Home Adaptive Learning Intelligent Tutoring Systems (a Decades-old Application of AI in Education)

Intelligent Tutoring Systems (a Decades-old Application of AI in Education)

8
SHARE

Artificial Intelligence in Education

While “AIED” may seem relatively new, a deeper look shows us it has been developing for over half a century

In the last few years, numerous developments have led to a growing awareness of the maturity of Artificial Intelligence. Self-driving cars and personal assistants like Alexa and Siri are some of the consumer-facing technologies that have helped to fuel this awareness. This knowledge can also bring with it a certain dystopian fear about robots and technology “taking over”.

While we should always strive to be cautious with new technologies, our concerns should also be tempered by understanding the long curve of development that typically precedes these seemingly overnight maturings of technology.

I've been reading Artificial Intelligence in Education, a 2019 publication by Wayne Holmes, Maya Bialik, and Charles Fadel, that explores implications of AI in the realm of teaching and learning. It has been an enlightening read (and I should add, a very affordable one – so many academic publications are exorbitantly priced, but this one was only $14.95 in paperback form on Amazon).

One of the things that really struck me as I read this book is the long history of applications of artificial intelligence in education, stretching all the way back to work by Pressey, Skinner, and others the 1950s.

I do not intended to summarize the entire book here in a blog post, but rather to provide a brief overview of one specific interesting use of AI that is explored therein: the Intelligent Tutoring System.

Intelligent Tutoring System =
Domain Model + Pedagogy Model + Learner Model

One of the first things we learn in the section about Intelligent Tutoring Systems is that such systems require three vital elements. Combining a specific knowledge domain model with a pedagogy model (representing knowledge of teaching and learning) and then integrating a Learner Model differentiates such systems from the Computer Assisted Instruction tools that preceded Intelligent Tutoring Systems. Learner Models can combine things that are known about the student with information or patterns gleaned via machine learning from other students who have used the system.

Before exploring various specific examples of ITS, there is a brief discussion about studies looking at the efficacy of such tools. One meta study notes:

“Developers of ITSs long ago set out to improve on the success of CAI [Computer Assisted Instruction] tutoring and to match the success of human tutoring. Our results suggest that ITS Developers have already met both of these goals”1

The book goes on to note that when pooling the results of numerous meta studies, ITS appears to have bested the results of whole-classroom teaching, but not (yet) the results of one-to-one teaching. This is impressive and important to be aware of.

Examples of ITS

There are dozens of examples of Intelligent Tutoring Systems provided in the book. Here are a handful of them:

Mathia: Developed at Carnegie Mellon University, Mathia was designed by cognitive scientists to help deliver a successful math experience to each individual student, while “providing you with all the real-time feedback and assessments you need to understand where your students are at and where they're headed.” (p. 109)

ASSISTments: This free mathematics tutoring tool was developed at Worcester Polytechnic Institute and works with curriculum from 2nd through 12th grade. Teachers can use a variety of resources to create assignments for students. Students receive immediate feedback and guidance on assignments through the software. Teachers can then use assessment and analytics to target class time to help address student needs. (Sounds like a great addition to the flipped math classroom). Learn more here: new.assistments.org.

alta: This fully integrated, adaptive learning courseware is a product from Knewton and is aimed at higher education (knewton.com/what-is-alta/). It is intended to work with several different subjects in the domains of math, economics, and chemistry. Alta's pedagogy is based on item response theory and it utilizes a mastery learning approach.

Area9 Lyceum (area9lyceum.com): This unique approach recreates learning content and delivers it on their platform, which uses “continuous self-assessment” (the user's confidence in their responses are used as part of the adaptive process), in addition to other adaptive factors. The solution appears to have become popular in professional/corporate training.

Toppr: This application is based out of India and offers personalized learning across a wide range of grades and subjects. Toppr uses machine learning based on student responses, personalizing questions and adjusts the speed of presentation to optimize the experience for each student.

These look like powerful tools. Where do we go from here?

With the many offerings available, some of which have been around for a decade or more, I am surprised by how rare the use of these tools still seems to be. I have not come across much in the way of discussions or written dialogue about these applications. There appears to be an imbalance between the power and potential of intelligent tutoring systems (i.e. adaptive learning) and the adoption of the same.

This doesn't seem to be a cost issue as some of these platforms are relatively inexpensive (alta is $39 for college students) or even free (ASSISTments).

Is the problem lack of awareness? Reluctance to change? I ask the reader – are you aware of these types of platforms? Do you have experience with them? Do you know of reasons educators and schools have not more aggressively adopted them? (Thanks for any feedback you care to offer!)

———————————————

1Kulik, J.A. & Fletcher J.D. (2015). “Effectiveness of intelligent tutoring systems a meta-analytic review.” Review of Educational Research, 0034654315581420. https://doi.org/10.3102/0034654315581420

8 COMMENTS

  1. Hi Robert –

    Thanks for the feedback. Teachers concerns about these types of tools are certainly one of the reasons adoption has been slow. This is really a shame since tools like these can help to supplement a teacher’s effectiveness. They are by no means a threat – they are a tool to help students learn more effectively, which what it’s all about, right?

  2. Interesting article and provoking question, I was not aware of the resources you have listed but I have been actively looking for how to implement the same in Kenya but with a low tech solution.

    I have recently come across a few but have experienced resistance from teachers which I suspect is because apart from providing a better experience for the student it makes it easier to evaluate the teacher’s performance across the school and even district.

    This implies that the teachers fear of being found out to be ineffective is the reason for the resistance to the implementation of adaptive solutions any other digital solutions.

  3. Hi Kelly, thanks for the review. Much appreciated. In my opinion, one of the reasons that ITS and other AI technologies have not been taken up as widely as one might expect is that they don’t actually always achieve what they claim to achieve. There’s a big credibility gap here. Other than that, I agree wholeheartedly with Mike Sharples’ summary.

  4. Thanks for the excellent, informed feedback Mike. Makes sense. Technology adoption can be a bear, and resistance/reluctance from teachers takes some special handling and embracing to overcome. Seems like such an untapped potential though. It will certainly be interesting to see how this progresses in coming years, both in the US and abroad (and indeed China is a special case as they are investing heavily in AI).

  5. I worked on researching and developing Intelligent Tutoring Systems in the mid 1980s. You ask good questions at the end of your blog. There have been adoptions of ITSs into classrooms, notably the Cognitive Tutors from Carnegie Learning. Comparisons with classroom teaching usually show significant gains in learning outcomes for ITS. The reasons that ITS systems haven’t been more widely adopted are threefold:
    – It can take years to design and implement an effective ITS. The Cognitive Tutors are based on over 30 years of research into modeling teaching and learning, in particular to build computer representations of learners’ cognitive processes and misconceptions.
    – Such formalisation of knowledge can only be done for well-defined domains, that change relatively slowly. For example, it would be hard to build an effective cognitive tutor to teach about the social and environmental effects of climate change, as there isn’t a standard body of knowledge that can be formalised.
    – Adopting an ITS into a school requires investment in both technology and teacher development. This can be a minefield, with school districts unwilling to make the investment in computers for individual learners and teachers resisting technology that they perceive to be threatening their jobs. Individualized learning in classrooms can be effective if managed well, but can also be boring for students and disruptive for teachers.

    In brief, ITS can work for some domains (such as Math, Science and Languages), but needs considerable investment in designing the systems, in teacher development, and in making the case for individualized learning. Schools and Governments usually have other priorities.

    Expect to see more widespread adoption coming from China!

LEAVE A REPLY

Please enter your comment!
Please enter your name here