The Limits of Adaptive Learning: What can robots teach us?

This paper is a work-in-progress. Last Updated Dec 23, 2019

When we look into the ambiguous essence of technology, we behold the constellation, the stellar course of the mystery…The question concerning technology is the question concerning the constellation in which revealing and concealing, in which the coming to presence of truth, comes to pass. But what help is it to us to look into the constellation of truth? We look into the danger and see the growth of the saving power…Human activity can never directly counter this danger. Human achievement alone can never banish it. But human reflection can ponder the fact that all saving power must be of a higher essence than what is endangered, though at the same time kindred to it.

Martin Heidegger, The Question Concerning Technology (12)

What can and can’t robots teach us?

In the very title of Niel Selwyn’s new book we are provoked with the question, Should Robots Replace Teachers? This fundamental question playing out across the world as technology is more available, diverse, and efficient, and demands for greater and greater educational access are quickly outstripping the available supply. Over the rapid-fire five chapters Selwyn walks readers through the benefits and concerns embedded in the increasing automation of teaching and learning processes in the 21st century. There are many excellent and thought-provoking sections as he explores not only the current uses of technology, but how they are and will influence the culture of education. Throughout the text the resounding concern is how the pedagogical role of educators will be able to perform under the technology conditions now being developed. This is the fundamental question at the heart of this paper: What should and shouldn’t be computer assisted in the 21st century classroom? What are the limits, especially in humanities education, to what the robots can and should do? We will start with a brief overview of the history of artificial intelligence in education, then explore adaptive and personalized learning design, finally looking at how the differences in knowledge domains in the humanities may be a constraint on the ability of current artificial intelligence in education to design for robust content awareness inline with the best practices of humanities pedagogy.

While fully recognizing that certain administrative and some evaluative tasks are better suited to automation, how do we reconcile the limits of Artificial Intelligence in Education (AIED) with the ever-increasing demand from administrators, government officials and the edtech industry to show efficacy and measurable outcomes?

“Human knowledge is emotionally laden insight into the inexorable complexity and ambiguity that attaches to the situations in which human beings find themselves.” (Janik, p. 3) Especially in the study of the humanities (literature, history, philosophy and the so-called soft sciences) the content knowledge is embedded not only situationally but performatively. In order to address this nature Selwyn (2019, p. 126) suggests that,

“we need to create educational environments that allow human teachers to work in the embodied, creative, expressive and relational ways that only human teachers can. We need to develop educational settings that facilitate the expert things that teachers can do and that technology cannot. This implies spaces and times that realize the value of the human embodiment of knowledge, the unique experience of being in the presence of an expert human other, the human modelling of thinking, and the social and affective bases of meaningful learning. These spaces should be collaborative, communal and cooperative. Education can be arranged in ways that allow human teaching to be reinvented as a high-quality, dignified and empowered work – ‘tak[ing] advantage of the uniquely human qualities of creativity, ideation and communication’.

 Neil Selwyn, Should Robots Replace Teachers?

A Brief Introduction to Artificial Intelligence in Education

In the schematic of Roger’s technology adoption curve (21), Artificial Intelligence in education seems to be on the cusp between early adopters and early majority. The difficulty in categorizing the adoption of AIED is that there is a wide variety of applications, some of which have reached late majority (digital grading) and some which are clearly still in the early adopter phase. 

With over 60 years of development, the large scale adoption of computer assisted teaching has grown increasingly ubiquitous within educational communities, you only have to look at the increasing importance placed on computer based assessments by government officials. 

The first teaching machines began emerging in the 1950s but never reached adoption stage. These early first automated machines where not intelligent, but they served the same function as many of the drill-and-practice software used today to reinforce basic concrete skills.  (Selwyn, 2016, p. 75) Designed along behaviorist principles, these reward-based, gamified of computer-assisted learning platforms use feedback loops to incentivize behavior changes. 

In “Artificial Intelligence in Education” by Wayne Holmes, Maya Bialik, and Charles Fadel (2019, p. 82), they discuss the breadth of what AIED encompasses in the modern educational landscape:

“AIED includes everything from AI-driven, step-by-step personalized instructional and dialogue systems through AI-supported exploratory learning, the analysis of student writing, intelligent agents in game-based environments, and student supported chat bots to AI-facilitated student/tutor matching that puts students firmly in control of their own learning. It also includes students interacting one-on-one with computers, whole-school approaches, students using phones outside the classroom and much more…”

Wayne Holmes, Maya Bialik, and Charles Fadel, Artificial Intelligence in Education

With this breadth of applications and underlying design philosophies it is a large project to begin to wrap your head around what the AIED can do, let alone what it can’t. For the purposes of our exploration into the applications within the humanities landscape, the logical place to start is with adaptive learning, since that is a clear place where good pedagogy could meet AIED and hold its own.

Adaptive Learning Theory

 Adaptive learning evolved out of the Skinner’s teaching machines, which were not in any way adaptive to answers, and started its development in the 1950s as a training tool. In these early machines, the questions would shift based on the answers given by the student. These early starts eventually became Computer Assisted Instruction (CAI) and ultimately Intelligent Tutoring Systems (ITS). 

Today ITS are the most common application of AI in education. The limit though, is clear in Holmes (2019, p. 102), as they describe ITS, “Generally speaking, ITS provide step-by-step tutorials, individualized for each student, through topics in well-defined, structured subjects such as mathematics or physics.”

The architecture of ITS can be described as having the following component parts: the domain model (the content to be learned), the pedagogy model (the effective approaches to teaching this content), and the learner model (what is known about this particular learner. The ITS algorithm then draws on these three models to create a sequence of personalized learning activities for each student.

So, what is the problem? What makes the humanities different? There are two aspects we need to explore the pedagogy mode and the domain model.

The Pedagogy of STEM v. Humanities

Forthcoming…

The Knowledge Domains within STEM v. Humanities

Forthcoming…

The Limits of Adaptive Learning for the Humanities

Forthcoming…

The Future Possibilities

Forthcoming…

%d bloggers like this: