OK Google, help me learn: an exploratory study of voice-activated artificial intelligence in the classroom

ABSTRACT

Artificial intelligence (AI) will be in the future lives of children at school today. Voice-activated intelligent personal assistant devices are used in the home and could be useful in the classroom. This article explores how two groups of New Zealand children aged 7–12 engaged with Google Home devices in their classroom. Interactions recorded through the devices were analysed to evaluate how the devices were used, how easy they were to use and how the use aligns with the purposes of education. A framework for analysis was developed from Davis’s Technology Acceptance Model and Biesta’s three functions of education. The children used the devices across the three functions of education. They anthropomorphised the device, talking respectfully, sharing jokes and asking for its opinion. Use was influenced by the social context of the classroom and teacher beliefs about education, and limited as the devices were not designed for the classroom.

KEYWORDS:

• Digital assistants
• primary school
• anthropomorphise
• TAM
• artificial intelligence

Introduction

Interactions are changing in the post-human classroom where digital technologies are becoming an integral part of the teaching and learning context (Selwyn, 2016). Oral/aural classroom interactions have traditionally occurred between students, between a student and a teacher, and among students and teachers. Such interactions have a cognitive and social dimension for student learning (Mercer, 2008) and the introduction of digital tools to the classroom alters classroom interactions (Harper, 2018; Øvereng & Gamlem, 2022).

The integration of voice-activated intelligent personal assistant devices into classroom contexts is likely to have implications for teaching and learning in the future. Technology such as Amazon Echo, Google Home and Apple HomePod use natural language processing to create an interaction between human and computer in the home. They are stand-alone devices with Internet of Things connectivity and voice recognition that use algorithms to respond to questions and carry out multi-step and complex tasks requested by users including answer questions, tell jokes, send messages to people, other devices or applications, set reminders and play music (Canbek & Mutlu, 2016). Voice-activated intelligent personal assistant devices have become affordable and intuitive to use in the domestic context and are needing to be evaluated and critiqued for their usefulness in classroom settings (Selwyn, 2023; Williamson & Eynon, 2020) for educational purposes.

Literature review

This review considers how digital assistants have evolved, oral language interactions, how voice-activated intelligent devices are used in the classroom context, and a framework to evaluate usefulness.

Digital or virtual assistants are embedded in online environments to provide navigational assistance using a range of functions such as chatbots in shopping websites or a virtual character to guide users towards specific learning goals in curriculum websites. Anthropomorphic virtual assistants within software programs, such as the Microsoft Word paper clip, were designed to help users of the software by increasing their ability to cope with information overload (Moreale & Watt, 2003), allowing independent or agentic use of computer programs. In formal learning settings, virtual assistants provide motivational benefits to learners in certain situations (Schroeder et al., 2013) to help users achieve academically. However, virtual assistants are primarily text based with either a computer-generated voice that uses text-to-speech software or a recorded human voice that lacks interactive functions (R. O. Davis et al., 2021).

Digital tools evolve over time as technological advances allow for innovation. Voice-activated intelligent devices are the next generation of digital assistants which are interactive, with embedded the Internet of Things and artificial intelligence (AI). They are designed to mimic human oral interactions (Dousay & Hall, 2018), which is important in the primary school setting where children’s use of oral language in the classroom helps to develop their language abilities and thinking (Mercer, 2008). Voice-activated intelligent devices provide a format for learning through oral interactions that differs to laptop text-based technology. Such interactions could contribute to children’s language development and thinking.

Oral communication is inherently social, and children’s oral interactions in the world differ from adults. Children naturally anthropomorphise non-human entities to increase the predictability and comprehension of what would otherwise be an uncertain world (Epley et al., 2007). To encourage interaction, voice-activated social robots are designed with a human-sounding voice to respond to questions (Duffy, 2003). While the degree that devices in classroom contexts are anthropomorphised appears to have little influence on learning outcomes (R. O. Davis et al., 2021; Schroeder et al., 2013), it can influence children’s willingness to interact and collaborate in a language learning context (Underwood, 2017). Anthropomorphism may therefore increase children’s acceptance and use of interactive technologies for social reasons (Epley et al., 2007).

Classroom use

Voice-activated intelligent devices have been introduced to the classroom context. Amazon Echo Dots were placed in 90 classrooms in a US school district to explore how teachers and administrators used the devices for subject and task-based activities (Dousay & Hall, 2018). The teachers used applications such as music software, computational thinking and maths games for curriculum-specific teaching and generic functions such as reminders, news briefings, hosting a game or telling a story. The students used the devices to check spelling and ask questions. In this study, the teachers believed the devices were useful as a teaching assistant to deliver the curriculum; however, they reported limitations such as the algorithms were unable to respond to complex maths problems.

Voice-activated devices have also been used in second language learning settings (Dillon, 2018; Dizon, 2017). In an English as a Foreign Language classroom context, children interacting with the device collaborated, persevered, reformulated questions and self-corrected (Underwood, 2017). The device became an ‘English Helper’, where children used the tool to practise their use of oral English and asked for help without the need to approach their teacher, which enhanced student agency in the classroom. Use in the language class was limited to talking to the device rather than talking with the device as they were designed to respond to a question rather than hold a conversation (Underwood, 2021). The importance of oral language interactions in the primary classroom context extends beyond second language learning.

Concerns have been raised about how appropriate AI devices are for education (Godhe et al., 2019; Perrotta & Selwyn, 2020; Selwyn, 2023). Voice-activated intelligent devices were developed for the home consumer market with algorithms designed to maximise sales and profit and therefore may not be compatible with the classroom learning environment. There is an ethical need for educators and policymakers to be aware of what voice-activated intelligent devices are telling children and how children interact with devices (Perrotta & Selwyn, 2020) to keep students safe and make decisions about possible or appropriate integration into classroom contexts.

Evaluating usefulness

Technology should be critiqued before being introduced into schooling contexts as in the past the promise of what digital tools can accomplish was oversold (Cuban, 2001; Selwyn, 2023). In this study usefulness was evaluated drawing from two theoretical models: the Technology Acceptance Model (TAM) and Biesta’s three functions of education (Biesta, 2009).

The TAM (F. D. Davis, 1989) identifies two aspects of technological acceptance which influence intention to use: the ease of use and usefulness, which are moderated by external variables. The model has evolved over time and been tested across contexts to become prominent in understanding predictors of human behaviour towards potential acceptance or rejection of a technology (Granić & Marangunić, 2019) and applied at an organisational level as both a predictor to use, pre-implementation and post-implementation (Venkatesh & Bala, 2008). While the model was developed for the business context, it has been applied to education where it has been found to be a useful predictor of whether a technology will be used in classroom practice by teachers with teacher beliefs as an external factor that influences use (Scherer et al., 2019). The underpinning focus on intention to use means that research using the TAM model usually measures perceptions rather than actual use (Legris et al., 2003). However, it can be used as a tool to frame a critique of technology. This research explored the actual use during implementation rather than perceptions of potential use.

Within the TAM model, perceived usefulness is influenced by ease of use because the easier a tool is to use, the more useful it can be (F. D. Davis, 1989). In a classroom setting, digital technology is usable when it is intuitive, age appropriate, and students can use it autonomously. With AI technologies, usefulness and ease of use are dependent on the algorithms which determine how a device will respond during an interaction.

In the context of a classroom, technologies should have an educational purpose. Biesta (2009) identified three purposes of education: qualification (gaining academic knowledge), subjectification (developing the student as an individual) and socialisation (preparing students as future citizens). Technology that enables academic achievement and develops student confidence, autonomy or understanding of use of technology in society could be considered educationally useful.

We were interested in examining the research question: how useful are voice-activated digital devices in a primary classroom environment?

Method

Case study method (Stake, 2013) was adopted to examine use of a voice-activated intelligent agent in the classroom context. A qualitative methods approach was used in the two case studies with data from Google Home transcripts of what the students said and how the Google Home responded and teacher interviews. The study was designed to focus on the interactions between students and the technology to identify the usability and the usefulness of technology. The Google Home devices respond to voice requests in four ways: first by using the Internet of Things, the interconnection of everyday devices via the internet, enabling them to send and receive data, such as using Google Home to play music, turn on lights or send a message. Secondly, drawing on inbuilt native skills to respond to simple requests such as setting a timer, answering spelling or simple maths questions, or joke telling. The third way is web-scraping, which is quoting from a webpage to answer an information inquiry. The fourth response is either not responding or stating it does not understand.

Procedures

The two case study classrooms were selected through convenience sampling. A general call over social media and through word of mouth asked teachers to volunteer to participate in the research. All students in the class had to agree to participate with parental consent for a class to join the study as it would not be possible to distinguish which students were asking questions and only consented data could be used. Ethical approval was granted for this study through the [Victoria University of Wellington] human ethics committee reference 27,378. While initially it was thought this would make recruitment for the study difficult, this was not the case owing to the enthusiasm of the teachers and students to use this technology in the classroom.

The research was undertaken in New Zealand, where schools and teachers have flexibility and a degree of autonomy in how they implement the New Zealand Curriculum (Ministry of Education, 2007) to meet the needs of the local context and students in the classroom. Each teacher is likely to have differing approaches to how they teach and manage their learning environment. Therefore, case study examination of two contexts was chosen.

Recruitment and participants

The two case studies selected were in contrasting contexts. The first case study consisted of one teacher and 29 students in a class where a range of subjects are taught including English, the arts, health and physical education, learning languages, mathematics and statistics, science, social sciences and technology. The students were aged 7–8 and a third were English as an additional language learners. The second case study consisted of 70 students and 4 teachers who were in a shared environment with a project-based curriculum approach that included the same subjects as the first case study. The students in case study two were aged 8–12 years.

Google Home Mini devices were placed in the two case study classrooms for six weeks in the middle of the academic year (Figure 1). These devices were selected as they are intuitive to use, affordable, complement the Google suite that teachers were familiar with, and the voice-activated aspect encourages communal usage; students can use them around and with other students.

Figure 1. Google Home in the classroom environment.

The study was designed to enable the collection of student use data with minimal intervention or direction from teachers or other adults. The researcher met with the students and teacher of each class before the research began and consent had been given. Students were given a short set of guidelines to clarify digital citizenship expectations, and teachers were empowered to restrict use of the device at their discretion, something that the Classroom One teacher did. During this meeting students were given examples of enquiries for spelling and maths, and ways they could use the devices to help them with their learning were discussed. The students could use the device as they desired within the rules of their classroom context and their school’s IT policy. Teachers did not schedule students to use the device or provide them with activities encouraging them to do so. The devices were non-invasive, being small and not responding unless spoken to, and students went to the device when they wanted to use it.

Data collection

Data for this study included student interactions with the device and teacher interviews. The student questions or interactions were recorded by the devices and uploaded to the cloud. A researcher downloaded and transcribed the oral record of interactions from the Google Home App. Interviews were conducted with one teacher from each case study to explore their observations of use and the role of context during the time the devices were in the classroom.

Data analysis

Data was analysed through a priori codes developed from a framework of analysis (Figure 2). The framework was developed by integrating the Technology Acceptance Model (F. D. Davis, 1989) and Biesta’s (2009) three purposes of education where qualification is curriculum-related learning, socialisation is learning to use the device and subjectification is learning about the technology (Figure 2).

Figure 2. Framework of research. Aspects of usefulness, ease of use and actual use examined in this study.

The utility value of a technology is a strong predictor for whether a technology will be used in a classroom setting (Backfisch et al., 2021). In this study, usefulness was framed across Biesta’s three purposes of education (2009). The qualification purpose was simplified to curriculum-focused activities, socialisation, and subjectification purpose was aligned with students learning about the technology and how to use the technology for different purposes with a view that this knowledge and these skills prepare students individually and collectively for their future in a digital society. Ease of use was examined through the functionality and whether the device was able to respond to student interactions. The resulting framework (Figure 2) was used to guide analysis of data.

Results

The students used the devices in their classrooms without prompting. The children in case study one had greater potential access as there was one device shared with 29 students; however the children in case study two where one device was shared with 70 students used the device more often (Table 1). The use of the device was not evenly spread across curriculum areas or consistently across time. For example, the case study two teacher noticed a significant number of enquiries from a small group of students occurring at transition times and unstructured parts of lessons. The use of the devices evolved over time as the students learnt how to phrase questions and explored different functions and limitations of the device.

Table 1. Number of enquiries.

	Number of students	Number of enquiries	Average number of enquiries per student per week
Case study one	29	846	5
Case study two	70	2531	6

The transcripts were examined to identify successful responses to queries to explore usability. A successful interaction was where a question was answered correctly, whether the student was satisfied with the answer or not. Based on these criteria, 68.6% of all enquiries were answered successfully. Enquiries answered correctly included spelling, word definition and basic mathematical sums. Unsuccessful enquiries included requests to play music where a default music provider had not been set up, and incomplete or incoherent questions. It is possible that some enquiries which were phrased correctly by students and not answered, such as those about aliens, and insults to other students, were deliberately not answered. Non-curriculum enquiries coded as ‘curiosity’ were answered by the device if the student’s grammar was sound, but in cases where students phrased questions poorly, the device appeared to not understand them. This success rate could increase over time as children learn how to phrase questions, better understand the limitations of the devices, and the device learns through its AI how to respond to a wider range of enquiries (see Kepuska & Bohouta, 2018).

The teachers in the two case studies took different approaches to managing device use which influenced usage patterns. While the devices have ‘parental controls’ for native tasks and web-scraping, the teachers remained vigilant about appropriate internet use. The teacher in case study one expressed concern about the children using the device for literacy and numeracy answers instead of ‘using their own strategies’. The teacher believed the device would cause distractions or reduce critical thinking. Students were able to use the device freely in mornings, during breaks and briefly after school. During learning time (including transitions), the use was monitored by the teacher and teaching assistant, with the expectation it would only be used to support assigned curriculum tasks. The teacher discussed her concerns with her class and later reflected that this likely influenced students’ decisions to use the device. The teacher in case study two believed that the introduction of the technology would expand children’s understanding of AI devices and the students were encouraged to be curious and innovative with their questions. The students were given free use of the device whenever they were in their classroom, and the teachers monitored student use retrospectively for student safety through the Google Home App.

Usefulness

The transcripts were analysed to identify what the students were asking and how the devices were responding. The device was considered educationally useful when it was used for academic, socialisation or subjectification purposes (Biesta, 2009), and students were able to get a response from the device to their questions or interactions.

Academic learning

Children in both case studies used the devices for academic learning. The transcripts of the interactions between students and the device were coded against subject areas in the New Zealand Curriculum (Ministry of Education, 2007). Te reo Māori is a national language in New Zealand and was tracked independently to explore if the device understood student queries in or about this language and if the students asked questions in or about te reo Māori. While the devices were dominantly used for literacy learning, the students were asking or instructing Google to help them with learning across curriculum areas (Figure 3). English or literacy enquiries dominated device use.

Figure 3. Enquiries by learning area of the curriculum.

The device was found to be useful for academic purposes. Students were able to get responses to enquiries about English and mathematics including spelling, word definitions and basic maths operations, information or advice such as information about Bollywood dancing, beatboxing and tips for giving speeches (for example see Table 2). The devices drew on four different functions when responding to student interactions. The Internet of Things function was used when students requested a change in volume, a timer to be set or to play music. The students asked questions which drew on native skills such as how to spell words or simple maths questions. Web-scraping was used to answer questions and included the device quoting text from an internet source such as Wikipedia. The device did not answer some queries, either because the student was not understood, or because the algorithm did not enable a relevant response.

Table 2. Asking for advice on speeches.

Student Enquiry	Device response
what are the give me some speech topic OK Google…	You are the beautiful thing I’ve even seen.
what are some good speech topics	I found 16 on the website prepscholar.com. Here are the first 5. It’s Something You Know About…
what is a good speech	On the website tiedunote.com, they say: An ideal speech is one which is delivered slowly and in a usual tone…
thanks	you are extremely welcome.
what is a good speech topic to write about?	On the website blog.prepscholar.com, they say: After you’ve chosen your topic keep these three tips in mind…

The usefulness for learning languages such as te reo Māori was limited as the device did not understand the language or questions about the language. This limits the usefulness of a device in a context where indigenous languages are considered an essential part of classroom discourse (Eley & Berryman, 2019).

Socialisation and subjectification

The voice-activated devices aligned with the educational purpose of developing the individual and preparing students for their future in a digital environment. The students understood how to use the device to control the environment. For example, the teacher in case study one connected the device to her Spotify account and the classroom Smart TV, allowing students to play music and YouTube videos through the device. Case study two students requested the device play music, but as no music provider was set up these attempts were unsuccessful.

Students explored the extent the devices could respond to their enquiries by posing questions not related to classroom activities and asking for advice and opinions. Students in both case studies asked the device to tell them jokes (case study one n = 34, case study two n = 78). Although the device was able to tell them jokes on demand, whether the students found these humorous was unclear, and repetition of jokes was evident. A few weeks into the research, students in case study 2 began telling the device their own jokes, but it was not able to react to these or interact with a ‘knock-knock’ style of joke. Students tried rewording but were not able to get the device to understand their joke attempts. In testing the limits of the device students also changed their names, asking the device to refer to them as a specific name such as ‘pineapple cheeseburger’, which it did. There was evidence that the students were learning about limitations of the technology.

The classroom teachers did not find students’ use of curiosity enquiries problematic. In case study two, the teacher noted that these interactions were made during transition times by a group of boys who would otherwise apply their creative humour and inquisitiveness elsewhere in the classroom. This teacher believed these interactions were important as the students were learning about the technology, therefore socialising the students into a future digital society.

The teacher in case study two believed that it was the school’s responsibility to ensure that students were confident using digital tools and interacting with AI to prepare them for their future in a digitised society. Introducing the digital devices to students provided opportunities to explore how AI works, such as how the wording of a question and consistency of feedback influences a device’s future responses. The classroom provided a supportive environment for the students to explore the limitations and possibilities of AI in their lives. Therefore, learning was not only through the questions that were asked of the devices, but also through the experience of interacting with the AI.

Ease of use

Both case study teachers reported that students found the devices easy to use and did not require assistance to set them up in their classrooms. In case study two, where students had used Google Voice on Chromebooks, the transition to using Google Home for enquiries was not significant even though the task and administration functions were different. However, some points of frustration for students created by the limitations of the device were identified in the transcripts, and two examples are outlined.

The first example is one of inconsistent answers to spelling enquiries (Table 3). Two students finished enquiries with ‘please’ and received different responses. In the first example the device returns the spelling for ‘me please’, but in the second instance it correctly recognises that ‘please’ was not part of the spelling enquiry. While it is possible that the device learned from students speaking patterns or received a software update between the two events, this can also be viewed as an inconsistency in the way the device responded. This example is important when considering the level of respect that educators and parents may set for young people when interacting with robots as it may discourage them from using please or thank you.

Table 3. Case study one, students’ inconsistent spelling assistance.

a. (dated: 23/08/19, at 12.01 PM)
can you spell me please?	Me please is spelled m-e-p-l-e-a-s-e.
b. (dated: 04/09/19, at 9.57 AM)
can you spell through please?	Through is spelled t-h-r-o-u-g-h.

A second example of student frustration and perseverance with the device’s limitations was identified when a student asked the device for help in calculating the radius of a circle (Table 4). The student in case study two made an initial enquiry at 10.52 am, and, after a series of 30 enquiries ended the exchange at 11.12 am without seeming to have an adequate answer to complete the task. The student began their enquiry by asking the device ‘measure a circle basically’. The device responded with a web-scraping, during which the student picked up the keywords ‘radius’ and ‘circumference’, but the response was beyond the student’s current understanding. The student then asked the device ‘OK Google how do you measure a circle for kids?’ and the device responded with a simplified answer, but not with instructions the student could follow. The student attempted to rephrase the question, and the device responded with relevant information, but the continued dialogue makes it clear the responses from the device were not helping the student with their task. While the device answered the enquiry, it was not at an appropriate level. This 20-minute interaction focused on one question between the device and student illustrates perseverance that would be unusual in an interaction between peers or a student and a teacher. The device is not designed specifically for education and was unable to draw on an age- or development-appropriate response, which limits the usefulness of the technology.

Table 4. Persistence.

Attempt	Student Enquiry	Device Response
1	measure a circle basically.	On the website mathgoodies.com, they say: However, using computers, has been calculated to over a trillion digits past the decimal point. The distance around a circle is called the circumference.
2	Ok Google how do you measure a circle for kids?	On the website ducksters.com, they say: Wow, we get a circle! Radius – the radius is the distance from the centre …
3	Ok Google what’s the area of a circle?	According to Wikipedia: In geometry, the area enclosed by a circle of radius…
4	what’s the formula of how I make it a circle when…	Sorry, I’m not sure how to help with that yet.
13	what is the formula to a circle and say it slowly?	I can only help with the first request. On the website purplemath.com, they say: The centre-radius form the circle equation…
17	what is C = pi d?	According to Wikipedia: In geometry, the circumference of a circle is the distance around it.
28	Ok Google what is the radius of a circle?	On the website mathopenref.com, they say: The radius of a circle is the length of the line from the centre to any point on its edge.

Overall, the devices were usable in that students were able to use them independently with minimal guidance from the teachers or the researcher. The students used the devices for curriculum-based learning, to control the environment and to explore social interactions with a machine. The device successfully responded to two thirds of the enquiries with limitations. Both case study teachers reported that the devices had a positive impact on their students for learning and continued to use the devices after the research concluded and inspired other teachers in their schools to do the same.

Discussion

Change in educational activity when a technology is introduced can be a gradual process, with the use initially replacing or replicating existing pedagogical practices and beliefs (Tondeur, 2020). The curriculum-based interactions with the voice-activated devices were similar to learning activities carried out without the devices. The children asked direct curriculum questions rather than tangential questions, as they did in a study where teachers were directing the use of voice-activated devices (Dousay & Hall, 2018). The device replaced other tools and technologies for curriculum enquiries such as spelling, word definitions, maths and research questions for social studies and science topics. For example, in case study two the children were familiar with using Google Voice function on their laptops for spelling and transferred this practice to the device. The devices’ web-scraping function for curriculum learning and research activities mirrored question and response interactions that occur when typing into a search engine. The students, like those in the study by Underwood (2017), demonstrated agency in their use of the devices.

The use of the devices was influenced by teacher beliefs, the design of the devices and the social context of peer influence in the classroom.

Teacher beliefs

The use of digital technologies in the classroom is influenced by teachers’ pedagogical beliefs (Ertmer et al., 2012). In this study the devices placed in the classrooms were not specifically integrated into teaching planning and were tools that students could choose to use for learning. However, the teachers’ beliefs and practices did create a difference in perceived utility value and use between the two case studies. Perceived usefulness can be influenced by external factors (F. D. Davis, 1989), and in this case use was mediated by the teachers’ beliefs about effective pedagogical practices.

Design of the devices

Children have been found to anthropomorphise voice-activated intelligent personal devices (Underwood, 2017). The students appeared to sometimes treat the AI device as human-like; they spoke using the same type of manners as they would to a person, such as saying ‘please’, something they are unlikely to use if talking to animals or electronic devices. The attempts to share jokes is also an example of anthropomorphising. The ability to tell jokes is a design feature deliberately built in to improve the social experience of interacting with AI (Duffy, 2003). Some of the enquiries suggest that the students expected a response that was human-like, for example, requesting help to choose a speech topic. However, unlike a human, the device is non-judgemental; this may have contributed to some students’ use and perseverance as the risk of embarrassment is lowered.

Social context

The influence of the two key social indicators in the TAM2 model, subjective norm and image, lessen over time when using a new technology (Venkatesh & Davis, 2000). The case studies captured these forces at their strongest during the first six weeks that the children had access to the devices. For students to treat the device as human, they needed to believe that such behaviour would be accepted or encouraged by their peers (subjective norm) and make them look ‘cool’ (image). These social factors may have influenced the anthropomorphising in case study two where the children explored the personality of the device including its friendships with other devices and food and colour preferences. The device’s responses were often sarcastic, which is a deliberate design feature because ‘a robot who comes across as too intelligent may be perceived as more selfish or too prone to weakness as humans’ (Duffy, 2003, p. 178). Facilitating anthropomorphism may increase the usefulness of technological agents by creating social bonds that increase a sense of social connection within a classroom environment.

Duffy asserts that ‘a robot’s functionality in our physical and social space is clear. It can augment our social space rather than “take over the world”’ (Duffy, 2003, p. 184). However, when the social pressure (subjective norm) of the classroom environment is considered, the power the device holds may be stronger than intended. This influence can be positive for both human–device and human–human interactions. Examples from both case studies show students using manners even when it meant the device was less likely to understand their enquiry. In case study one, in the final week of data gathering, the device was dropped and students asked if it was okay afterwards, suggesting they cared about its well-being beyond checking if it was still working. This may also indicate that over previous weeks some form of familiarity-based empathy had formed. Luckin et al. (2016) noted that the ability to communicate politely with machines as well as humans may be an important employment skill.

Appropriateness

Students’ willingness to see the device as more than a programmed machine comes with its own risks and ethical considerations. Some instances of the device being asked for personal advice and reassurance is concerning as the device response is reliant on algorithms. For example, a student in case study two asked the device for advice on dealing with their parents’ separation. In this case the teachers were checking the audio recordings regularly and following up on any concerning enquires. The ethics and implications of designing educational technology to be anthropomorphised need to be considered along with a critique of algorithms (Perrotta & Selwyn, 2020) and cultural bias, especially when the technology is not specifically designed for educational use (Celik, 2023). Cultural responsiveness is beginning to be built into health-based applications (see Bruno et al., 2018), which could increase the appropriateness of voice-activated educational technologies of the future.

Ethical considerations also exist when introducing devices into the classroom with the knowledge that behaviour will be influenced by subjective norm and image perceptions held by students. In this study it was clear that a technology may be useful and easy to use, but it may not be appropriate for all educational contexts. Technology acceptance models such as the TAM (F. D. Davis, 1989) were made for industry, not education. Age-appropriate responses, understanding indigenous languages used in classrooms and responding to sensitive questions are considerations of appropriateness that should inform whether a technology is acceptable for a classroom context. In educational research ‘appropriateness’ is a fitting construct that should be considered separately to usefulness when deciding whether a technology is acceptable for a setting involving children such as a school.

Having established that the devices can be useful in the classroom context, we now explore implications of the integration of voice-activated artificial intelligence devices into the schooling context.

Oral language

There is potential for pedagogical change in classrooms as educational technologies expand to include voice-activated artificial intelligence. Technology-enhanced learning environments can accommodate different modes of accessing learning (Slemmer, 2002). The use of verbal interactions is a unique feature of the voice-activated intelligent personal assistants. Self-verbalisation that occurs as students ask and reformat questions is an aspect of self-regulation and has been correlated to positive achievement outcomes (Schunk, 1999), and the use of oral language in the classroom can develop children’s thinking (Mercer, 2008). Students in this study illustrated persistence in their oral questioning of the device which may not have been tolerated by a peer or teacher with limited time or patience, although a human may have provided an appropriate response in a much shorter time. The reframing of questions by students to get a response may have a positive effect on their learning.

The introduction of voice-activated responsive devices may further alter classroom experience by placing additional emphasis on oral/aural interactions and literacy, which can be beneficial as learning through discourse has been found to have a positive effect on students’ academic achievement and social development (Mercer, 2008; Premo et al., 2021). The increased opportunity for students to verbalise questions to the virtual ‘teaching assistants’ about process and content could also influence pedagogy as the introduction of multimedia resources did last century (Jewitt, 2002). Students being able to use oral language for their learning when interacting with a device provides opportunities for discipline-specific literacy development (Love, 2009). Interestingly, the children in this study used the devices for literacy-based enquiries more than other subject areas, which may reflect the existing oral aspect of a language curriculum. Voice-responsive technologies appear to be beneficial for children who are language learners (Underwood, 2017) and those whose oral language skills are more advanced than their literacy skills, including younger children. A voice-activated intelligent personal assistant could be a powerful tool for literacy and language development.

As the technology develops educational functionality, it could become a ‘virtual learning assistant’ rather than an ‘intelligent personal assistant’. The device was able to take most requests from students and instantly respond with oral communication matching that of an adult without demanding use of the teachers’ time. However, because the device does not hold information on the students’ previous questions that emerging ‘chat’ devices do, it did not tailor its suggestions to the individual student or scaffold learning in the way that a virtual learning assistant might (R. O. Davis et al., 2021). Ideally, a virtual learning assistant would have functions that enable personalisation of responses during a conversation style of interaction. If the devices could use machine learning to draw together individuals’ past queries, academic and social progressions during interactions, it could become a powerful learning assistant to guide students’ cognitive and social development. This will have potential to change educational practices and the experience that children have in classrooms of the future.

Conclusion

Voice-activated intelligent personal assistants which were designed for the commercial home market were introduced into classroom contexts, and with minimal teacher direction children used them for educational purposes. Generic digital technologies in education settings may be perceived to be useful, but they may not be appropriate for classroom use with children. Appropriateness includes aligning with values of an education system such as inclusion of indigenous languages and being age appropriate. In addition, there needs to be careful consideration of the algorithms in the design to limit bias, assumptions or tensions (Perrotta & Selwyn, 2020) to ensure such technologies are appropriate for the classroom environment. Educators with expertise in curriculum, bias and pedagogy should be involved in the design of voice-activated intelligent devices to ensure they are appropriate for classroom contexts (Fusco et al., 2021).

If voice-activated AI devices are developed specifically with educational functions and appropriate algorithms, they may become intelligent learning assistants that have an impact on classroom practices in the future. However, further research is required into the implications of students creating social bonds with digital agents.

Acknowledgments

The authors would like to acknowledge the children and teachers who willingly participated in this research.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Notes on contributors

Laura Butler

Laura Butler is a PhD candidate with a background as a primary school teacher in New Zealand and the UK. Her research interests include artificial intelligence and student perspectives of digital technology in the classroom.

Louise Starkey

Louise Starkey is interested in the future of education. She is also interested in complexity theory, educational policy and practice associated with teaching and learning in the digital age. Her research includes policy and practice in the schooling and university sectors