ligence Adoption: AI-readiness at Firm-Level. PACIS 2018 Pro- org/​10.​1108/​JSTPM-​03-​2017-​0008
ceedings. 37.  https://​aisel.​aisnet.​org/​pacis​2018/​37. Accessed 22 Chatterjee, S., Ghosh, S. K., Chaudhuri, R., & Chaudhuri, S. (2020).
Oct 2021 Adoption of AI-integrated CRM system by Indian industry: From
Ameen, N., Tarhini, A., Reppel, A., & Anand, A. (2021). Customer security and privacy perspective. Information and Computer Secu-
experiences in the age of artificial intelligence. Computers in rity. https://​doi.​org/​10.​1108/​ICS-​02-​2019-​0029
Human Behavior, 114(August 2020), 106548. https://​doi.​org/​10.​ Chatterjee, S., Chaudhuri, R., & Vrontis, D. (2021a). Usage intention
1016/j.​chb.​2020.​106548 of social robots for domestic purpose: From security, privacy, and
Araujo, T., Helberger, N., Kruikemeier, S., & de Vreese, C. H. (2020). legal perspectives. Information Systems Frontiers. https://​doi.​org/​
In AI we trust? Perceptions about automated decision-making by 10.​1007/​s10796-​021-​10197-7
artificial intelligence. AI and Society, 35(3), 611–623. https://​doi.​ Chatterjee, S., Rana, N. P., Khorana, S., Mikalef, P., & Sharma, A.
org/​10.​1007/​s00146-​019-​00931-w (2021b). Assessing Organizational Users' Intentions and Behav-
Aribarg, A., & Schwartz, E. M. (2020). Native advertising in online ior to AI Integrated CRM Systems: a Meta-UTAUT Approach.
news: Trade-offs among clicks, brand recognition, and website Information Systems Frontiers, (7491). https://​doi.​org/​10.​1007/​
trustworthiness. Journal of Marketing Research, 57(1), 20–34. s10796-​021-​10181-1
https://​doi.​org/​10.​1177/​00222​43719​879711 Dwivedi, Y. K., Hughes, L., Ismagilova, E., Aarts, G., Coombs, C.,
Awa, H. O., Ojiabo, O. U., & Orokor, L. E. (2017). Integrated technol- Crick, T., & Williams, M. D. (2021). Artificial Intelligence (AI):
ogy-organization-environment (T-O-E) taxonomies for technology Multidisciplinary perspectives on emerging challenges, opportu-
adoption. Journal of Enterprise Information Management, 30(6), nities, and agenda for research, practice and policy. International
893–921. https://​doi.​org/​10.​1108/​JEIM-​03-​2016-​0079 Journal of Information Management, 57(August 2019), 101994.
https://​doi.​org/​10.​1016/j.​ijinf​omgt.​2019.​08.​002

13
Information Systems Frontiers

Elbanna, A., Dwivedi, Y., Bunker, D., & Wastell, D. (2020). The Kar, A. K., & Kushwaha, A. K. (2021). Facilitators and barriers of arti-
search for smartness in working, living and organising: Beyond ficial intelligence adoption in business – Insights from opinions
the “Technomagic”: Editorial for special issue of information sys- using big data analytics. Information Systems Frontiers. https://​
tems frontiers. Information Systems Frontiers, 22(2), 275–280. doi.​org/​10.​1007/​s10796-​021-​10219-4
https://​doi.​org/​10.​1007/​s10796-​020-​10013-8 Kim, B., Park, J., & Suh, J. (2020). Transparency and accountability
El-Haddadeh, R., Fadlalla, A., & Hindi, N. M. (2021). Is there a place in AI decision support: Explaining and visualizing convolutional
for responsible artificial intelligence in pandemics? A tale of two neural networks for text information. Decision Support Systems,
countries. Information Systems Frontiers. https://d​ oi.o​ rg/1​ 0.1​ 007/​ 134, 113302. https://​doi.​org/​10.​1016/j.​dss.​2020.​113302
s10796-​021-​10140-w Kumar, P., Dwivedi, Y. K., & Anand, A. (2021). Responsible Artifi-
Enholm, I. M., Papagiannidis, E., Mikalef, P., & Krogstie, J. cial Intelligence (AI) for value formation and market performance
(2021). Artificial intelligence and business value: A literature in healthcare: The mediating role of patient’s cognitive engage-
review. Information Systems Frontiers. https://​doi.​org/​10.​1007/​ ment. Information Systems Frontiers. https://​doi.​org/​10.​1007/​
s10796-​021-​10186-w s10796-​021-​10136-6
Erskine, M. A., Khojah, M., & McDaniel, A. E. (2019). Location selec- Kushwaha, A. K., & Kar, A. K. (2021). MarkBot – A language
tion using heat maps: Relative advantage, task-technology fit, and model-driven chatbot for interactive marketing in post-modern
decision-making performance. Computers in Human Behavior, world. Information Systems Frontiers. https://​doi.​org/​10.​1007/​
101, 151–162. https://​doi.​org/​10.​1016/j.​chb.​2019.​07.​014 s10796-​021-​10184-y
Foshay, N., Mukherjee, A., & Taylor, A. (2007). Does data warehouse Kushwaha, A. K., Kar, A. K., & Dwivedi, Y. K. (2021a). Applica-
end-user metadata add value? Communications of the ACM, tions of big data in emerging management disciplines: A literature
50(11), 70–77. https://​doi.​org/​10.​1145/​12977​97.​12978​00 review using text mining. International Journal of Information
Gide, A. (1967). Building trust in artificial intelligence. Angewandte Management Data Insights, 1(2), 100017. https://​doi.​org/​10.​
Chemie International Edition, 6(11), 951–952., 72(1), 5–24. 1016/j.​jjimei.​2021.​100017
Gupta, M., Parra, C. M., & Dennehy, D. (2021a). Questioning racial Kushwaha, A. K., Kumar, P., & Kar, A. K. (2021b). What impacts
and gender bias in AI-based recommendations: Do espoused customer experience for B2B enterprises on using AI-enabled
national cultural values matter? Information Systems Frontiers. chatbots? Insights from Big data analytics. Industrial Marketing
https://​doi.​org/​10.​1007/​s10796-​021-​10156-2 Management, 98(September), 207–221. https://​doi.​org/​10.​1016/j.​
Gupta, S., Kamboj, S., & Bag, S. (2021b). Role of risks in the develop- indma​rman.​2021.​08.​011
ment of responsible artificial intelligence in the digital healthcare Li, Y., & Kettinger, W. J. (2021). Testing the relationship between
domain. Information Systems Frontiers. https://​doi.​org/​10.​1007/​ information and knowledge in computer-aided decision-mak-
s10796-​021-​10174-0 ing. Information Systems Frontiers. https://​doi.​org/​10.​1007/​
Gupta, S., Modgil, S., Bhattacharyya, S., & Bose, I. (2022). Artificial s10796-​021-​10205-w
intelligence for decision support systems in the field of opera- Li, J., Huang, J., Zheng, L., & Li, X. (2020). Application of artificial
tions research: Review and future scope of research. Annals of intelligence in diabetes education and management: present status
Operations Research, 308(1–2), 215–274. https://d​ oi.o​ rg/1​ 0.1​ 007/​ and promising prospect. Frontiers in Public Health, 8, 173.https://​
s10479-​020-​03856-6 doi.​org/​10.​3389/​fpubh.​2020.​00173
Hong, Y. H., Teh, B. H., Vinayan, G., Soh, C. H., Khan, N., & Ong, T. Lin, H. F., & Lee, G. G. (2005). Impact of organizational learning
S. (2013). Investigating the Factors Influence Adoption of Internet and knowledge management factors on e-business adoption. Man-
Banking in Malaysia: Adopters Perspective. International Journal agement Decision, 43(2), 171–188. https://d​ oi.o​ rg/1​ 0.1​ 108/0​ 0251​
of Business and Management, 8(19). https://d​ oi.o​ rg/1​ 0.5​ 539/i​ jbm.​ 74051​05819​02
v8n19​p24 Liu, F., Lai, K.-H., Wu, J., & Duan, W. (2021a). Listening to online
Im, I., Kim, Y., & Han, H. J. (2008). The effects of perceived risk and reviews: A mixed-methods investigation of customer experience
technology type on users’ acceptance of technologies. Informa- in the sharing economy. Decision Support Systems, 149, 113609.
tion and Management, 45(1), 1–9. https://​doi.​org/​10.​1016/j.​im.​ https://​doi.​org/​10.​1016/j.​dss.​2021.​113609
2007.​03.​005 Liu, R., Gupta, S., & Patel, P. (2021b). The application of the prin-
Jöhnk, J., Weißert, M., & Wyrtki, K. (2021). Ready or Not, AI ciples of responsible AI on social media marketing for digital
Comes— An interview study of organizational AI readiness fac- health. Information Systems Frontiers. https://​doi.​org/​10.​1007/​
tors. Business and Information Systems Engineering, 63(1), 5–20. s10796-​021-​10191-z
https://​doi.​org/​10.​1007/​s12599-​020-​00676-7 Loh, X.-K., Lee, V.-H., Loh, X.-M., Tan, G. W.-H., Ooi, K.-B., &
Kai-ming au, A., & Enderwick, P. (2000). A cognitive model on atti- Dwivedi, Y. K. (2021). The dark side of mobile learning via
tude towards technology adoption. Journal of Managerial Psy- social media: How bad can it get? Information Systems Frontiers,
chology, 15(4), 266–282. https://​doi.​org/​10.​1108/​02683​94001​ (0123456789). https://​doi.​org/​10.​1007/​s10796-​021-​10202-z
03309​57 Luo, X., Tong, S., Fang, Z., & Qu, Z. (2019). Frontiers: Machines vs.
Kar, A. K. (2020). What affects usage satisfaction in mobile payments? humans: The impact of artificial intelligence chatbot disclosure on
Modelling user generated content to develop the “Digital service customer purchases. Marketing Science, 38(6), 937–947. https://​
usage satisfaction model.” Information Systems Frontiers. https://​ doi.​org/​10.​1287/​mksc.​2019.​1192
doi.​org/​10.​1007/​s10796-​020-​10045-0 Manfreda, A. (2020). Smart city adoption: An interplay of construc-
Kar, A. K., & Aswani, R. (2021). How to differentiate propagators of tive and adverse factors. Lecture Notes in Business Information
information and misinformation–Insights from social media ana- Processing, pp. 179–192. https://​doi.​org/​10.​1007/​978-3-​030-​
lytics based on bio-inspired computing. Journal of Information 44322-1_​14
and Optimization Sciences, 42(6), 1307–1335. https://​doi.​org/​10.​ Meske, C., & Bunde, E. (2022). Design principles for user interfaces in
1080/​02522​667.​2021.​18801​47 AI-Based decision support systems: The case of explainable hate
Kar, A. K., & Dwivedi, Y. K. (2020). Theory building with big data- speech detection. Information Systems Frontiers. https://​doi.​org/​
driven research–Moving away from the “What” towards the 10.​1007/​s10796-​021-​10234-5
“Why.” International Journal of Information Management, 54, Mikalef, P., Lemmer, K., Schaefer, C., Ylinen, M., Fjørtoft, S. O.,
102205. https://​doi.​org/​10.​1016/j.​ijinf​omgt.​2020.​102205 Torvatn, H. Y., … Niehaves, B. (2021b). Enabling AI capabilities
in government agencies: A study of determinants for European

13
Information Systems Frontiers

municipalities. Government Information Quarterly, (June). https://​ Subramaniam, C., Shaw, M. J., & Gardner, D. M. (2000). Product mar-
doi.​org/​10.​1016/j.​giq.​2021.​101596 keting and channel management in electronic commerce. Informa-
Mikalef, P., & Gupta, M. (2021). Artificial intelligence capability: tion Systems Frontiers, 1(4), 363–378. https://​doi.​org/​10.​1023/A:​
Conceptualization, measurement calibration, and empirical study 10100​61924​822
on its impact on organizational creativity and firm performance. Sun, Y., & Jeyaraj, A. (2013). Information technology adoption and
Information and Management, 58(3), 103434. https://​doi.​org/​10.​ continuance: A longitudinal study of individuals’ behavioral
1016/j.​im.​2021.​103434 intentions. Information & Management, 50(7), 457–465. https://​
Motorny, S., Sarnikar, S., & Noteboom, C. (2021). Design of an intel- doi.​org/​10.​1016/j.​im.​2013.​07.​005.
ligent patient decision aid based on individual decision-making Tiwary, N. K., Kumar, R. K., Sarraf, S., Kumar, P., & Rana, N. P.
styles and information need preferences. Information Systems (2021). Impact assessment of social media usage in B2B mar-
Frontiers. https://​doi.​org/​10.​1007/​s10796-​021-​10125-9 keting: A review of the literature and a way forward. Journal of
Nicolaou, A. I., & McKnight, D. H. (2006). Perceived information Business Research, 131, 121–139. https://​doi.​org/​10.​1016/j.​jbusr​
quality in data exchanges: Effects on risk, trust, and intention to es.​2021.​03.​028
use. Information Systems Research, 17(4), 332–351. https://​doi.​ Trocin, C., Mikalef, P., Papamitsiou, Z., & Conboy, K. (2021). Respon-
org/​10.​1287/​isre.​1060.​0103 sible AI for digital health: A synthesis and a research agenda.
Oh, O., Eom, C., & Rao, H. R. (2015). Research note —role of social Information Systems Frontiers, (May). https://​doi.​org/​10.​1007/​
media in social change: An analysis of collective sense making s10796-​021-​10146-4
during the 2011 Egypt revolution. Information Systems Research, Venkatesh, V., Brown, S., & Bala, H. (2013). Bridging the qualitative-
26(1), 210–223. https://​doi.​org/​10.​1287/​isre.​2015.​0565 quantitative divide: Guidelines for conducting mixed methods
Önkal, D., Gönül, M. S., & Lawrence, M. (2008). Judgmental adjust- research in information systems. MIS Quarterly, 37(1), 21–54.
ments of previously adjusted forecasts. Decision Sciences, 39(2), Venkatesh, V., Brown, S. A., & Sullivan, Y. W. (2016). Guidelines for
213–238. https://​doi.​org/​10.​1111/j.​1540-​5915.​2008.​00190.x conducting mixed-methods research: An extension and illustra-
Pillai, R., Sivathanu, B., & Dwivedi, Y. K. (2020). Shopping inten- tion. Journal of the Association of Information Systems, 17(7),
tion at AI-powered automated retail stores (AIPARS). Journal of 435–495.
Retailing and Consumer Services, 57, 102207. https://​doi.​org/​10.​ Vimalkumar, M., Sharma, S. K., Singh, J. B., & Dwivedi, Y. K. (2021).
1016/j.​jretc​onser.​2020.​102207 “Okay google, what about my privacy?”: User’s privacy percep-
Pu, P., & Chen, L. (2007). Trust-inspiring explanation interfaces for tions and acceptance of voice based digital assistants. Computers
recommender systems. Knowledge-Based Systems, 20(6), 542– in Human Behavior, 120(October 2020), 106763. https://​doi.​org/​
556. https://​doi.​org/​10.​1016/j.​knosys.​2007.​04.​004 10.​1016/j.​chb.​2021.​106763
Ridhwan, K. M., & Hargreaves, C. A. (2021). Leveraging Twitter data Votto, A. M., Valecha, R., Najafirad, P., & Rao, H. R. (2021). Artificial
to understand public sentiment for the COVID-19 outbreak in intelligence in tactical human resource management: A systematic
Singapore. International Journal of Information Management literature review. International Journal of Information Manage-
Data Insights, 1(2), 100021. https://​doi.​org/​10.​1016/j.​jjimei.​ ment Data Insights, 1(2), 100047.
2021.​100021 Wang, C. H. (2016). A novel approach to conduct the importance-satis-
Riedel, R., Wiers, V., & Fransoo, J. C. (2012). Modelling dynamics in faction analysis for acquiring typical user groups in business-intel-
decision support systems. Behaviour & Information Technology, ligence systems. Computers in Human Behavior, 54, 673–681.
31(9), 927–941. https://​doi.​org/​10.​1080/​01449​29X.​2010.​535969 https://​doi.​org/​10.​1016/j.​chb.​2015.​08.​014
Robey, D. (1979). User attitudes and management information system Wang, W., Chen, L., Xiong, M., & Wang, Y. (2021). Accelerating AI
use. Academy of management Journal, 22(3), 527–538. 255742. adoption with responsible AI signals and employee engagement
Ruggieri, S., Bonfanti, R. C., Passanisi, A., Pace, U., & Schimmenti, mechanisms in health care. Information Systems Frontiers. https://​
A. (2021). Electronic surveillance in the couple: The role of doi.​org/​10.​1007/​s10796-​021-​10154-4
self-efficacy and commitment. Computers in Human Behavior, Wendt, C., Adam, M., Benlian, A., & Kraus, S. (2021). Let's connect
114(August 2020), 106577. https://​doi.​org/​10.​1016/j.​chb.​2020.​ to keep the distance: How SMEs leverage information and com-
106577 munication technologies to address the COVID-19 crisis. Infor-
Sharma, S., Rana, V., & Kumar, V. (2021). Deep learning based seman- mation Systems Frontiers, (0123456789). https://d​ oi.o​ rg/1​ 0.1​ 007/​
tic personalized recommendation system. International Journal of s10796-​021-​10210-z
Information Management Data Insights, 1(2), 100028. Wijnhoven, F. (2021). Organizational learning for intelligence amplifi-
Shet, S. V., Poddar, T., Wamba Samuel, F., & Dwivedi, Y. K. (2021). cation adoption: Lessons from a clinical decision support system
Examining the determinants of successful adoption of data analyt- adoption project. Information Systems Frontiers, (0123456789).
ics in human resource management – A framework for implica- https://​doi.​org/​10.​1007/​s10796-​021-​10206-9
tions. Journal of Business Research, 131(August 2020), 311–326. Yuthas, K., & Young, S. T. (1998). Material matters: Assessing the
https://​doi.​org/​10.​1016/j.​jbusr​es.​2021.​03.​054 effectiveness of materials management IS. Information and Man-
Silver, S. D. (2020). Dynamics of negative evaluations in the informa- agement, 33(3), 115–124. https://d​ oi.o​ rg/1​ 0.1​ 016/S
​ 0378-7​ 206(97)​
tion exchange of interactive decision-making teams: Advancing 00028-1
the design of technology-augmented GDSS. Information Systems
Frontiers. https://​doi.​org/​10.​1007/​s10796-​020-​10063-y Publisher's Note Springer Nature remains neutral with regard to
Spathis, C., & Ananiadis, J. (2005). Assessing the benefits of using an jurisdictional claims in published maps and institutional affiliations.
enterprise system in accounting information and management.
Journal of Enterprise Information Management, 18(2), 195–210.
https://​doi.​org/​10.​1108/​17410​39051​05799​18

13
Information Systems Frontiers

Amit Kumar Kushwaha  is a part-time research scholar in Informa- Arpan Kumar Kar  is Amar S Gupta Chair Professor in Indian Institute
tion Systems at the Indian Institute of Technology Delhi, India, and a of Technology Delhi, India. His research interests are in the domain
full-time AI/ML practitioner with over 13 years in the industry, cur- of data science, digital transformation, internet platforms and public
rently working in the capacity of data science practice head of Hewlett policy. He has authored over a 170 peer reviewed articles and edited 9
Packard. A gold-medallist engineering graduate with a Hons. Degree, books. Out of these publications, over 50 publications are in ABDC-A,
he completed his Masters (MBA) from SIBM, Pune. His interests lie CABS-3 and Clarivate WoS Q1 quality journals, including 11 ABDC-
in research cutting across business applications of AI/ML innovations A* publications. He is the Editor in Chief of International Journal
and productionalization of these solutions for the consumption of end- of Information Management Data Insights, published by Elsevier. He
users. He has published more than 7 peer-reviewed papers in high- is also Associate / Coordinating Editor in Int. Journal of Electronic
quality journals and conference proceedings. He has received the best Government Research, Journal of Public Affairs, Information Systems
conference paper award at the ISIC conference held in Delhi 2021. He Frontiers and Global Journal of Flexible Systems Management. He
has also worked part-time as a guest faculty in various educational received the Research Excellence Award by Clarivate Analytics for
institutes and is currently associated as well. highest WoS citations in India (2015–2020). He also received the Bas-
ant Kumar Birla Award for the count of ABDC A*/ABS 4 publications
Ruchika Pharswan  is associated with Indian Institute of Technology in India (2014–2019). He received the Best Seller Award from Ivey
Delhi as a PhD Scholar in the Bharti School of Telecommunications Cases / Harvard Business Publishing in 2020. He has received over
Technology and Management. She is working in the space of Govern- 20 other awards like 3 IFIP Best Papers, ACM ICEGOV Best Paper,
ance of Artificial Intelligence. Before joining IIT Delhi, she has done Tata Consultancy Services Gold Medal, Project Management Institute
her M.Tech and B.Tech in Computer Science. Award, AIMS Research Award, IIT Delhi Teaching Award, 6 Outstand-
ing Reviewer Awards from Elsevier and many more.
Prashant Kumar  is currently a research scholar at the Department
of Management Studies, Indian Institute of Technology Delhi, India.
Before this, he has worked as a research assistant at National Institute
of Technology, Arunachal Pradesh, India. He has published articles
in Industrial Marketing Management, Journal of Business Research,
Renewable and Sustainable Energy Review. His research interests are
renewable energy, data analytics, customer experience, and artificial
intelligence.

13

You might also like