78: Intelligent Tutoring Systems (ITS) at the University of Memphis w/ Dr. Arthur Graesser

ABOUT THIS EPISODE

Dr. Arthur Graesser, Professor in the Department of Psychology and the Institute of Intelligent Systems at the University of Memphis joined the podcast to talk about the student benefits of intelligent tutoring systems, his work with AutoTutor, and creating cognitive disequilibrium as part of the learning process.

So people are trying to scale this up in classrooms, in other words, and it turns out the face and the natural language or human computer interaction, those two channels can recover about ninety percent of the learner centered emotions. You're listening to enrollment growth university from Helix Education, the best professional development podcast for higher education leaders looking to grow enrollment at their college or university. Whether you're looking for fresh enrollment growth techniques and strategies or tools and resources, you've come to the right place. Let's get into the show. Welcome back to enrollment growth university, a proud member of the connect Edu podcast network. I'm Ericleson AVP of marketing at Helix Education, and we're here today with Dr Art Gracer, professor in the Department of Psychology and the Institute of Intelligence Systems at the University of Memphis and recent winner of the Harold W mcgarade junior prize and education art. Welcome to the show. I'm very happy to be here. Huge congratulations on your Harold W mcgarade junior prize and I'm so excited to talk with you today about what you were recognized for. Your work with intelligent tutoring systems. Before we dig into that art, can you give the listeners a little bit better understanding of both? The University of meant this and you rule there. Well, I'm part of the Institute for Intelligence Systems, which is very interdisciplinary. In this one center you have people from many different departments, psychology, computer science, engineering, English, physics, the list goes on, and what we do is build things and test them out on people and try to apply our science of learning to these artifacts. It sounds like you've a pretty great GIG and I know you're doing some incredible work. Can you start off this conversation by giving us a definitional level set? What exactly are intelligent ...

...tutoring systems its? Well, intelligent tutoring systems have been created for a few decades, but they're now really taking off with artificial intelligence. So these are systems that track the knowledge and the skills and the strategies and sometimes they did, the emotions of learners at a very fine grain level and then formulates tasks and conversation in ways that will help them learn. And it formulates these using modern advances of artificial intelligence. So you mentioned the evolution of its over the last decade. Specifically summer, wondering if these systems maybe this missing link in our attempt to effectively scale deep learning, because when just reading the book or listening to a lecture aren't enough for some students to be able to grasp higher level concepts, faculty and tutors today can provide very customized and personalized one on one tutoring, but in order to efficiently scale this or provide virtual tutoring for online learners. Talk about the benefits of a system like auto tutor. Well, I'd say there's three benefits. One is there can be subject matters where it's hard to get human tutors, like in our city of Memphis, there's not many people. Are Not enough people to teach physics. Yeah, so these systems can be available to tutor people, the hundreds of thousands, millions of people, and that can be any time. It can be twenty four seven. So that's one advantage. Another advantage is the systems can track people and their psychological characteristics at a very fine grand level. It's often hard for...

...a human tutor to do that. In fact they can't do that very well. And so you know the richness of tracking and the sensitivity on what to do next is very refine tuned with these systems. And and there's just one other thing that's very important is we find that these intelligent tutoring systems, if if they're conversational, they can produce learning gains about the same as human tutors, and which is amazing. Yeah, our can you talk about this concept of cognitive dis equilibrium and why it's something that you are intentionally trying to create for students through auto tutor? Well, it turns out that students, even college students, don't really know how well they're understanding material. In fact, the correlation between their impression of how much they understand and their actual understanding, that correlation is only about point to seven, and Meta analyzes have shown this. So what you have to do is have an interactive system. Yeah, and there's many forms of interactivity, but what can happen is when students interact, they can discover that, hmm, maybe I don't know this as much as I thought I should have, and they can discover the limits of their knowledge. That creates cognitive disequilibrium when there's a gap between how they try to do a task and the performance on that task, or if there's two agents, these computer agents, that disagree over a topic. That conflict puts people in cognitive disequilibrium and then they think about it. It kind of encourages to...

...think at a deeper level and whereas if they were just reading or listening, it just may go by so quickly and they'll just let it go, whereas the interactivity that kind of forces the hand of their knowledge and put some incognitive disequilibrium. It get them to think in deeper waters. In one other point I might say is when people normally read and normally listen, they tend to be cognitive misers and process it at US pretty superficial or shallow level. You have to have some interactive activities to to get them to think more. Yeah, are that dual agent dissonance in autitutors so intriguing to me? So let's talk about auto tutors built an emotional monitoring capabilities that make sure that, with these dual agents, that you're not overly frustrating students when intentionally putting them in situations of confusion and cognitivist equilibrium, and and what the system does when it sends the student frustration. Well, let me say first of all that, there's really for emotions that these systems or a human tutor need to worry about when people learn. One of them is frustration, another is confusion, another one is boredom or disengagement, and another one is when they're in the flow. I mean they're a flow experiences when they're just, you know, thinking at the right rate and pace and learning that you want the flow where people are so absorbed that time and space disappear, and fatigue. But the other emotions just should do something about. Obviously, if they're bored, you've got to kind of shake them up a little bit or give them more demanding exercises.

If they are confused, you of course to want to get them confused, but not overly or chronically confused it. You want it to be selective confusion and in fact you want it to be at the optimal zone of confusion, so it not too much, not too little, just right now. When they get frustrated, there's kind of an interesting interaction because you probably known people who do games. They can be frustrated. However, if they conquer that frustration it's a real high. Yeah, so it depends how interesting the subject matter is is to them. If they're very interested in it, they can tolerate some frustration. However, if it's not, that's not so good. And so you really want to engineer the emotions at a fine grain level, just as you try to have a fine grain engineering on knowledge acquisition. And so, knowing that there is this certain kind of range you want to keep students in, how is auto tutor sensing those different emotions and and how is it kind of tweaking the intelligence in the background to react to it effectively? Okay, well, first let me say a little bit about detecting the emotions. My colleagues and I, especially a colleague, Sydney de Mellow, have actually tried to track many channels of communication. We have had tried to have automatic tracking of natural language interaction and also speech intensity, but also the facial expressions and even the posture are you can learn a lot from the posture how people are sitting or how close their noses to the material. If it's...

...very close, they're more engaged, if it's far away, they're less engaged, and then there's various dynamics. Well, what we've done is compared the different channels and we find that the two most important channels are the natural language interaction and human computer interaction and the timing. That's important, but also the facial expressions. Sometime, and for example, when you're confused, part of your eyebrows go up and part of the forehead wrinkle. Yeah, and that can be detected and actually they have it now. We're pretty cheap. Cameras on computers can be pretty reliable and detecting the confusion. So people are trying to scale this up in classrooms, in other words. And it turns out the face and the natural language or human computer interaction, those two channels can recover about ninety percent of the learner centered emotions. So you get the emotions. The next question is, what do you do? And we have a bunch of these rules on production, rules on what auto tutor does in response. For example, if they are frustrated, in pretty frustrated, you need to give them a good hint and kind of direct them on a better trajectory. If they're bored, your better increase the razzle dazzle or make the that the problems to work on a little bit more challenging. Yeah, or if their performance is poor, maybe it's too hard for them and they're just getting pie eyed. So you you want to gage the difficulty of the material to what they do and their emotions. The other thing is confusion, and that's where you need to kind of track other characteristics of them to see how much...

...confusion they can take. One interesting thing is some students don't know enough about the subject matter to be confused. Yeah, yeah, so that's where you need to try to use these agents or some other part of the learning environment to actually create this you know, contradictions or surprising things that contradict what they believe, or an argumentation back and forth, and so we create that. If we think that they're comprehending the material at a very shallow level, then we try to launch these agents to create some cognitive disequilibrium art. It's incredibly exciting technology. You mentioned a little bit earlier that while its has been around the last thirty years, it's starting to get a lot better where we're seeing comparable results between its and oneonone live human tutors. What are some of the results that you've seen from students who have used auto tutor this far? Well, we've actually tested auto tutor or systems that evolved from auto tutor in terms of learning gains and motivations. And what we find is that the autotutor improves learning compared to reading a textbook by about a letter grade. I could get into the exact quantification but that's not necessary. But that's a pretty good improvement. And you know, if you were to launch auto tutor in some environments, they the students, might really like it, but other categories of students might not like autotutor. And let me explain. The lower ability learners and...

...lower socioeconomic status learner they really like the agents, the conversational agents, a lot. However, for the extremely bright or knowledgeable student and some of the students who are in the older generation, they sometimes don't like the talking heads. So sometimes because they have to wait for it finished speaking. So sometimes you have to take away the talking heads and just have more like a chat interaction, depending on the learners. But we're very encouraged by the fact that many populations really like the agents and it's not only fun and motivating but it also helps them learn. Let me just mention two groups where we've had real success. One is struggling adult readers. There's a lot of adults who don't read at a level that will get them a decent job. And so we have auto tutor to improve their comprehension and they really like these agents quite a bit. And a lot of these struggling adult readers are in college. We've estimated about thirty eight percent of the college students, if at an institution of you know, four year college or university, about thirty eight percent of them don't read at an eighth grade level. Wow, and so the remediation of these tutors like auto tutor could be tremendous. They can go to what some people call a sandbox and other people call a gymnasium, where they can go to this portal and exercise their skills on a variety of skills, not only...

...reading and math but also stem topics, and so that that's one possible future for universities and other institutions, because there are literacy problems, you know, not only in reading and writing but also in new mercy. So if there is one of these virtual gymnasiums or sand boxes where people who are not doing so well can go there and exercise their their skills and strategies and knowledge, I think that systems like auto tutor and other intelligent tutoring systems would be a great that would be a great advance for universities. Are you mentioned a couple of these use cases that you're seeing the most successful right now? Looking forward, what are you the most excited about with auto tutor in similar systems, what you use cases and higher read you see the most potential for? I see a potential of clusters of these intelligent tutoring systems, so it's not just one system but a group of them, because some learners kind of prefer or benefit more from one system than another. So if there's some fluidity to kind of match the right intelligent tutoring system or other learning environment with the student, then that synchrony can be really exciting because you know some just like classes and teachers, sometimes it just doesn't work out and that's where another sort of system can go in there and they may kind of resonate with it more. So I think the big exciting thing is this personalized kind of matching the right learning environment on a topic for for a student, and to do that I think...

...you need a cluster of these intelligent tutoring systems and other learning resources. Are It's so exciting we're finally getting to the future that the movies have talked about four years, thanks to you and your colleagues. Any next steps? Advice for other institutions who are listening to this and considering implementing virtual tutoring on their campuses? Yes, I will recommend knowing which community they should turn to. Like, for example, there's the artificial intelligence and education community, to just look at those systems that are built from that community. And let me mention one other the Department of Defense actually has a site you can go to. That's called Gift Tutoring Dot Org. Gift tutoring to t's their gift tutoring dot Org and there is a series of seven books that talk about intelligent tutoring systems adsea where they've had gee's over a hundred different major researchers in intelligent tutoring systems, right chapters and and talk about the systems that have been built. So if you want to learn about the latest in intelligent tutoring enterprise, I would go to the gift tutoring dot Org. Are thanks so much for your time today. What's the best place for listeners to connect with you if they have any follow up questions. The best is to just google, Google Art Gracer Memphis and you will come to my website and you'll be able to explore what we do. You can, anybody can, at any time go to auto tutor dot org and you'll see example auto tutor systems. I encourage all the listeners to start exploring arts doing some amazing things. Thanks against so much for...

...joining us today. Are Thank you so much, Eric. Thank you so much for allowing me to share my passions over the last forty years. Absolutely thank you. Attracting today's new post traditional learners means adopting new enrollment strategies. Helix educations data driven, enterprise wide approach to enrollment growth is uniquely helping colleges and universities thrive in this new education landscape, and Helix has just published the second edition of their enrollment growth playbook with fifty percent brand new content on how institutions can solve today's most pressing enrollment growth challenges, downloaded today for free at Helix educationcoms playbook. You've been listening to enrollment growth university from Helix Education. To ensure that you never miss an episode, subscribe to the shown itunes or your favorite podcast player. Thank you so much for listening. Until next time,.

In-Stream Audio Search

NEW

Search across all episodes within this podcast

Episodes (264)