Content uploaded by Dragan Gasevic
Author content
All content in this area was uploaded by Dragan Gasevic on May 13, 2024
Content may be subject to copyright.
Advancing Theory in the Age of Arficial Intelligence
Shane Dawson ( https://orcid.org/0000-0003-2435-2193 )
Centre for Change and Complexity in Learning,
University of South Australia
Srecko Joksimovic ( https://orcid.org/0000-0001-6999-3547 )
Centre for Change and Complexity in Learning,
University of South Australia
Caitlin Mills ( https://orcid.org/0000-0003-4498-0496 )
Department of Educational Psychology
University of Minnesota
Dragan Gašević ( https://orcid.org/0000-0001-9265-1908 )
Centre for Learning Analytics and Department of Human Centred Computing
Faculty of Information Technology
Monash University
George Siemens ( https://orcid.org/0000-0002-9567-9794 )
Centre for Change and Complexity in Learning,
University of South Australia
Introduction
Arficial Intelligence (AI) is rapidly advancing. Decades of research is now crossing over into
praccal applicaon in society, as evidenced by the rise of generave AI. Educaon has not been
immune to the uptake. For educaon this presents a duality of potenal outcomes. First, AI
brings the potenal to revoluonize teaching methods, assessment and learner engagement.
Second, AI also harbors numerous unancipated longer term impacts on student learning and
the educaon system broadly. For instance, creave thinking and problem solving are crical in
modern complex sengs. When AI begins to serve as an acve partner in sustained social,
creave, and intellectual pursuits, the impacts remain unknown. An over-reliance on AI systems
may result in a decline in many of the traits that make us human. These traits include
self-regulaon, metacognion, goal orientaon, planning, creave brainstorming, and a range
of skills that could be negavely impacted by automaon or machine take over. Effecve
deployment of AI systems in educaon requires theorecal lenses to guide and direct both
research and pracce. In short, theory provides the guard rails to ensure that principles, values,
and trusted constructs guide adopon of AI in educaonal sengs.
The papers in this special secon argue for the cricality of theory in the design, development
and deployment of AI in educaon. In so doing we queson the connued relevance and value
of exisng theories of learning when AI becomes prominent in classrooms. We call for new
frameworks, models and ways of thinking, ones that include the presence of non-human agents
that are more like an acve partner than a simple technology. Theories provide a foundaon for
understanding the complexies of the educaonal process and can inform the design and
development of AI systems that align with educaonal goals and principles. As AI uptake in
educaon increasingly impacts teaching and learning there are quesons on the connued
relevance of exisng learning theories. The integraon of theorecal frameworks into the
development and implementaon of AI-based educaonal systems is essenal for advancing
the field and achieving opmal learning outcomes . Does the adopon of AI in educaon require
modificaons or revisions in how we learn? Or is a complete restructuring required, resulng in
the need for new theories? Importantly, what should theory offer educators when AI is
included? How these quesons are pursued and addressed have important implicaons. If an
exisng theorecal lens, such as cognivism, is used to evaluate the role of AI, theory will need
to guide new cognive tasks and funcons. For example, how do cognive processes such as
coding, memory, and recall, change when declarave knowledge is no longer the primary intent
of educaonal acvies? AI’s ease of access to this type of knowledge makes it less
instruconally important than before. Similarly, do exisng views of social construcvism cover
AI as a social agent in learning? Or do small theorecal lenses, such as community of inquiry,
sufficiently explain the inclusion of AI in classrooms or do they need to be updated or even
enrely rethought?
Theory old and new:
Technology has been heralded as a needed innovaon to improve educaonal pracce. Indeed,
the use and sophiscaon of technologies to support mul-modal learning has advanced
significantly in recent mes, as have insights gained from learning analycs and data science.
Technologies are now highly embedded in educaon and across all levels of schooling, ranging
from early years to higher educaon and professional upskilling. Although there is an
abundance of experimental and exploratory research invesgang the use of these specific
technologies, relavely few studies have leveraged technological advances to directly challenge
or expand on learning and educaon theory. Our understanding of learning theory has
remained relavely constant, despite the rapidly changing educaon context and almost
ubiquitous access to, and availability of, technologies and informaon. This can be seen through
a review of the research literature in the fields of Educaon Technology, Learning Analycs (LA)
or AI in Educaon, where a plethora of works are focused on predicve models, tesng of a
novel technology, or evaluaons of impact. Comparavely, there are far fewer examples of
research interrogang or posing new theories on learning experiences where instructors and
designers integrate and complement the work of arficial agents.
The lack of crical engagement from research in educaon technology, and in parcular LA and
AI, with theory and challenging perspecves may stem from an over reliance on constrained
data sets and tradional research methods (Barmote et al., 2023; Perroa and Selwyn, 2020;
Poquet, et al., 2021). For instance, the majority of LA research findings tend to be derived from
relavely small-scale, or single course studies (Dawson, et al., 2019). As an example in LA,
Dawson and colleagues (2019) draw on Hevner’s research maturity model to demonstrate that
LA has stalled in a cycle of exploratory works. Although such exploratory research is crical for a
field and brings much creavity and rapid trial and tesng, there is also a need to progress
works towards large scale replicaon studies and the establishment of new theory and revision
of exisng theories. However, the praccalies of undertaking AI in educaon and LA research is
funneled more towards the analysis of individual courses in lieu of mul-age and
muldisciplinary data sets derived from tradional methodologies (Arocha, 2021; Dawson, et
al., 2019; Jacobson et al, 2016). The theorecal framing of these works and interpretaon of
findings are oen based on an educaon theory that was conceptualized in a markedly different
era and learning context. Convenonal learning theories were conceived for learning contexts
with limited technology mediaon, let alone the use of advanced technologies and intelligent
agents to the extent that we are seeing today. These technologies have a significant influence
on how courses are designed and delivered alongside instruconal recommendaons. More
crically, they influence the concept of agency and autonomy as recommendaons are oen
made without awareness of the unique developmental needs of each student and without
regard to the longer term impacts of metacognive skills being replaced by automated systems.
Quesons arise regarding the exisng framing of learning theories and whether they remain
applicable in an educaon system that provides for automaon, recommendaons and
intervenon of learning misconcepons or for that maer establishing common standards and
judgment of learning. The lack of alignment between convenonal theories and advanced
technologies mediang new learning experiences points clearly to the need for a revision and
postulaon of new frameworks, views, theories that include these agents as members of the
learning ecosystem.
Papers in this Special Secon:
Papers in this special secon highlight the ways in which technology in general, and AI in
parcular, are transforming learning environments. The papers bring new insights into the
applicaon of AI and educaon technologies to aid learning processes. In so doing, they
collecvely raise quesons about the applicability of tradional learning theories in an
educaon context increasingly powered and influenced by non-human agents.
Jarvela et al (this issue), pose new ways of thinking about the integraon of human and AI
collaboraon in socially shared regulaon (SSRL). The authors explore the intersecon of human
and AI collaboraon in SSRL research. Jarvela and colleagues propose a hybrid model of
human-AI teaming that holds great promise for enhancing learning outcomes. By leveraging the
strengths of both human and AI collaborators, this approach has the potenal to revoluonize
the way we approach teaching and learning, parcularly in the context of online and hybrid
learning environments. The authors provide a comprehensive overview of the theorecal
underpinnings of SSRL and offer praccal suggesons for the design and implementaon of
hybrid human-AI learning systems.
The paper by Saqr (this issue) tackles an exisng challenge for AI and LA research. While the
development of predicve models of academic performance is now a relavely common
pracce, demonstrang improved change remains elusive. As Winne has argued, the predicve
model falsely assumes all students react in similar ways. In this context, Saqr argues that the
convenonal approach of developing predicve models that assume all students respond
similarly is inadequate. Instead, the author proposes incorporang within-person and
between-person variance to establish more accurate and praccal models. By doing so, Saqr's
work provides a promising avenue for improving the effecveness of AI and LA research in
promong student success.
Kio et al’s paper (this issue) tled "Using causal models to bridge the divide between Big Data
and Educaonal Theory" explores the gap between educaonal theory and big data analycs in
educaon. The authors argue that while big data can provide valuable insights into learning
processes, it is limited by its lack of causal reasoning. To bridge this gap, the authors propose
using causal models, which allow for a deeper understanding of the underlying mechanisms
that drive learning. They provide several examples of how causal models can be used to analyze
educaonal data, and argue that such an approach can help to improve the accuracy and
reliability of educaonal analycs.
The paper by McMaster and Kendeou (this issue) starts from an exisng framework that is used
by many schools and school districts to promote evidence-based instruconal pracces -
Mul-Tiered Systems of Support (MTSS). MTSS has the potenal to assist in making precise and
prompt diagnosc and instruconal choices, as well as customizing intervenons for children
who require intensive learning support. To support the effecve and efficient learning in schools
they propose the theory-based integraon of learning analycs and data-driven educaonal
technology into MTSS. The present a use case that demonstrates how (a) MTSS can be used to
guide the use of technology in educaonal processes and (b) how educaonal technology can
be integrated into the MTSS framework.
The paper by Gibson et al. (this issue) synthesizes exisng learning theories and proposes a new
theory to inform the design of AI systems and improve computaonal modeling that can
enhance the role of arficial intelligence (AI) in facilitang learning processes. At the core of the
theory is a causal learning model that is designed to explain how learning occurs across micro,
meso, and macro levels. The authors highlight the “natural” role of learning in elaborated
exploraon and filling of niches in a larger environment via incremental steps and leaps of
progress. The paper emphasizes the importance of big data, compung power, and deep
computaonal learning models in addressing quesons about the roles of AI in society.
Rahm-Skågeby and Rahm (this issue) discuss how AI in educaon can be seen and analysed as
"policies frozen in silicon." The authors suggest that policies exist as both soluons and
representaons of problems. The paper provides a heurisc lens for analyzing and
understanding AI technologies, and how they funcon as proposed soluons to specific
problems based on different ideas about how educaon and learning are enacted. The aim of
this paper is to improve theorecal and analycal approaches in the educaonal system as AI
becomes more prevalent, and to gain insights into how AI will impact how we assess and
measure learning.
In Bearman & Ajjawi (this issue), the authors argue that intenonally framing AI as a ‘black box’
may help students learn to deal with its inherent indeterminacy in an AI-mediated world. The
paper uses relaonal epistemology as a lens to frame AI, highlighng that it should be
understood based on its interacon with humans at a parcular moment in me (as opposed to
how the AI was constructed as a separate enty). The paper gives examples to illustrate their
point, making the case against over focusing on “explainable AI” as a way to understand an
AI-mediated world by arguing that: a) explainability does not necessarily equate to transparency
or understanding, and b) the idea of explainability may require us to assume that such
knowledge is fixed and measurable — both of which the authors suggest might “miss the point”
from a relaonal knowledge view of AI. The paper makes the case for how pedagogy should
focus on what AI and humans do together , including how to orient quality standards within
sociotechnical ensembles, designing rubrics for ambiguity and complexity, and developing
digital literacies for an AI-mediated world.
Hollander et al. (this issue) provide an outlook for how AI and computaonal linguiscs can
guide reading development for diverse populaons of learners. The authors organize a
framework that captures the complexies of reading development including the need to
consider reading as a developmental process that involves a complex set of knowledge, skills,
strategies, and disposions to become “proficient”. The paper provides a helpful literature
review that outlines the state of the literature in terms of foundaonal theory and current
educaonal technologies, and discusses some of the barriers and opportunies that come along
with tradional school organizaon. With future R&D in mind, the authors outline the key
problems, consequences, AI opportunies, and desired outcomes for literacy which will
undoubtedly be a helpful guide as the field progresses in coming years.
Hilbert et al. (this issue) address the need for engagement in STEM learning, with an increased need in
student self regulaon. They build on an extensive history of science of learning research that uses
digital track data to create cognive constructs that provide insight into engagement, social networks,
community, and metacognion. In this paper, they detail the importance of regularity of engagement as
a strong predictor of course outcome and the effects of using a science of learning to learn intervenon
to foster student SRL and ongoing engagement. Their results suggest promising and sustained effects of
this training, raising the need for consideraon of theorecal approaches that integrate behavioral
observaons with cognive constructs in digital educaon.
Beauer et al. (this issue) address a crical area of learning related to feedback. Feedback is central to
guiding student progress and tradional approaches rely primarily on human observaon. With the
development of large language models, and natural language processing in general, new opportunies
exist to offer feedback to learners. They detail how textual arfacts can be enhanced by AI feedback.
They offer a framework to connect feedback processes to adapve student support. As digital learning
grows in importance in educaonal sengs, inclusion of more diverse and mul-modal arfacts will
require a similar updang of theory and constructs to ensure feedback as a driver of overall student
success.
Finally, Giannakos and Cukurova (this issue) explore the potenal of mulmodal learning analycs
to understand collaborave problem-solving processes in educaonal contexts. The authors
highlight the limitaons of relying solely on self-reported data or manual observaon, and argue
that incorporang data from mulple sources (e.g., video recordings, eye-tracking, and
physiological sensors) can provide a more complete picture of how students collaborate and
learn. They also discuss the challenges and opportunies of using machine learning to
automacally extract meaningful paerns from these rich, mulmodal datasets.
CONCLUSION:
The papers in this special secon highlight the transformave potenal of AI in educaon. The
speed and sophiscaon of AI is revoluonizing educaon. This brings numerous noted, yet sll
unrealised, benefits for teaching methods, assessment, and learner engagement. However, the
introducon and speed of deployment of AI in educaon also poses challenges and potenal
negave impacts on student learning and the educaon system as a whole. The deployment of
AI in educaon requires theorecal frameworks to guide and direct both research and pracce.
The lack of crical engagement with theory in educaon technology research will severely
impede progress towards new theories that can more effecvely account for the changes and
complexies of AI integraon into learning experiences. As AI connues to impact teaching and
learning, it raises quesons on the connued relevance of exisng learning theories and the
need for new frameworks, models, and ways of thinking that incorporate non-human agents as
acve partners. Effecve integraon and tesng of new theorecal frameworks into AI-based
educaonal systems is essenal for advancing the field.
References
Arocha, J. F. (2021). Scienfic realism and the issue of variability in behavior. Theory &
Psychology, 31(3), 375-398.
Barmote, K., Howard, S., Gašević, D. (Eds.) (2023). Theory informing and arising from learning
analycs . New York: Springer.
Chen, X., Xie, H., Zou, D., Hwang, G.J. (2020). Applicaon and theory gaps during the rise of
arficial intelligence in educaon. Computers and Educaon: Arficial Intelligence , 1
Dawson, S., Joksimovic, S., Poquet, O., & Siemens, G. (2019). Increasing the impact of learning
analycs. In Proceedings of the 9th internaonal conference on learning analycs & knowledge
(pp. 446-455).
Garrison, D. R. (2016). E-learning in the 21st century: A community of inquiry framework for
research and pracce . Taylor & Francis.
Jacobson, M. J., Kapur, M., & Reimann, P. (2016). Conceptualizing debates in learning and
educaonal research: Toward a complex systems conceptual framework of learning. Educaonal
psychologist, 51(2), 210-218.
Perroa, C., & Selwyn, N. (2020). Deep learning goes to school: Toward a relaonal
understanding of AI in educaon. Learning, Media and Technology , 45(3), 251-269.
Poquet, O., Kio, K., Jovanovic, J., Dawson, S., Siemens, G., & Markauskaite, L. (2021).
Transions through lifelong learning: Implicaons for learning analycs. C omputers and
Educaon: Arficial Intelligence, 2, 100039.
Zawacki-Richter, O. Marín, V.I., Bond, M., Gouverneur, F. (2019). Systemac review of research
on arficial intelligence applicaons in higher educaon – where are the educators?
Internaonal Journal of Educaonal Technology in Higher Educaon , 16 (1),
10.1186/s41239-019-0171-0
A preview of this full-text is provided by Wiley.
Content available from British Journal of Educational Technology
This content is subject to copyright. Terms and conditions apply.