Factors influencing the adoption of artificial intelligence-based digital assistants
Published: 2019, Issue 3/2019
Abstract. The aim of our research is to investigate the adoption and use of artificial intelligence-based digital assistants, and to identify the factors influencing adoption and use. Our secondary objective is to summarise the scientific literature published to date regarding digital assistants. Through our conceptual model, we seek to answer the question of how the five independent variables examined, which originate from the UTAUT2 model, explain the development of behavioural intention to use. The five independent variables in question are performance expectancy, effort expectancy, social influence, facilitating conditions, and hedonic motivation. The study was conducted on a sample of N=123, the majority of whom were university students. Partial least squares – structural equation modelling (PLS-SEM) was used to analyse the data. We concluded that performance expectancy, effort expectancy, social influence, and hedonic motivation significantly influence behavioural intention to use. These five independent variables explain 69% of the variance in the dependent variable.
Keywords: artificial intelligence, digital assistant, technology adoption, behavioural intention to use, UTAUT2 model
Publisher: Hungarian Economic Association of Romania and the Hungarian Institute of the Faculty of Economics and Business Administration, Babeș-Bolyai University ISSN: 1582-1986 www.econ.ubbcluj.ro/kozgazdaszforum
Factors influencing the adoption of artificial intelligence-based digital assistants SEER LÁSZLÓ1 – VIZELI IBOLYA2 The aim of our research is to investigate the adoption and use of artificial intelligence-based digital assistants, and to identify the factors influencing adoption and use. Our secondary objective is to summarise the scientific literature published to date regarding digital assistants. Through our conceptual model, we seek to answer how the five independent variables examined, which originate from the UTAUT2 model, explain the development of behavioural intention. The five independent variables under consideration are performance expectancy, effort expectancy, social influence, facilitating conditions, and hedonic motivation. The research was conducted on a sample of N=123, the majority of whom were university students. Partial least squares structural equation modelling (PLS-SEM) was used to analyse the data. We concluded that performance expectancy, effort expectancy, social influence, and hedonic motivation significantly influence behavioural intention. These five independent variables explain 69% of the variance in the dependent variable.
Keywords: artificial intelligence, digital assistant, technology adoption, behavioural intention, UTAUT2 model.
JEL codes: M31, M39, O33, L86.
Introduction
In recent decades, digital technology has developed at an unprecedented rate. In the era of digitalisation, people's daily habits, corporate workflows, and even the structure of the labour market are undergoing significant transformations (Fehér 2016).
There is no universally accepted definition of the concept of digitalisation. Brennen and Kreiss (2016) base their definition of digitalisation on social life; according to them, the starting point is how people interact with one another. These interactions are increasingly moving away from analogue technologies (traditional letters, telephone calls) and towards digital ones (e-mail, chat, social media), thus digitising many areas of life (work, leisure) (Brennen–Kreiss 2016).
According to the formulation by the innovation research organisation Gartner, digitalisation occurs through those digital technologies that companies use to change their business models in order to secure new revenue streams and value-adding opportunities (Gartner 2017). It is a generally accepted fact that many companies today are rethinking their business models, and these new business models are heavily influenced by digital technologies.
The development of digital technology is expected to continue at an increasing rate, which will also have a significant impact on the multifaceted system of relationships between consumers and companies. According to another forecast by Gartner, by 2020, 85% of interactions between consumers and companies will take place without the consumer coming into contact with people on the part of the company (Gartner 2011). Therefore, in order to satisfy consumer expectations to an increasingly high standard, companies must place great emphasis on considering their increasingly diverse needs and expectations when automating processes that affect consumers.
In this regard, marketing sub-sectors and professionals play an outstanding role, as they are the ones who understand consumer needs. One of the key technologies of this process automation is artificial intelligence, within the framework of which several technologies can be discussed (e.g. machine learning, deep learning, neural networks, natural language processing, etc.) (Bostrom 2014).
The focus of our research is the investigation of the phenomenon of the principles according to which consumers accept or reject a new, digital technological solution. The subject of the research is the investigation of the consumer acceptance of so-called digital assistants based on artificial intelligence and the factors influencing such acceptance.
The literature on technology acceptance is extremely rich (Nysveen–Pedersen 2016); therefore, our investigation of the willingness to accept new technology is based on the existing, extensive literature and, within it, on the Unified Theory of Acceptance and Use of Technology (UTAUT2).
The relevance of the topic lies in the fact that the gradual rise of artificial intelligence exerts a significant impact on both people's daily lives and on companies. This affects the field of marketing in three ways: firstly, through the radical change of marketing workflows (as numerous activities become automatable). Secondly, consumer expectations are also undergoing a transformation because various artificial intelligence systems are beginning to become part of their daily lives (digital assistants such as Netflix, Alexa, and Siri). Thirdly, similar to many other fields, the marketing profession itself is being completely transformed due to artificial intelligence-based technologies (Kotler et al. 2017).
According to Brynjolfsson and McAfee (2014), many people may lose their jobs during technological development, and many jobs will cease to exist. This primarily affects those areas where it is not necessary for the employee to possess specialised skills and abilities (for example, workers on production lines). However, new jobs will also be created that require specialised skills and competencies, all of which will not be impossible to acquire (Brynjolfsson–McAfee 2014).
At the same time, however, significant time can be saved in both production and everyday life through automation and the other significant added value of artificial intelligence-based technologies; the derivative benefits of this are extremely diverse (Brynjolfsson–McAfee 2014). We are researching one specific part of the impact of this phenomenon on consumers, as uncovering this is timely for the marketing profession.
In conclusion, investigating consumer expectations in the era of the rise of artificial intelligence can provide useful information for marketing professionals. Furthermore, the topic is insufficiently researched at both international and domestic levels, despite the fact that it exerts a significant impact on the lives of consumers and companies, and completely transforms certain professions, including the marketing profession.
Literature Review
Definition of Artificial Intelligence
Artificial intelligence refers to the set of technologies that are capable not only of the programmatic processing of diverse information but also of the development and adaptation of their own processes. Thus, artificial intelligence-based software is a programme that is capable of developing and adapting itself within given parameters (Mijwel 2015; Bostrom 2014).
The birth of the science of artificial intelligence can be traced back to 1950 (Nguyen–Sidorova 2017), as this was when Alan Turing’s seminal work, *Computing Machinery and Intelligence*, was published. In the article, Turing formulated the model that became known as the Turing test, the essence of which is as follows: a computer passes the test if a (human) subject participating in the test, during a text-based, question-and-answer interaction with another (human) subject and a computer, cannot determine with absolute certainty which participant is the computer and which is the human (Turing 1950). In the following years, much of the research related to artificial intelligence focused on creating a machine capable of producing or imitating abilities associated with human intelligence (Nguyen–Sidorova 2017). Such abilities include, for example, communication in some language, the storage of knowledge and access to knowledge, prediction derived from multivariate data analysis, making rational decisions, and adapting to inputs from the environment (Russell–Norvig 2010).
According to the summary by Nguyen and Sidorova (2017), the science of artificial intelligence has three main aspects. Firstly, artificial intelligence encompasses automated reasoning, natural language, image processing, knowledge representation, and machine learning (Bishop 1995; Jain et al. 1996). Secondly, most artificial intelligence research focuses on the creation of models, representations, and algorithms capable of reproducing the aforementioned abilities identifiable with human intelligence; however, a large part of these can be assimilated into traditional computer science applications and thus form part of behavioural-style research (Gregor–Benbasat 1999; Se et al. 2007). Thirdly, the science of artificial intelligence is characterised by rapid developmental stages coupled with high expectations, which are subsequently followed by long, very slow periods of development and a lack of public interest (Bostrom 2014).
Classification of artificial intelligence systems
According to Bostrom (2014), artificial intelligence systems can be classified into three categories: Artificial Narrow Intelligence (ANI), Artificial General Intelligence (AGI), and Artificial Super Intelligence (ASI).
ANI constitutes the first level of artificial intelligence systems, which are characterised by providing outstanding performance in a specific domain.
An example of this is the "supercomputer" developed by IBM, named Deep Blue, which defeated world chess champion Garry Kasparov during a chess match in 1997 (Gurkaynak et al. 2016; Vardi 2012). A significant portion of existing artificial intelligence systems can be classified into this category; for instance, Google’s RankBrain system and Apple’s assistant, Siri, qualify as ANI.
Furthermore, according to Gurkaynak et al. (2016), AGIs are Human-Level AIs, i.e., computers that are intelligent in every respect as a human is, and are capable of performing any task requiring intellectual work that a human is also capable of solving.
AGIs are expected to be capable of solving complex problems arising in various fields, as well as independently perceiving and controlling their own thoughts, concerns, feelings, strengths, weaknesses, and inclinations (Pennachin–Goertzel 2007).
Finally, ASIs are systems that are significantly more intelligent than humans in virtually every field, including scientific creativity, general wisdom, and social skills (Bostrom 2014).
Most artificial intelligence researchers believe that following the development of AGIs, they will further develop themselves into ASI systems, a phenomenon referred to in the literature as an intelligent explosion or the singularity (Yudkowsky et al. 2010; Urban 2015). The impact of artificial intelligence on the transformation of marketing processes: After defining artificial intelligence, it is important to determine its significance in the transformation of marketing processes. According to Olson and Levy (2018), artificial intelligence reshapes the field of marketing at three levels: interaction, understanding, and reasoning.
Firstly, artificial intelligence affects the interaction between consumers and companies in two ways: on the one hand, through searches, and on the other, through the activities of bots (search technologies). Artificial intelligence provides consumers with the opportunity to communicate with brands in a more natural and personalised manner through various smart services and systems possessing agentic competencies (agents), such as bots, digital assistants, smart speakers, and in the future, connected cars (Olson–Levy 2018).
The primary application area of chatbots (software capable of adapted responses) is customer service and customer relationship management; however, as these tools evolve and communicate more effectively, they may become capable of participating in every stage of the consumer decision-making process, creating significant added value for the consumer through their involvement (e.g. reducing search time, facilitating more rational decisions, etc.). Digital assistants and smart speakers are capable of understanding who the user is, what they are doing, and where they are doing it; they answer questions, perform tasks requested by the consumer, and are capable of predicting consumer desires based on their behavioural patterns (Sterne 2017).
The widespread adoption of assistants contributes to the transformation of processes such as content production and the personalisation of products and services, as well as the collection and utilisation of insights; furthermore, they reshape advertising production through the creation of new advertising interfaces. For brands, these assistants become important tools in building strong consumer–brand relationships and in satisfying consumer needs (Olson–Levy 2018; Sterne 2017).
Secondly, artificial intelligence enables companies to understand consumers in a more natural manner based on text, audio, image, video, and gesture-based inputs. Microsoft’s Cognitive Services, which is used by 600,000 developers, allows marketing professionals to easily add intelligent enhancements to their applications, devices, and platforms, such as emotion and sentiment detection, vision and speech recognition, language understanding, and knowledge and search (Olson–Levy 2018).
Finally, artificial intelligence offers marketing professionals the opportunity to process vast quantities of data (e.g. big data) and identify patterns that are invisible to the human eye. The processing of massive datasets enables the identification of insights that professionals can utilise for the design and optimisation of campaigns and media mix strategies. Furthermore, data analysis allows for the generation of forecasts based on established search and behavioural patterns, which can contribute to the success of various advertising campaigns (Olson–Levy 2018).
In summary, artificial intelligence is significantly reshaping consumer habits and corporate activities, thereby transforming marketing processes and presenting marketing professionals with new challenges, while simultaneously providing them with new tools. One of the most important such tools is the digital assistant, which will be presented in detail below.
Defining digital assistants
In English-language literature, the most frequently used term is
Intelligent Personal Assistant (IPA), which encompasses those applications that provide assistance to the user based on various inputs (for example, the user's voice), namely by answering questions, making recommendations, and performing various tasks (Baber 2002). In English-language literature, the terms Conversational Agents, Virtual Personal Assistants, Personal Digital Assistants, Voice-Enabled Assistants, or Voice Activated Personal Assistants are used as synonyms for IPA (Cowan et al. 2017).
In Hungarian-language literature, there is no uniformly accepted term for these types of applications, while in colloquial language, the most frequently used terms are virtual or digital assistant and intelligent assistant. Hereafter, the term digital assistant (DA) will be used.
A digital assistant is a so-called software agent that assists (human) users in performing professional/administrative (messaging, setting alarms, scheduling, food ordering), technical (complex tasks such as managing home automation systems via smart device control, or monitoring the user's health status by analysing signals emitted by wearable devices), and social (communication between the user and the assistant) tasks by automating and facilitating their daily activities (Saad et al. 2016; Santos et al. 2016). According to Spohrer et al. (2017), digital assistants are decision support tools capable of extending and enhancing human capabilities and expertise through a profound understanding of the environment, thereby enabling them to make suggestions to users that contribute to better, data-driven decision making (Sterne 2017).
The operating principle and use of digital assistants
Digital assistants combine speech recognition, language understanding, dialogue management, language generation, and speech synthesis to respond to queries, requests, and questions from users (Siddike et al. 2018). In terms of their operation, digital assistants utilise speech recognition and natural language processing algorithms: in the first step, the DA listens to and records the user's command via a built-in microphone (speech recognition); subsequently, the recording is sent via the internet to a natural language processing server, which interprets the recorded audio (language understanding) and sends back an appropriate response (text output) (Han–Yang 2018).
According to Lison and Meena (2014), in the stage following speech recognition and language understanding, the DA selects the (response) operation most appropriate for the input (the user's question, request, or instruction) by taking into account the meaning of the recognised expression, the current state of the dialogue, and previously shared information between the user and the system (via the dialogue manager); this is followed by the planning of the content intended for the user, i.e., the generation of the response (via speech synthesis) and its output (Lison–Meena 2014). Some DAs (for example, Siri and Google Assistant), provided they have permission to access data, also utilise information available on the user's mobile phone to construct appropriate responses (Milhorat et al. 2014). In recent years, not only voice but also emotion recognition techniques have been developed, which operate by recording and interpreting the user's facial expressions (Knight 2016).
Digital assistants available on the commercial market can be accessed in four different ways: 1. as a specific function embedded within an operating system (Apple Siri, Google Now), 2. in the form of downloadable mobile applications (Amazon Alexa), 3. installed on specialised hardware devices (smartphones, smartwatches, smart speakers, PCs, laptops) (Amazon Echo, Google Home), and 4. in the form of robots (JIBO). Currently, DA types in the form of intelligent speakers are the most popular on the market (Han–Yang 2018). Furthermore, voice-based digital assistants (Siri, Google Assistant, Microsoft Cortana, and Amazon Alexa) are widely available applications via smartphones, which provide assistance in solving tasks related to the user's home (Amazon Echo, Google Home) or vehicle (the integration of Google Assistant into certain Hyundai vehicle models) (Cowan et al. 2017).
Table 1 summarises the tasks that some of the previously mentioned digital assistants are capable of performing, as well as the platforms that a given assistant uses while performing those tasks. The significance of digital assistants is further demonstrated by the fact that, according to forecasts, the value of the DA market will reach $4.61 billion by 2020 (Kamitis 2016).
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Apple Siri
Table or figure – see the PDF version of the issue for its exact content.
IBM Watson
· Watson Engagement Advisor: interacts with the user, listens to their questions, and provides solutions · Watson Explorer: accesses databases, analyses structured and unstructured datasets · Watson Discovery Adviser: establishes connections between data points · Watson Oncology: analyses patients' medical data in order to provide evidence-based treatment recommendations
Table or figure – see the PDF version of the issue for its exact content.
· answers questions, makes suggestions, and performs tasks in accordance with user requests · is capable of both voice- and text-based communication with the user · Google Home: a voice-based device that answers user questions based on Google searches, starts music players, and performs other everyday tasks · conducts voice-based communication · starts music players, compiles to-do lists, sets alarms, broadcasts podcasts and audiobooks, and provides information on weather, traffic conditions, and other real-time events · is capable of controlling certain types of smart devices
Table or figure – see the PDF version of the issue for its exact content.
· can set reminders, use maps, write and send emails, and run searches from Microsoft · informs the user about flight bookings and the latest news from its platform Source: Siddike et al. 2018
Summary of research examining the use of digital assistants In recent years, several studies related to digital assistants have been published. According to a survey by Creative Strategies, 98% of iPhone users have used Siri, the digital assistant; however, only 30% are those who use the DA regularly, and this phenomenon is not only observed in the case of Siri, but also affects other DAs (Alexa, Cortana, Google Assistant) (Milanesi 2016). Luger and Sellen (2016) examined users who regularly use their digital assistant and concluded that users' expectations of the DA are significantly lower than what the technology can provide. Kiseleva et al. (2016) investigated whether satisfaction with the tasks performed by the DA varies depending on the areas of use and circumstances.
Cowan et al. (2017) examined those users who have tried the DA but use it only occasionally. Among the reasons for irregular use are factors such as cultural habits, norms, and the tension or anxiety regarding what others might think (social embarrassment) (Cowan et al. 2017). According to Babic et al. (2018), one of the most important factors influencing the widespread adoption of VAs is quality, which in this case is the combination of usability and user experience.
Manikonda et al. (2018) examined user preferences and privacy concerns regarding DAs, and established that several privacy issues arise during the use of DAs; in such instances, users often return the product to the vendor, mute the microphone of the device on which the DA was used, or narrow down the list of tasks performed by the DA (for example, only setting an alarm clock). Siddike et al. (2018) pointed out that the use of a DA improves the performance of the individual (user), meaning that those users who conduct successful interactions with the DA may develop their knowledge and intelligence. Furthermore, they demonstrated that the key to DA success lies in trust, privacy, accuracy, and ease of use (Siddike et al. 2018).
Jiang et al. (2015) were the first to develop an automated method through which satisfaction with tasks performed by voice-based digital assistants can be measured, and based on which different DAs and the tasks they perform can be compared. The method can identify sessions performed by the DA that the consumer evaluates as either appropriate or inappropriate (Jiang et al. 2015). Using user data from a commercial DA, Sano et al. (2016) developed a model capable of predicting potential users' commitment to a DA. Han and Yang (2018) developed a comprehensive research model that explains the factors related to users' acceptance and use of DAs. Based on their results, the factors influencing the acceptance of DAs include interpersonal attraction—comprising task attraction, social attraction, and physical attraction—as well as security and data privacy risks (Han–Yang 2018).
Some research concerning DAs has focused on the examination of specific groups. Wulf et al. (2014) investigated the use of digital assistants among the elderly, while Lovato and Piper (2015) focused on children. Babic et al. (2018) examined the perceived user experience and performance regarding digital assistants applied within the higher education system, with the studied group consisting of students.
In summary, the majority of existing research examining digital assistants has aimed to investigate satisfaction with the DA and its influencing factors, the identification of factors influencing DA acceptance, and the uncovering of problems arising during use and their underlying causes. Few scientific works have dealt with the investigation of behavioural intention towards the DA and its influencing factors, despite the fact that intention is closely correlated with actual usage (Venkatesh et al. 2003). According to Siddike et al. (2018), DAs will be capable of fulfilling all kinds of occupations and social roles within service systems in the future, marking the beginning of the era of smart, people-centered service systems. That is, digital assistants contribute to the transformation of consumer habits and, consequently, of services and service processes. Service systems are becoming more people-centered thanks to digital assistants, and both companies and marketing professionals must react to this process.
According to Coskun et al. (2017), the presentation of the Unified Theory of Acceptance and Use of Technology suggests that, in order to ensure the adoption and adaptation of new technologies by the general public, it is necessary to identify and understand the expectations of future users. Therefore, identifying the factors influencing the intention to use digital assistants is an important and necessary step in technology implementation. For the purpose of examining the acceptance and behavioural intention of users towards DAs, the Unified Theory of Acceptance and Use of Technology (UTAUT) proves to be the most appropriate.
The UTAUT model was developed by Venkatesh in 2003 through the review and integration of eight previous theories, which are as follows: the Theory of Reasoned Action (TRA), the Theory of Planned Behaviour (TPB), the Technology Acceptance Model (TAM), the motivational model, a model combining TAM and TPB, the Model of PC Use, Rogers' Diffusion of Innovations theory, and Social Cognitive Theory (Williams et al. 2015; Keszei–Zsukk 2017).
According to the summary by Keszei and Zsukk (2017), the aim of the UTAUT model is to provide managers with a tool through which they can estimate the probability of success associated with the introduction of a new technology, understand the factors related to the adoption of that technology, and plan potential intervention steps. The four elements of the model (Figure 1)—performance expectancy (PE), effort expectancy (EE), social influence (SI), and facilitating conditions (FC)—are factors that directly influence behavioural intention (BI) and indirectly influence actual use behaviour (UB). The extent of the impact of the four aforementioned factors is influenced by gender, age, experience, and the degree of voluntariness of use, which appear as moderating factors in the model.
Performance expectancy expresses the extent to which the system assists the user in achieving better job performance. Effort expectancy expresses how much effort the user believes will be required to use the system. Social influence expresses how important the user feels the opinions of important people are while using the system. Finally, facilitating conditions comprise the factors that have a direct impact on actual use, namely the extent to which these are available to the user (Venkatesh 2003; Keszei–Zsukk 2017).
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Above all, it can be concluded that the greater the enjoyment or pleasure associated with the use of technology, the more likely consumers are to adopt the given technology, and the higher their behavioural intention to use it.
Research methodology
Research questions and hypotheses
Following a review of research concerning digital assistants, the following research questions were formulated: 1. Which factors influence the formation of potential users' behavioural intention towards digital assistants? 2. To what extent do the identified factors influence the behavioural intention towards digital assistants?
In order to answer the aforementioned research questions, a conceptual model applied in this study was developed based on the UTAUT2 model (Figure 3). Following a review of the relevant literature, five variables were identified that may possess the greatest explanatory power regarding behavioural intention towards digital assistants. The arrows between the variables in the conceptual model indicate the research hypotheses, which are as follows: Hypothesis 1 (H1+): There is a significant and positive relationship between the performance expectancy of the digital assistant and the behavioural intention to use it. Hypothesis 2 (H2+): There is a significant and positive relationship between the effort expectancy associated with the use of the digital assistant and the behavioural intention to use it. Hypothesis 3 (H3+): There is a significant and positive relationship between social influence regarding the use of the digital assistant and the behavioural intention to use it. Hypothesis 4 (H4+): There is a significant and positive relationship between facilitating conditions promoting the use of the digital assistant and the behavioural intention to use it. Hypothesis 5 (H5+): There is a significant and positive relationship between hedonistic motivation related to the use of the digital assistant and the behavioural intention to use it.
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Data collection was carried out between April and May 2019 using an online snowball sampling method via the Google Forms service. The final sample size was N=123. The data were processed using the SPSS statistical software and the Smart PLS 3.0 software (Ringle et al. 2015).
Data processing and results
During the questionnaire-based research, data were collected from a total of 123 respondents from Romania and Hungary. The majority of respondents (63.4%) were female; 46.3% were under the age of 21, while 48.8% were between 22 and 31 years old; 87.8% live in urban areas; regarding educational attainment, 42.3% had completed high school or vocational school, and 39% had completed undergraduate studies. The majority of respondents (74.4%) were students, while 14.6% were employees. Examining monthly gross income, it appears that 50.4% have a monthly income of less than 1000 RON (which is understandable, as most respondents are still students and do not have their own earned income), 14.6% live on an amount between 1001 and 1500 RON, while 12.2% earn over 3000 RON. 69.1% of the respondents had already heard of digital assistants, whereas 28.5% had not yet encountered the phenomenon (Table 4). Table 4. Summary of respondents' sociodemographic data Non
Table or figure – see the PDF version of the issue for its exact content.
Occupation
Table or figure – see the PDF version of the issue for its exact content.
Table 5 summarises the descriptive statistical analysis pertaining to the conceptual model. In the case of statements regarding expected performance, the mean values range between 2.98 and 3.46, meaning that the majority of respondents selected the "partially disagree" and "neutral to me" options.
Interestingly, for the statement "the digital assistant would be useful in my daily life", the mean is 3.46, indicating that most respondents chose the "neutral to me" and "partially agree" options; whereas for the statement "using the digital assistant would help me to be more productive", the value is 2.98, meaning most chose "partially disagree" and "neutral to me". Regarding statements concerning expected required effort, the mean shows a value between 4.11 and 4.26, which implies that most responses to the given statements were "partially agree" and "strongly agree". This suggests that the majority of respondents believe that using the DA would be easy, that they would easily learn to operate and use the device, that interaction with the DA would be clear and understandable to them, and that they would easily become proficient in its use.
For statements regarding social impact, the mean shows a value between 1.89 and 1.96, thus the overwhelming majority either did not agree at all or partially disagreed with the related statements. In the case of statements regarding facilitating conditions and hedonistic motivation, the value ranges between 3.28 and 3.74, meaning the majority selected the "neutral to me" and "partially agree" responses. Regarding behavioural intention, the mean is between 2.74 and 2.98, meaning most respondents partially disagree or are neutral regarding whether they would like to use, or plan to use, a DA in the future, as well as whether they would frequently use the DA while performing their tasks (Table 5).
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Testing validity and reliability
Due to the small sample size (123), we employed partial least squares structural equation modelling (PLS-SEM) to examine the relationships between the variables (Hair et al. 2014). The PLS-SEM algorithms were executed using the SmartPLS 3.0 statistical software (Ringle et al. 2015).
During the modelling process, we tested the existence and strength of the previously assumed relationships between six variables. We investigated the extent to which expected performance of the digital assistant, expected required effort, social influence, facilitating conditions, and hedonistic motivation (independent variables) influence the intention to use the digital assistant and its formation (dependent variable). It is important to note that the PLS-SEM methodology employs a different computational method than the methodology used by the SPSS statistical analysis programme for measuring the validity and reliability between latent variables (the variables in the research model: VT, VSZE, TH, EF, HM, HSZ) and their indicators (the statements used in the questionnaire research to measure the variables in the research model: VT1, VT2, VT3, VT4, VSZE1, VSZE2, VSZE3, VSZE4, TH1, TH2, TH3, EF1, EF2, EF3, EF4, HM1, HM2, HM3, HSZ1, HSZ2, HSZ3), as well as between the indicators themselves (Henseler–Sarstedt 2013); thus, it provided the opportunity for a more recent interpretation of the variable indicators.
In the process of running the PLS algorithm, we used the default PLS algorithm path weighting scheme. The data were processed in z-standardised form. The maximum number of iterations was set at 300, and for the bootstrapping algorithm, we worked with a setting of 500 bootstrap resamples (Hair et al. 2014; Henseler et al. 2014).
The first step in analysing the research model is estimating the reliability and validity between the latent variables and the indicator variables, as well as between the indicators themselves. To this end, we examined the metrics recommended by the latest PLS methodology for these cases (Hair et al. 2014; Henseler et al. 2014). The aforementioned metrics are as follows: · indicator reliability: analysis of the values (path loadings) between the constructs and their indicators; · internal consistency reliability:
Cronbach's Alpha, composite reliability, and the Dijkstra and Henseler $\rho_A$ coefficient; · convergent validity: AVE (average variance extracted) index; · overall model fit assessment: SRMR index.
The most important validity metrics are summarised in Table 6. For the evaluation of indicator reliability, a value of 0.7 was considered the benchmark (Hair et al. 2014); therefore, those indicators that did not reach this value (EF4 and HM1) were removed from the model, and the analysis was re-run. In the analysis performed without the EF4 and HM1 indicators, all indicators met the 0.7 threshold criterion for indicator reliability.
Internal consistency reliability was examined based on Cronbach's Alpha, composite reliability, and the Dijkstra and Henseler $\rho_A$ coefficient. The composite reliability, Cronbach's Alpha, and the Dijkstra-Henseler $\rho_A$ indices all possess values above the 0.7 threshold in every case, and are thus acceptable.
The AVE indicator, which examines convergent validity, is acceptable in all cases, as it possesses values above the 0.5 threshold; thus, convergent validity is sufficient for every construct. Model fit was assessed using the SRMR indicator (Henseler 2015). The indicator must have a value below 0.8 for the fit to be considered acceptable. In the present case, the value is 0.059, meaning the model fit is acceptable. Finally, discriminant validity was examined (Appendix 2).
Hypothesis testing
Following the estimation of the reliability and validity between the latent variables and the indicator variables, as well as among the indicators themselves, the next step is to run the bootstrapping algorithm to test the research hypotheses. The results of the procedure are summarised in Table 7.
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Factor weight
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Based on Table 7, the following relationships exist:
The perceived performance (PP) of the digital assistant has a significant and positive effect on intention to use (path coefficient = 0.421, p < 0.01); thus, the first hypothesis is accepted (H1).
The perceived effort required (PER) regarding the use of the digital assistant has a significant and positive effect on intention to use (path coefficient = 0.145, p < 0.05); thus, the second hypothesis is accepted (H2).
The social influence (SI) regarding the use of the digital assistant has a significant and positive effect on intention to use (path coefficient = 0.265, p < 0.01); thus, the third hypothesis is accepted (H3).
Facilitating conditions (FC) for the use of the digital assistant do not have a significant effect on intention to use (path coefficient = 0.018, p > 0.05); thus, the fourth hypothesis is rejected (H4). Hedonistic motivation (HM) regarding the use of the digital assistant has a significant and positive effect on intention to use (path coefficient = 0.262, p < 0.01); thus, the fifth hypothesis is accepted (H5).
The variables in the research model contribute to the formation of intention to use as follows: perceived performance explains 42.1% of the variance (path coefficient = 0.421), perceived effort required explains 14.5% (path coefficient = 0.145), social influence explains 26.5% (path coefficient = 0.265), and hedonistic motivation explains 26.2% (path coefficient = 0.262) of the variance in intention to use. In the case of facilitating conditions, the relationship was not significant, with a path coefficient value of 0.018, or 1.8%. The conceptual model established by us explains 68.9% of the variance in intention to use digital assistants (R² = 0.689) (Figure 4).
Evaluation of results and conclusions
Summarising the results of the research, it can be stated that four factors were successfully identified which exert a significant and positive effect on the intention to use digital assistants. These are as follows: perceived performance, social influence, hedonistic motivation, and perceived effort required.
The four variables collectively explain 68.9% of the variance in intention to use, which we consider to be a good result.
The perceived performance of the device exerts the greatest effect on the intention to use digital assistants (t-value = 5.287) in the examined sample
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Table or figure – see the PDF version of the issue for its exact content.
Cowan, B. R.–Pantidi, N.–Coyle, D.–Morrissey, K.–Clarke, P.–Al-Shehri, S. 2017. “What can I help you with?”: Infrequent users’ experiences of intelligent personal assistants. In: Proceedings of the 19th International Conference on Human-Computer Interaction with Mobile Devices and Services, 1–12. Fehér, K. 2016. Digitalisation and new media. Trends, strategies, illustrations. Budapest: Akadémiai kiadó.
Gartner 2017. High-Tech Manufacturing Supply Chain innovators 2017: Demand-Driven Digitalization. https://www.gartner.com/doc/3724117?srcId=1-3931087981, accessed: 2019. 04. 07.
Gartner 2011. Gartner Customer Summit 2011. https://www.gartner. com/imagesrv/ summits/docs/na/customer-360/C360_2011_brochure_FINAL.pdf, accessed: 2019. 02. 26.
Gregor, S.–Benbasat, I. 1999. Explanations from intelligent systems: Theoretical foundations and implications for practice. MIS Quarterly: Management Information Systems 23(4), 497–530.
Gurkaynak, G.–Yilmaz, I.–Haksever, G. 2016. Stifling artificial intelligence: Human perils. Computer Law & Security Review 32(5), 749–758.
Hair, J. F.–Hult, J. T. M.–Ringle, C.–Sarstedt, M. 2014. A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM). London: Sage.
Han, S.–Yang, H. 2018. Understanding adoption of intelligent personal assistants:
A parasocial relationship perspective. Industrial Management & Data Systems 118(3), 618–636.
Henseler, J. 2015. PLS Path Modeling: Introduction and Applications. ADANCO Seminar, 2015 November 3-5, Cluj-Napoca.
Henseler, J.–Ringle, C. M.–Sarstedt, M. 2014. A new criterion for assessing discriminant validity in variance-based structural equation modeling. Journal of the Academy of Marketing Science 43, 115–135.
Henseler, J.–Sarstedt, M. 2013. Goodness-of-fit indices for partial least squares path modeling. Computational Statistics 28(2), 565–580.
Jain, A.–Mao, J.–Mohiuddin, K. 1996. Artificial neural networks: A tutorial. IEEE Computer 29(3), 31–34.
Jiang, J.–Awadallah, A. H.–Jones, R.–Ozertem, U.–Zitouni, I.–Gurunath Kulkarni, R. 2015. Automatic online evaluation of intelligent assistants. In: Proceedings of the 24th International Conference on World Wide Web, 506–516.
Kamitis 2016. Intelligent Personal Assistant: Products, technologies and market – 2017. http://kamitis.cluster014.ovh.net/IMG/pdf/ipa_-sample.pdf, accessed: 2019. 04. 30.
Keszey, T.–Zsukk, J. 2017. Consumer acceptance of new technologies. An overview and critical evaluation of Hungarian and international literature. Vezetéstudomány. Budapest Management Review 48(10), 38–47.
Kiseleva, J.–Williams, K.–Hassan, A.–Crook, A.–Zitouni, I.–Anastasakos, T. 2016.
Predicting User Satisfaction with Intelligent Assistants. In: SIGIR ‘16: Proceedings of the
39th International ACM SIGIR conference on Research and Development in Information Retrieval – July 2016, 45–54.
Knight, W. 2016. Amazon working on making Alexa recognize your emotions.
MIT Technology Review. https://www.technologyreview.com/s/601654/amazon-workingon-making-alexa-recognize-your-emotions/, accessed: 2019. 04. 30. Kotler, P.–Kartayala, H.–Setiawan, I. 2017. Marketing 4.0: Moving from Traditional to Digital. Hoboken, NJ: Wiley.
Lison, P.–Meena, R. 2014. Spoken dialogue systems: The new frontier in human-computer interaction. ACM Crossroads 21(1), 46–51.
Lovato, S.–Piper, A. 2015. “Siri, is This You?”: Understanding Young Children’s Interactions with Voice Input Systems. In: Proceedings of the 14th International Conference on Interaction Design and Children (IDC ’15), 335–338.
Lowenthal, J. 2010. Using mobile learning: Determinates impacting behavioral intention. The American Journal of Distance Education 24(4), 195–206.
Luger, E.–Sellen, A. 2016. Like having a really bad PA: The gulf between user expectation and experience of conversational agents. In: Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems, 5286–5297.
Manikonda, L.–Deotale, A.–Kambhampati, S. 2018. What’s up with Privacy?: User Preferences and Privacy Concerns in Intelligent Personal Assistants. In: Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society, 229–235.
Marchewka, J. T.–Kostiwa, K. 2007. An Application of the UTAUT Model for Understanding Student Perceptions Using Course Management Software. Communications of the IIMA 7(2), 93–104.
Mijwel, M. 2015. History of Artificial Intelligence. PhD Thesis. Baghdad: Baghdad College of Economics Sciences University.
Milanesi, C. 2016. Voice Assistant Anyone? Yes please, but not in public! http:// creativestrategies.com/voice-assistant-anyoneyes-please-but-not-in-public/, accessed: 2019. 04. 26.
Milhorat, P.–Schlögl, S.–Chollet, G.–Boudy, J.–Esposito, A.–Pelosi, G. 2014. Building the next generation of personal digital assistants. In: IEEE 1st International Conference on Advanced Technologies for Signal and Image Processing (ATSIP), 458–463.
Nguyen, Q. N.–Sidorova, A. 2017. AI capabilities and user experiences: a comparative study of user reviews for assistant and non-assistant mobile apps. In: Twenty-third Americas Conference on Information Systems.
Nysveen, H.–Pedersen, P. E. 2016. Consumer adoption of RFID-enabled services.
Applying an extended UTAUT model. Information Systems Frontiers 18(2), 293–314.
Olson, C.–Levy, J. 2018. Transforming marketing with artificial intelligence. Applied Marketing Analytics 3(4), 291–297.
Pennachin, C.–Goertzel, B. 2007. Contemporary Approaches to Artificial General Intelligence. In: Artificial General Intelligence, Heidelberg: Springer.
Ringle, C.–Wende, S.–Becker, J-M. 2015. SmartPLS 3. http://www.smartpls.com, accessed: 2019. 05. 10.
Russell, S. J.–Norvig, P. 2010. Artificial Intelligence: A Modern Approach. Hoboken, NJ: Pearson Education.
Saad, U.–Afzal, U.–El-Issawi, A.–Eid, M. 2016. A model to measure QoE for virtual personal assistant. Multimedia Tools and Applications 76(10), 1–21.
Sano, S.–Kaji, N.–Sassano, M. 2016. Prediction of prospective user engagement with intelligent assistants. In: Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics 1, 1203–1212.
Santos, J.–Rodrigues, J. J.–Casal, J.–Saleem, K.–Denisov, V. 2016. Intelligent personal assistants based on internet of things approaches. IEEE Systems Journal 12(2), 1793–1802.
Se, U.–Xiao, B.–Benbasat, I. 2007. E-Commerce Product Recommendation Agents:
Use, Characteristics, and Impact. MIS Quarterly: Management Information Systems 31(1), 137–209.
Siddike, M.–Kalam, A.–Spohrer, J.–Demirkan, H.–Kohda, Y. 2018. People’s interactions with cognitive assistants for enhanced performances. In: Proceedings of the 51st Hawaii International Conference on System Sciences, 1–19.
Siow, B. S. 2016. Validated UTAUT2 questionnaire for elderly. https://www.researchgate.net/publication/307907913_Validated_UTAUT2_questionnaire_for_elderly#fullTextFileContent, accessed: 2019. 03. 11. Spohrer, J.–Bassano, C.–Piciocchi, P.–Siddike, M. 2017. What makes a system smart? Wise? Advances in The Human Side of Service Engineering. New York, NY: CRC Press.
Sterne, J. 2017. Artificial Intelligence for Marketing: practical applications. Hoboken, NJ: Wiley.
Turing, A. M. 1950. Computing Machinery and Intelligence. Mind, New Series 59(236), 433–460.
Urban, T. 2015. AI revolution – road to superintelligence. https://waitbutwhy.com/2015/01/artificial-intelligence-revolution-1.html, accessed: 2019. 03. 11.
Vardi, M. Y. 2012. Artificial intelligence: past and future. Association for Computer Machinery 55(1), 1–16.
Venkatesh, V.–Morris, M. G.–Davis, G. B.–Davis, F. D. 2003. User acceptance of information technology: Toward a unified view. MIS Quarterly 27(3), 425–478.
Venkatesh, V.–Thong, J. Y.–Xu, X. 2012. Consumer acceptance and use of information technology: extending the unified theory of acceptance and use of technology. MIS Quarterly 36(1), 157–178.
Wang, Y.–Wu, M.–Wang, H. 2009. Investigating the determinants and age and gender differences in the acceptance of mobile learning. British Journal of Educational Technology 40(1), 92–118.
Williams, M. D.–Rana, N.–Dwivedi, Y. K. 2015. The unified theory of acceptance and use of technology (UTAUT): A literature review. Journal of Enterprise Information Management 28(3), 443–488.
Wulf, L.–Garschall, M.–Himmelsbach, J.–Tscheligi, M. 2014. Hands Free – Care Free: Elderly People Taking Advantage of Speech-only Interaction. In: Proceedings of the 8th Nordic Conference on Human-Computer Interaction: Fun, Fast, Foundational (NordiCHI ’14), 203–206.
Yudkowsky, E.–Salamon, A.–Shulman, C.–Kaas, S.–McCabe, T.–Nelson, R. 2010.
Reducing long-term catastrophic risks from artificial intelligence. https://intelligence.org/files/ReducingRisks.pdf, accessed: 2019. 03. 11.
Appendix 1. Consumer acceptance and use of digital assistants – questionnaire 1. Have you ever interacted with any type of digital assistant application? (for example, Siri, Alexa, Google Assistant, Cortana, etc.) * yes * no * I do not remember 2. Please rate your level of agreement with the following statements (1 – "strongly disagree", 2 – "somewhat disagree", 3 – "neutral", 4 – "somewhat agree", 5 – "strongly agree").
1 A digital assistant would be useful in my daily life. Using a digital assistant would help me accomplish things that are important to me.
Using a digital assistant would help me complete my tasks more quickly.
Using a digital assistant would help me be more productive.
I would find it easy to learn how to use a digital assistant.
Interacting with a digital assistant would be clear and understandable to me.
Using a digital assistant seems easy to me. I would find it easy to become proficient in using a digital assistant.
People who are important to me think I should use a digital assistant.
Table or figure – see the PDF version of the issue for its exact content.
Text extracted automatically from the PDF; for tables, figures and exact layout see the PDF of the issue.
