The idea of intelligent tutors is not new, nor is the technology. From the 1940’s, Alan Turing and other early innovators of computing systems envisioned these intelligent machines would be used to teach humans, with personalized learning being a specific pedagogical affordance of the technology (Ferster, 2014, 2017; Shute, 1994; Woolf, 2010). Historically, development and implementation of intelligent tutoring systems (ITS) has aimed towards mimicking or substituting for what has been considered the “gold standard” in education: one-on-one learner-teacher interaction (Baker, 2016; Ferster, 2014, 2017; Roll & Wylie, 2016; Woolf, 2010). This “gold standard” remains a persistent challenge within distance education (DE) contexts, though recent Internet communications technology (ICT) has helped solve some of these challenges, enabling more direct learner-teacher (and learner-learner) interaction (Garrison, Anderson & Archer, 2010; Simonson, Smaldino, Albright, & Zvacek, 2012; Woods & Baker, 2004). How might ITS build on this progress?
An important question that has shaped inquiry regarding the role of ITS and machine intelligence (MI) in education and learning is, “Can machines teach?”, followed closely by the question “Should machines teach?” (Ferster, 2014, 2017). Challenges in answering these questions may stem from the embedded assumption: “Can (should) machines replace teachers?”. If we ignore this hidden question, and deal with the explicit question alone, the answer is conceptually straightforward: Yes, machines can and do (and, within the 21st century context, should) teach. Consider the question from a basic behaviourist perspective: Working one-to-one with an ITS, a student inputs an answer to the question/problem the ITS generates, and receives immediate feedback from the ITS, which leads the student to modify his/her actions (Ferster, 2014, 2017; Laurillard, 2012) In this scenario, the teacher (and teacher intervention) is not eliminated, but rather repositioned, and how the ITS is implemented will impact to where (i.e. what point in the learner-ITS-teacher interaction cycle (Laurillard, 2012).
The important question then becomes: “HOW can/should machines (ITS) teach?” (Baker, 2016; Ferster, 2014, 2017; Roll & Wylie, 2016; Woolf, 2010). This is the question I will attempt to answer in this paper. I will pay particular attention to how ITS might solve some important problems for DE, while simultaneously challenging the idea of “distance” (historically considered to be any separation of learner and teacher) (Simonson, et al., 2012). For instance, can the separation inherent in learner-ITS-interaction actually mediate “distance” and improve immediacy and teacher presence, if the integrated technology and pedagogy afford teacher interventions that reach each learner and are tailored to their individual needs? (Buckreus, 2017; Garrison, et al., 2010; Rizzotto, 2017; Wolf & Baker, 2004). Widespread integration of ITS across learning environments may render the historic distinctions between face-to-face (f2f) and DE contexts immaterial, with “distance” utilized to enhance learning in both contexts.
Intelligent Tutoring Systems (ITS): A Description
The basic concept is simple: ITS is meant to stand-in for direct one-to-one learner-teacher interaction (the “gold standard”), providing proxy teacher intervention through automated immediate feedback and scaffolding that is responsive to the learner’s individual needs, capacity, and pace, as the learner (typically) works independently paradigm (Baker, 2016; Ferster, 2014, 2017; Roll & Wylie, 2016; Woolf, 2010). ITS therefore has the potential to solve two persistent problems arising from the industrial model of education: 1.) Learner-teacher ratios that do not enable consistent one-to-one learner-teacher interaction; 2.) Reliance on a one-way knowledge transfer format (Ferster, 2014; Johannssen, 1999; Mitrovic, n.d.). Both problems shape teacher presence (per the Community of Inquiry framework), and inhibit learning by limiting the frequency and processes of social interaction, which social constructivism and situated cognition describes as the means through which learners construct knowledge (Bredo, 1994; Jonassen, 2011; Perera, 2011; Simonson, 2012; Woolfolk, Winne, & Perry, 2011). Historically, teacher presence and interaction have been particularly challenging problems in distance education (DE) contexts (Garrison, et al, 2010; Simonson, et al., 2012).
The implications of answering “Yes, machines can (should) teach)” become more salient when the goals for contemporary education are considered, along with expectations for changing teacher roles within this new paradigm (Baker, 2016; Ferster, 2014, 2017; Roll & Wylie, 2016; Woolf, 2010). Roll and Wylie (2016) point to focus being resituated from product (i.e. domain knowledge) to process (e.g. metacognition, critical thinking, self-regulation, collaboration, motivation), across diverse formal and informal learning environments that people traverse throughout their lifetime (p. 589). Rather than being the primary source of domain knowledge, teachers take on a supportive role, guiding learners in constructing their own learning (independently and/or collaboratively) through seeking, evaluating, and synthesizing knowledge from diverse sources (Roll & Wylie, 2016; Siemens, 2005). Amidst this new learner-teacher relationship, machines can serve as an intermediary source for accessing domain knowledge, and as supportive tools helping teachers create/design learning experiences (Baker, 2016; Roll-Wylie, 2016).
Implementation has differed in different countries and regions. Kronk (2018) describes that while in North America ITS use has been primary supplemental/supportive to the teacher role, in China ITS (with embedded MI) have taken over the teacher role completed in some cases. The ITS LAIX (pronounced “Likes”), a new Chinese startup for English language learning, is completely teacher-free in its delivery (Kronk, 2018).
Mitrovic (n.d.) describes that ITS applications facilitate communication and knowledge construction, with a high degree of control situated within the hands of the learner. In other words, ITS helps foster personalization and learner-centeredness through:
- Learning tailored to individual learner differences;
- Supporting motivation;
- Supporting self-direction; and,
- Pacing learning with individual learners’ capacity and existing knowledge and skills (Ferster, 2014, 2017; Laurillard, 2012; Reigeluth & Carr-Chellman, 2009).
Mitrovic (n.d. slide 11) identifies some common ITS features the foster learner-centeredness include:
- “Adaptive sequencing
- Adaptive problem generation
- Student diagnosis
- Adaptive feedback generation
- Adaptive problem generation
- Fading of scaffolding
- Adaptive dialogue management”
The one-to-one interaction “gold standard” is clearly reflected in conventional ITS structure, which embeds a student model comprising a qualitative rendering of the individual learner’s persona (behaviour, emotions, existing skills and knowledge), which informs how the ITS applies the domain model and tutoring model (Baker, 2016; Brusilovsku, 1999; Ferster, 2014, 2017; Huang & Chen, 2016; Mitrovic, n.d.; Nye, 2015; Roll & Wylie, 2016):
Figure 1. Conventional ITS structure (Bourdeau & Grandbastien, 2010)
Notes for Figure 1: Domain Model = content knowledge; Student Model = compilation of data regarding student behaviour, existing knowledge and skills, emotions, etc.; Tutoring Model = pedagogical strategies
Figure 1 provides a sense of how these three models intersect and are mutually informing. Ferseter (2017, para. 12) describes that ITS contains “a semantically connected conceptualization of the content to be taught, a way of knowing what the learner does and doesn’t understand, and a delivery method that adapts that instruction accordingly.” Figure 2 depicts this in application:
Figure 2. Typical ITS Framework (modified from Brusilovsky, 1999)
Applications of ITS align well with constructivism and the ideal of personalized learning by leveraging technological affordance for integrating multi-model learning objects and tasks (Perera, 2011; Rizzotto, 2017). For example, Shufti, an ITS developed at the University of Alberta for students learning to identify lesions in radiologic images, capitalizes on kinetic user input to foster the learner’s development of both visual and psychomotor skills (Johnson & Zaiane, 2012, 2017). The learner is presented with a radiological image(s), and moves either their finger or the curser around the image on the screen to identify lesions. A virtual thermometer appearing on-screen adjacent to the radiologic image provides an adaptive visualization of the learner’s identification accuracy: As the student’s finger or cursor gets closer to the lesion in the image, the thermometer’s virtual mercury rises and approaches red on the blue-red spectrum (red = warmer = closer; blue = colder = farther away). The ITS prevents the learner from advancing to the next level in the activity, until the learner has achieved competent application of identification skills. Not only does Shufti support the visual and psychomotor skills needed for working with radiological images, but also enables students to gain experience using a tool that is used in clinical practice (Johnson & Zaiane, 2012, 2017).
Shufti is a an example of technological and pedagogical integration that redefines learning (per Puentedura’s SAMR model) (Puentedura, 2014). In other words, Shufti enables learning that would not be possible without the use of technology. Shufti’s interface allows for authentic kinetic user input, and is responsive to this input in a way that no teacher could, and this would be especially true in a strictly DE context wherein the teacher could not easily observe the minutia of the student’s psychomotor skills as the student is working (Baker, 2016; Johnson & Zaiane, 2012, 2017; Roll & Wylie, 2016). Shufti demonstrates how the technological affordance of learner-machine interaction via ITS can be leveraged to foster constructivist learning.
However, as an application of MI, ITS in general have failed to leverage the affordances of MI for constructivism in human learning, and, as Baker (2016), Ferster (2014, 2017), and Woolf (2010) suggest, have fallen short of their promise to revolutionize education. These failures have little to do with the technological possibilities, and rather stem from limitations in how theory, pedagogy, and imagination have informed both ITS programming and implementation.
Figures 1 and 2 clearly reflect that the conventional ITS structure and implementation framework privileges domain knowledge, rather than focusing on the process learning that Roll & Wylie (2016) describe as a critical goals for 21st century education (discussed below).
In general, focus has been on how ITS can promote learner-centredness and enable personalized learning as solutions to the logistical challenges of one-to-one learner-teacher interaction within the backdrop of the industrial education model (Baker, 2016; Ferster, 2014, 2017; Roll & Wylie, 2016). However, this focus foments a teacher-centred approach, with the embedded assumption that one-to-one learner-teacher interaction is the “gold standard” (Baker, 2016; Roll & Wylie, 2016). As ITS development and implementation move forward, this assumption must be abandoned, so that focus can be resituated to how leaner-ITS interaction might more fully support learning towards the 21st century education goals that Roll-Wylie (2016, p. 508) identifies: Skills in communication, metacognition, critical thinking, self-direction, motivation, and collaboration. This will require changes in how ITS interacts with both teachers and learners, and in how ITS mediates learner-teacher interaction. Roll-Wylie (2016) asks why would we limit ITS use to mimicking human teachers, when ITS can do in learning contexts what humans cannot do?
It’s time for intelligent tutors to live up to their promised potential
As education shifts to become a process of lifelong learning (across formal, informal, and workplace learning environments), a foreseeable trajectory is that ITS will become more integrated with the other connected systems they become ubiquitous and ambient (Roll & Wylie, 2016; Rizotti, 2017). We have begun to witness the widespread presence of both rudimentary and sophisticated MI in our everyday interactions mediated by the Internet, and this will expand in concert with The Internet of Things.
ITS and MI have historically been intertwined, though to-date ITS has been slow to leverage the affordances that MI offers. However, the 2018 New Media Consortium Horizon Report for Higher Education (New Media Consortium, 2018) estimates a two-to-three year time to adoption for adaptive learning technologies and MI, linking this to expanding capacities for personalized learning. Complementing this, training targets around the world aims towards producing skilled MI programmers, as well as supporting educators in gaining MI competencies (Baker, 2016; Crowe, LaPierre & Kebritchi, 2017; Nye, 2015). The time is right, and the opportunity is here, for ITS to be re-imagined.
New Education Goals and Pedagogical Competencies for New Learning Environments
Roll and Wylie (2016) assert that the following three areas should inform ITS development, and ITS systems and pedagogical integration:
- Educational goals –
To address the needs of the 21st century learners and our global society, learning outcomes must transition from domain knowledge to adaptive process skills: Communication, critical thinking, metacognition, motivation, self-direction, and collaboration (Roll and Wylie, 2016, p. 589). New learning taxonomies and assessment strategies must also be developed and implemented, to effectively evaluate these process skills, and to better enable assessment outside formal education environments (Roll and Wiley, 2016).
For instance, since much domain knowledge resides in databases and is instantly accessible through connected devices (including mobile devices), learning and assessment must support learners in developing skills for finding, evaluating, synthesizing, sharing, and using information when needed, rather than the acquisition of factual knowledge (Roll & Wylie, 2016; Siemens, 2005).
- Teacher/Facilitator (Classroom) Practices –
Teacher competencies and education research must focus on how technology influences pedagogy, in concert with 21st century learning goals, including what aspects of teaching may be replaced by technology. Pedagogy, across learning contexts, must place greater emphasis on experiential learning, focusing on authentic learning content (e.g. everyday tasks and challenges), settings (e.g. at home, in the workplace, in public spaces), and manner (learner and peer actions and interaction) (Roll & Wylie, 2016).
In conjunction with this, Roll and Wylie (2016) describe that the complexity of learning activities and assignments must also increase, to include problem-based learning, collective knowledge construction, information seeking, and technology/information literacy. This shift will require teachers to attain new pedagogical competencies, including behavioural competencies such as facilitation (rather than domain knowledge transfer), interpersonal skills (e.g. emotion-based responses), intuition, and collaborative skills for working as part of instructional design teams or working collaboratively with students (Mitrovic, n.d.; Roll & Wylie, 2016; Siemens, 2005).
- Environments –
Roll and Wylie (2016) describe two dimensions relating to learning environments: First is the technology ecosystem, which embeds multiple spaces (contexts) that both learners and teachers move across in their everyday lives. Pedagogy must shape learning to parallel “organic” technology-mediated processes, fostering digital literacy that is relevant across the ecosystem.
Second is technological affordance for context awareness and novel forms of (collaborative) learning engagement (Roll & Wulie, 2016). As an example, Roll & Wylie (2016) describes mobile devices sensors (such as GPS) and novel input devices device (such as cameras), which can foster direct interaction among learners as they move within distal contexts; In this example, Roll and Wylie describes that an ITS could analyse how learners interact and how devices interact, and either directly adapt tasks to foster collaborative learning, or report to the teacher who would then provide design intervention (Baker, 2016; Roll-Wylie, 2016, p. 594).
A takeaway from the above is that social constructivist is enhanced through support/guidance, and that appropriate support/guidance for 21st century learning entails a partnership between teacher and machine (Baker, 2016; Roll & Wylie, 2016).
Compared to traditional f2f contexts, distance learning has historically required a high degree of learner independence, due to the inherent separation between learner and teacher (Simonson, et al., 2012). Within contemporary distance learning contexts, the process skills that Roll and Wylie (2016) identify, and technology literacy in general, are iteratively constructed as learners engage in learning activities, synchronously and asynchronously, independently and collaboratively. Crowe, LaPierre and Kebritchi (2017) point to both curriculum and administrative teacher/developer task support as an important benefit that ITS may offer for DE contexts.
Moving Practice Forward: Big Data & ITS Reporting to Leverage Human Intelligence
Baker (2016) argues that intelligent machines should not be the end-goal of technological advancement in education:
“Perhaps instead what we need, what we are already developing, is stupid tutoring system. Tutors that do not, themselves, behave very intelligently. But tutors that are designed intelligently, and that leverage human intelligence.” (p. 603)
In other words, Baker envisions intelligence amplification, whereby human decision-making is informed by machine-mediated data mining and learning analytics (Baker, 2016; Siemens, 2013; Siemens & Baker, 2012; Sin & Muthu, 2015). The idea is that, rather than designing sophisticated machines that embed complex student modeling that automated personalized learning (see: Figure 1), the machines are instead designed as tools for collecting and reporting complex data to humans for analysis to inform instructional design and teacher intervention.
Baker (2016) notes that ITS have unique affordances that could facilitate this approach, within both f2f and DE contexts, such as the ability to report to multiple human decision-makers simultaneously; and the ability to report to teachers in real-time, so that teachers can immediately implement interventions.
One implication is that ITS development focus on system design that aids human decision-makers identify when intervention is needed; what intervention is needed; how to implement the intervention; and why the intervention works (Baker, 2016). Designing systems that capture the richest possible data from learners, to inform teachers, instructional designers, etc., would be one aspect of this. The critical point here is that decision-making is situated within the hands of a human, rather than being an automated solution determined by the ITS (Baker, 2016).
Baker (2016) points to the ASSISTments ITS as an example where this iterative approach has been successful in supporting personalized learning and learner achievement. Teachers receive daily reports summarizing data drawn from student work during the previous day (such as frequent A/B tests, which Baker describes serve as mini randomized control trials), and use this data to tailor subsequent learning content and activities (Baker, 2016).
The impetus for exploring a different trajectory for ITS development and implementation stems from a number of observations of current ITS limitations:
- MI in facilitating learner-ITS interaction has not been as robust as anticipated;
- ITS have not been effectively trialed in semantically complex subjects, such as English such as English or history;
- ITS privilege domain knowledge;
- ITS do not yet integrate easily with learning management systems, because the programming requires specific technical expertise;
- ITS development is labour-intensive; and,
- Automated systems (and interventions) tend to be brittle (Baker, 2016; Ferster, 2014, 2017; Nye, 2015; Roll-Wylie, 2016).
Roll-Wylie (2016) describes that:
“Presently, [ITS] developers develop their own content. One rare exception is ASSISTments, which uses homework assignments from existing textbooks. However, this is very labour intensive. In addition, this effort is decontextualized by nature and harder to adapt and adopt. Instead, we suggest to build [ITS] that operates as a shell or an envelope for existing learning objects.” (p.594)
Roll-Wylie (2016) suggests these existing learning objects might include open content from MOOCs, Wikipedia, open educational resources (OER) repositories, browser add-on, and open assessment tools. Baker (2016) contends this approach might help with system scalability, something that has been problematic for systems embedding complex structural models.
Both system complexity and lack of openness are aspects that Baker (2016) describe contribute to automated ITS being brittle. Baker describes that:
“[A]n automated system cannot recognize when a model is clearly wrong; and if an automated intervention is not working, its difficult for the system to recognize and correct for this.” (p.607)
“[S]tudents change over time; automated systems need to be checked and revised over time.” (p. 608)
Baker (2016) points out that humans are flexible, however, and suggests that ITS be designed to foster a human-machine partnership leveraging the affordances of each: “What humans cannot do is sift through large amounts of data quickly, but, once informed, can respond quickly” (p. 608).
How Can (Should) ITS Teach? New Pedagogies for New Types of Interaction
There is a critical disconnect between ITS development/implementation and contemporary learning theory. Although existing ITS has been effective in supporting personalized learning, acquisition of domain knowledge has remained privileged as a primary learning outcomes, and this is clearly reflected in the conventional ITS model-based structure (Figure 1) and application framework (Figure 2). This structure is consistent with constructionism, in which knowledge is individually constructed through a learner’s interaction with their prior experience, and with constructivism, in which knowledge is individually constructed through a learner’s interaction with their environment (Jonassen, 1999; Perera, 2011). However, in focusing on the individual, ITS structures to-date have been inconsistent with social constructivism, in which learning is a collaborative process wherein individuals construct knowledge through interacting with others’ experience as part of the learning environment; and with situated cognition, in which knowledge is linked to action and specific context that is constituted through interaction with others (Bredo, 1994; Jonassen, 1999; Perera, 2011). It is collaborative environments that both Baker (2016) and Roll-Wylie (2016) envision for the future of ITS for both learners and teachers, to address the process learning that is critical for the goals of 21st century education (Baker, 2016; Roll-Wylie, 2016).
Laurillard (2012) advocates for teaching to be approached from the perspective of design science, an approach that is particularly appropriate for technology-mediated learning environments. The idea is for a teacher-design team to design learning experiences (leveraging the affordances of media/technology, matched to task and learning outcome) that enable the iterative cycles of interaction and knowledge construction/modification depicted in Figure 3, with teacher intervention inserted at regular points in the cycle. Although learner-teacher interaction is not direct, teacher presence nonetheless remains present. Of note, this framework depicts interaction with other learners and interaction with technology.
Figure 3. Conventional ITS Framework Mapped to Laurillard’s (2012) Conversational Framework
The ITS framework depicted in figure 3 centralizes interaction in a way that is consistent with both social constructivism and situated cognition, embedding the processes skills that are relevant to 21st century education and lifelong learning (Bredo, 1994; Jonassen, 1999; Perera, 2011; Roll & Wylie, 2016). However, in Laurillard’s (2012) Conversation Framework, knowledge is constructed through interaction with both others and with machines, wherein machines contribute to the creation of context/environment when they are incorporated into the learning experience. In other words, machines become part of the “social” environment and contributes to learners’ collective and individual knowledge construction.
Approaching ITS development and implementation from this perspective may be particularly important for DE contexts, wherein immediacy and teacher presence might be enhanced through aspects of ITS design premised on the iterative cycles of interaction and knowledge construction depicted in figure 3: Learner-teacher and learner-learner interaction may not need to be direct, if learner-ITS interaction is able to provide a collective environment comprising all (Garrison, et al., 2010; Laurillard, 2012; Simonson, et al., 2012). Distance, therefore, may not be the barrier it once was in DE, and may actually represent an enhancement that f2f contexts adopt through implementing ITS, if richer environments are able to be created through the use of technology.
Another shortcoming of conventional ITS structure is that most application frameworks privilege technical components, and omit the collaborative teacher/design team component, which I have added to Figure 2. The addition of this teacher/design team component to Figure 2 is a simple way to help developers re-imagine ITS possibilities for teaching and learning, providing a vision of where human decision-makers might be integrated within learner-ITS interaction, in line with Baker (2016) and Roll and Wylie (2016). Baker (2016) goes so far as to envision ITS that embeds automated supports for teachers, aggregating best practice guidelines or crowd-sourced recommendations, and flagging these to teachers when deviation is detected.
Some Cautionary Reflections
Britain’s industrial revolution between the 1700s and 1900s saw technological and industrial advancements that rapidly moved human development (economic and otherwise) forward, while at the same time creating new dimensions of social disenfranchisement and human suffering (Industrial Revolution, 2018). Advances in MI have brought us to the precipice of our next techno-industrial leap promising revolutionary changes to the human condition, but with perhaps even greater potential for human tragedy (as technology invades further into private, personal spheres) if we are not mindful and purposeful as we proceed.
Threats to Personal Freedoms
Our initial experiences with the scale and spread of Internet-based ventures powered by big data and MI have provided society with some experiential learning on the kind of threats to privacy that big data poses. With the student model continuing as a critical programming structure enabling ITS MI to support personalized learning, and with big data feeding into this, ITS will have the capacity to embed an increasingly detailed psychological profile of the student, including reflections of the student’s private thoughts (Pardo & Siemens, 2014; Rizzotto, 2017; Slade & Prinsloo, 2013).
Aside from obvious issues regarding data security, protection of and access to personal information, and how the use of personal information may impact the person, important issues relating to new types of learner-teacher interaction may arise from ITS MI advancement. For instance, what potential is there for the creation of (new) hierarchy within the learner-teacher relationship, privileging the teacher who has access to the student’s detailed psychological profile? (Buckreus, 2018). And how might this hierarchy influence student engagement in terms of social presence? For instance, how might an environment of surveillance influence student motivation?
Haridy (2018) describes how social presence is impacted in one f2f Chinese classroom where an MI-enabled device is used for behavior management; The device scans the room every 30 seconds, records and provides metrics to the teacher, which may be used to inform intervention. One student from this classroom describes his experience:
“Previously when I had classes that I didn’t like very much, I would be lazy and maybe take a nap on the desk or flick through other textbooks. But I don’t dare be distracted since the cameras were installed in the classrooms. It’s like a pair of mystery eyes are constantly watching me.” (Haridy, 2018, para. 5)
It is easy to envision how surveillance strategies like this could support teacher responsiveness and personalized learning overall, especially in distance learning contexts. However, the student quote above reminds us that, as MI development and implementation proceeds, we must be mindful of potential costs that may end up counteracting our benevolent purpose (Pardo & Siemens, 2014; Rizzotto, 2017; Slade & Prinsloo, 2013). The example described here demonstrates how ITS MI privileges extrinsic motivation, and has encouraged this student towards less-than-authentic social presence (Haridy, 2018; Garrison, et al., 2010).
Threats to Meta-skills and Metacognition
Siemens’ (2005) connectivism framework emerged to account for learning relating to new meta-skills that are relevant to our current and future information-based global society. These meta-skills include:
- “delineating patterns and connections within a mass of technology-mediated knowledge that is rapidly changing/increasing and only tenuously under the learner’s control;
- evaluating the value of content (i.e. whether information/knowledge is worth being learned); and
- determining when and what knowledge should be retired and replaced with updated knowledge (the meta skill of unlearning obsolete knowledge).” (Buckreus, 2014, para. 3)
As described above, the goals of education overall are shifting towards communication, critical thinking, self-regulation, and metacognition, all of which align with connectivism (Roll-Wylie, 2016; Siemens, 2005). One support that MI offers is the ability to scrub the internet to aggregate and tailor content to match individual learner needs, a time-saving affordance for both learner and teacher. (This integrated capacity could be especially supportive for DE learners, who may not have easy access to traditional institutional supports, such as subject matter librarians.)
An important meta-skill for our current and future information-based society – across learning, workplace, and personal contexts – will be learning how to use MI-enabled machines to perform these tasks for you (Siemens, 2005; Rizotto, 2017). However, fostering development of this meta-skill in the learner, without attending also to the development of the curatorial and evaluative meta-skills that Siemens identifies, may pose a risk to metacognitive development by situating too much of the strategic decision-making in the virtual hands of a machine. If the AI revolution is to position us, as members of society, to take on more strategic roles vis-à-vis machines, as Rizzotto (2017) describes, then one scenario might be an ITS MI that scrubs digital content and reports options to the learner, but with decision-making left to the learner regarding what content to select to best match their task-at-hand and learning goals (a process that itself could become a loci for skills assessment).
Threats to Digital and Educational Equity
DE, and mLearning in particular, have supported critical education gains in developing nations/regions, and ITS MI offer great potential for this trajectory continuing, with ‘the right’ implementation that includes design informed by diverse user contexts, with flexible and modifiable platforms (Nye, 2016).
In a recent systematic review, Nye (2015) reported that the majority of ITS research has focused on developed regions (Europe and North America) and on computer-based interfaces. Research bias, such as this, informing ITS development and privileging the already privileged, could neutralize the gains toward equity that DE and mLearning have been supporting (Nye, 2015). Not only could developing regions be excluded from access to important technological advances, but also from the 21st century learning these technologies support (Nye, 2015; Roll & Wylie, 2016). Broader ramifications could include reducing developing regions’ hard-won competitiveness within the global economy. A caveat to this is that technological and socioeconomic inequities persist even within developed regions, which DE and mLearning could have a role in addressing.
ITS research and development must proceed, informed by equity as a constitutive goal. In doing so, not only might worldwide digital and educational inequity be addressed (for common socioeconomic benefit), but technological and pedagogical innovations may be discovered that are relevant for learners in developed and developing regions alike (Nye, 2015).
The future of ITS will move towards more integrated development, offering new affordances for both learner and teacher, and leveraging new types of technology-mediated relationships that enhance immediacy and teacher presence to support learning. Future ITS development will need to address the process skills relevant to 21st century education, within more personalized learning contexts. Learning analytics and big data will aid in this personalization, and will influence teacher roles relative to ITS, supporting teachers in decision-making by collecting and reporting on data that teachers otherwise would not be able to access efficiently.
Though conventional ITS structure and implementation to-date has largely focused on constructivist learning (conceiving of knowledge construction centred on the individual), changing to target the process skills comprising 21st century education goals will move ITS development more towards social interaction and situated cognition, which emphasize interaction as the means for knowledge construction.
New technologies that better foster interaction and create richer experiential learning environments will help to mitigate the impact of separation between learner and teacher in DE contexts, rending the distinctions between f2f and DE contexts immaterial.
Baker, R. S. (2016). Stupid tutoring systems, intelligent humans. International Journal of Artificial Intelligence in Education, 26(2), 600-614. https://doi.org/10.1007/s40593-016-0105-0
Bourdeau, J., & Grandbastien, M. (2010). Modeling tutoring knowledge. Advances in intelligent tutoring systems (pp. 123-143). Springer, Berlin, Heidelberg.
Brusilovsky, P. (1999). Adaptive and intelligent technologies for web-based education. Ki, 13(4), 19-25. Retrieved from https://www.researchgate.net/publication/2493073_Adaptive_and_Intelligent_Tec hnologies_for_Web-based_Education
Buckreus, K. (2014, May 14). Connectivism: Informing distance education theory, from https://kelliralph.wordpress.com/2014/05/24/connectivism-informing-distance-education-theory-pedagogy-and-research/
Buckreus, K. (2018, April 23). The changing discourse of distance: How tripartite learner-AI-teacher interaction may leverage distance to affect immediacy in learning interactions.
Crowe, D., LaPierre, M., & Kebritchi, M. (2017). Knowledge based artificial augmentation intelligence technology: Next step in academic instructional tools for distance learning. TechTrends, 61(5), 494-506. https://doi.org/10.1007/s11528-017-0210-4
Desmarais, M. C., & Baker, R. S. (2012). A review of recent advances in learner and skill modeling in intelligent learning environments. User Modeling and User-Adapted Interaction, 22(1-2), 9-38. https://doi.org/10.1007/s11257-011-9106-8
Ferster, B. (2014). Teaching machines: Learning from the intersection of education and technology. Baltomire, MD: John Hopkins University Press.
Ferster, B. (2017, January 17). Intelligent Tutoring Systems: What Happened? [Weblog]. Retrieved from http://www.stagetools.com/bill/intelligent-tutoring-systems- what-happened/
Garrison, D. R., Anderson, T., & Archer, W. (2010). The first decade of the Community of Inquiry framework: A retrospective. The Internet and Higher Education, 13(1), 5-9. https://doi.org/10.1016/j.iheduc.2009.10.003
Haridy, R. (2018, May 28). AI in schools: China’s massive and unprecedented education experiment. News Atlas. Retrieved from https://newatlas.com/china-ai-education-schools-facial-recognition/54786/
Huang, J., & Chen, Z. (2016). The research and design of web-based intelligent tutoring system. International Journal of Multimedia and Ubiquitous Engineering, 11(6), 337-348. Retrieved from https://pdfs.semanticscholar.org/7f5f/53386328f92df03d312d3663830b169a0afe. pdf
Industrial Revolution. (2018, November 10). In Wikipedia. Retrieved from https://en.wikipedia.org/wiki/Industrial_Revolution
Johnson, S., & Zaiane, O. (2012). Intelligent feedback polarity and timing selection in the Shufti Intelligent Tutoring System. In International Conference on Computers in Education. Retrieved from https://www.semanticscholar.org/paper/Intelligent-feedback-polarity-and-timing-selection-Johnson- Zaiane/e6042a709ddffd5fddaa927a28a74d41275332ff
Johnson, S., & Zaiane, O. R. (2017). Learning to Analyze Medical Images: A Smart Adaptive Learning Environment for an Ill-Defined Domain. In Innovations in Smart Learning (pp. 103-112). Springer, Singapore.
Jonassen, D. (1999). Designing constructivist learning environments. In C.M. Reigeluth (Ed.) Instructional-Design Theories and Models, Vol. II (pp. 215-240). New Jersey: Lawrence Erlbaum.
Kronk, H. (2018, October 8). Chinese edtech has no reluctance with AI. eLearning Inside. Retieved from https://news.elearninginside.com/chinese-edtech-has-no-reluctance-with-ai/
Laurillard, D. (2012). Teaching as a design science: Building pedagogical patterns for learning and technology. New York: Routledge (Taylor & Francis Group).
MacLellan, C. J., Harpstead, E., Wiese, E. S., Zou, M., Matsuda, N., Aleven, V., & Koedinger, K. R. (2015, January). Authoring tutors with complex solutions: a comparative analysis of example tracing and simstudent. In Workshops at the 17th International Conference on Artificial Intelligence in Education AIED (pp. 35- 44).
MacLellan, C. J. (2017). Computational models of human learning: Applications for tutor development, behavior prediction, and theory testing (Doctoral dissertation, Carnegie Mellon University).
Mitrovic, T. (n.d.). Current trends in intelligent educational systems [Slides]. Retrieved from http://www.info2.uqam.ca/~nkambou_r/DIC9340/seances/seance1/intelligent-tutors.pdf (NOTE: LINK IS NOW BROKEN)
Nye, B. D. (2015). Intelligent tutoring systems by and for the developing world: A review of trends and approaches for educational technology in a global context. International Journal of Artificial Intelligence in Education, 25(2), 177-203. https://doi.org/10.1007/s40593-014-0028-6
Pardo, A., & Siemens, G. (2014). Ethical and privacy principles for learning analytics. British Journal of Educational Technology, 45(3), 438-450. https://doi.org/10.1111/bjet.12152
Perera, N. C. (2011, January 31). Constructivism, social constructivism and situated cognition: A sliding scale [Weblog]. Retrieved from https://nishancperera.com/2011/01/31/constructivism-social-constructivism- and-situated-cognition-a-sliding-scale-by-nishan-perera/
Puentedura, R. R. (2014 January). SAMR: A Brief Contextualized Introduction [slides].
Retrieved August 31, 2017, from
Reigeluth, C. M., & Carr-Chellman, A. A. (2009). Understanding instructional theory. In C.M. Reigeluth & A.A. Carr-Chellman (Eds.) Instructional-Design Theories and Models: Building a Common Knowledge Base, Vol. II (pp. 3-26). New York: Taylor and Francis.
Rizzotto, L. (2017, June 23). The future of education: How A.I. and immersive tech will reshape learning forever. Medium. Retrieved from https://medium.com/futurepi/a- vision-for-education-and-its-immersive-a-i-driven-future-b5a9d34ce26d
Roll, I., & Wylie, R. (2016). Evolution and revolution in artificial intelligence in education. International Journal of Artificial Intelligence in Education, 26(2), 582-599. https://doi.org/10.1007/s40593-016-0110-3
Siemens, G. (2005). Connectivism: A learning theory for the digital age. International
Journal of Instructional Technology and Distance Learning, 2(1). Retrieved from http://www.itdl.org/Journal/Jan_05/article01.htm
Siemens, G. (2013). Learning analytics: The emergence of a discipline. American Behavioral Scientist, 57(10), 1380-1400. https://doi.org/10.1177/0002764213498851
Siemens, G., & Baker, R. S. J. D. (2012). Learning analytics and educational data mining: Towards communication and collaboration. Proceedings of the 2nd International Conference on Learning Analytics and Knowledge, New York, NY: ACM.
Simonson, M., Smaldino, S., Albright, M., & Zvacek, S. (2012). Teaching and learning at a distance: Foundations of distance education (5th ed.). Boston, MA: Pearson.
Sin, K., & Muthu, L. (2015). Application of big data in education data mining and learning analytics: A literature review. ICTACT Journal on soft computing, 5(4).
Slade, S., & Prinsloo, P. (2013). Learning analytics: Ethical issues and dilemmas. American Behavioral Scientist, 57(10), 1510-1529. https://doi.org/10.1177/0002764213479366
Westerfield, G., Mitrovic, A., & Billinghurst, M. (2015). Intelligent augmented reality training for motherboard assembly. International Journal of Artificial Intelligence in Education, 25(1), 157-172. Retrieved from https://eric.ed.gov/?id=EJ1049515
Woods, R. H., & Baker, J. D. (2004). Interaction and immediacy in online learning. The International Review of Research in Open and Distributed Learning, 5(2). Retrieved from http://www.irrodl.org/index.php/irrodl/article/view/186/268
Woolf, B. P. (2010). Building intelligent interactive tutors: Student-centered strategies for revolutionizing e-learning. Burlington, Massachusetts: Morgan Kaufmann.
Woolfolk, A., Winne, P. H., & Perry, N. E. (2011). Education psychology (5th Canadian ed.). Newmarket, ON: Pearson Education Canada.
 Throughout this paper, I use the term “machine intelligence (MI)” instead of “artificial intelligence (AI)”, as MI is a more apt descriptor of intelligence type.
 The framework becomes more complicated with specific approaches to modeling, such as constraint-based modelling and other types of tutor modelling (Bourdeau, & Grandbastien, 2010; Brusilovsky, 1999; Desmarais & Baker, 2012; Huang & Chen, 2016; MacLellan, 2017; MacLellan, et al., 2015; Westerfield, Mitrovic, & Billinghurst, M, 2015). But this is a topic for another paper!
 Shufti2, the second iteration if this ITS, has been implemented in clinical settings, and is utilized for clinical diagnostics by radiologic imaging technicians and radiologists (Johnson & Zaiane, 2012, 2017)
 Substitution -> Augmentation -> Modification -> Redefinition
 Roll and Wylie (2016) notes that much of the ITS research conducted to-date has focused on domain-level learning, whereas research on motivation has been limited to measuring learner satisfaction, rather than other aspects of motivation such as self-efficacy.
 Including the ability to evaluate knowledge half-life (Siemens, 2005).
 Per the Community of Inquiry framework, referring to the authenticity with which individuals present themselves within online contexts (Garrison, et al., 2010).
 e.g. First Nations and Inuit communities in Canada.