Dr. Art Graesser

AutoTutor has been around for some time now. Can you give our newer readers a brief summary on what AutoTutor is and what the benefit is of having AutoTutor?

AutoTutor helps students learn by holding a conversation in natural language. An animated agent (talking head) communicates in speech and rudimentary gestures. AutoTutor keeps track of the knowledge of the learner, but also the learner’s emotions. AutoTutor is very sensitive to this knowledge and emotions. It formulates its actions in a fashion that is sometimes much more sensitive than humans.

AutoTutor helps students learn. AutoTutor improves learning by nearly a letter grade compared to reading a textbook or listening to a lecture, particularly for deep knowledge. Thousands of students have benefited from AutoTutor in the areas of computer literacy, critical scientific thinking, physics, and biology.

How did the idea for AutoTutor come about? What year was it conceptualized and how long did it take to reach a workable format?

In the early 1990s, I led a research team in analyzing how tutoring occurs in school systems. We analyzed the discourse patterns and tutoring strategies of real tutors. I discovered that the strategies that human tutors used were very effective, but they also were very simple. They were simple and systematic enough to convince me that they could be simulated by a computer. In 1996 I wrote a draft of a grant proposal to the National Science Foundation after soliciting feedback on the idea from colleagues in the interdisciplinary Institute for Intelligent Systems (IIS). The proposal was greatly improved by this team effort. In the summer of 1997, I received a telephone call to inform me that the NSF proposal was funded for $900,000. At that point, my life changed. I entered a new world of cutting-edge technology.

In what ways has AutoTutor been used?

AutoTutor has been developed to tutor thousands of students on many topics. Students learn about physics, computer literacy, biology, scientific reasoning, tactical military reasoning, and other topics. These are lengthy dialogues for difficult topics. AutoTutor is not the boring generation of computers of 30 years ago, asking multiple-choice questions and fill in the blank questions. Instead, AutoTutor tries to comprehend the students and responds adaptively. It takes 100 turns sometimes between the student and the computer to answer one difficult question.

Is it an ongoing project? What are the latest steps in the project?

Yes, AutoTutor is alive and well today. We are currently working on several grants from the National Science Foundation and the Institute for Education Sciences – several millions of dollars. We are currently working on two major challenges. First, how can AutoTutor detect the student’s emotions? We know the major emotions of students during learning are confusion, frustration, boredom, flow (deep engagement), delight and surprise. Confusion is the best predictor of learning – a sign of thinking. The computer automatically detects the student’s emotions (as good as human judges) on the basis of conversation patterns, facial expressions, speech signals and body posture. Once AutoTutor detects the emotions through these sensing devices, it intelligently responds with ideal dialogue moves and emotions. If the student is bored, AutoTutor produces engaging razzle-dazzle. If the student is frustrated, AutoTutor gives a good hint. If the student is confused, AutoTutor keeps the student confused for a little bit but not too long. Some versions of AutoTutor are empathetic, polite and supportive. Other versions of AutoTutor are confrontational. We at the U of M (Sidney D’Mello and I) have built the very first automated tutor that detects student emotions and responds emotionally. That’s right – right here in Memphis.

A second challenge is to design a system with two agents interacting with the student – a trialog. One computer agent is a tutor and the other a fellow student. If the human student has low knowledge, AutoTutor shows a good dialogue between a tutoragent and student-agent. If the human student is brilliant, then the human teaches the student-agent (students learn most from teaching) and the tutor-agent chimes in whenever the trialog deteriorates. If the human student is intermediate in knowledge, the tutoragent interacts with the human and the student-agent chimes in when the tutor is uncertain what the human knows. In essence we are designing systems with groups of agents who interact. We have built the very first systems with intelligent adaptive trialogs in natural language.

Is there one goal for AutoTutor for the future or is it an on-going project that will keep developing?

AutoTutor has blossomed into about a dozen projects, currently on more than $10 million in grant funding from the University. We are among three universities in the U.S. that are developing cutting-edge learning technologies with agents. Others are University of Southern California and MIT, who we collaborate with periodically. We are the only university that is systematically developing and testing the cutting-edge agent technologies on learning gains.

AutoTutor’s talking head, Marco, tutors students.

My colleagues in the IIS are taking the helm on these agent projects: Roger Azevedo, Zhiqiang Cai, Scotty Craig, Barry Gholson, Xiangen Hu, Max Louwerse, Danielle McNamara, Andrew Olney, and Vasile Rus. Others in the IIS don’t develop agents but inspire me with ideas: Eugene Buder, Rick Dale, Randy Floyd, Don Franceschetti, Stan Franklin, Iftekharuddin Khan, Roger Kreuz, Trey Martindale, Phil McCarthy, Kim Oller, and Mohammad Yeasin.

Who else at the University has been instrumental in developing AutoTutor?

AutoTutor requires an interdisciplinary effort. We build these with interdisciplinary teams of psychologists, computer scientists, computational linguists, learning scientists, artists, subject matter experts, and so on. Approximately 120 faculty, staff and students have worked on AutoTutor from seven departments. We hold hours of research meetings each week from folks in different departments.

We could never be able to do this without a supportive administration, notably the Provost and the Vice Provost of Research. They understand that cutting-edge research requires interdisciplinary teams that transcend isolated departments. These visionaries are doing their best to convince department chairs to look beyond their own micro-worlds. Interdisciplinary research is currently in vogue in funding agencies.

Have students been involved in the project and if so, in what capacity?

Approximately 100 students have worked on AutoTutor research. These students are graduate and undergraduate students from the psychology, computer science, education, engineering, physics, English, anthropology, art – the list goes on. These students have been coauthor on more than 150 publications and 150 conference presentations. Dozens of doctoral dissertations, masters theses and honors theses have been inspirited by the research with agents.

Has the FedEx Institute of Technology been helpful in further advancing AutoTutor? If so, how?

The Institute for Intelligent Systems (IIS) has been the heart and soul of AutoTutor development. The 20 or so interdisciplinary faculty provide the intellectual foundation. We invite students, faculty and the public to attend our “Cognitive Science” seminar at 4 p.m. on Wednesdays in Room 405 in the FedEx Institute of Technology. This meeting is the intellectual oasis of the IIS, where approximately 50 faculty and students from various departments attend, as well as some members of the public.

Tell us about yourself, your educational background and your research background.

I received my undergraduate degree at Florida State University, where I had a major in psychology and minors in mathematics, linguistics and philosophy. Math and computer science always came easy for me, but I was intrigued with questions about the mind, art and linguistics. This broad background prepared me for the interdisciplinary field of cognitive science, which I pursued in graduate training at University of California at San Diego. Imagine writing programs that simulate mental mechanisms of memory, learning, story comprehension, problem-solving and eye movements. The fusion of computational models and psychology is the essence of what I do. I am most known for my contributions in discourse and language comprehension, inferences, emotions, question asking and answering, tutoring, agents and advanced learning environments.

What is your favorite book and favorite author(s) if you have one?

Alan Lightman’s Einstein’s Dreams, David Macaulay’s The Way Things Work and Don Norman’s Things that Make us Smart and Emotional Design. Regarding fiction, I have more time for films and theatre than novels in print.

What hobbies do you take part in?

I love documentary films. My wife Nancy and I go to Hot Springs, Ark., each year for the Documentary Film Festival in mid-October. I love theatre in New York City and Memphis. Nancy and I travel to NYC two or three times each year to see productions on Broadway and Off Broadway. We have seats with our name in Circuit Playhouse and the new Playhouse on the Square. I love to see and play basketball. Each morning I have an intense 20-30 minute workout of Israeli exercises and full-court basketball in my driveway.

If you enjoy travel, what is your favorite destination, both nationally and internationally?

I have always traveled a great deal and currently go on about 30 professional trips each year, hoping to smuggle in some time for fun and frolic. In the U.S., New York City is my home away from home. I love Europe, particularly driving through the villages of France, Spain, Germany and Eastern Europe, or the Mediterranean coast. Recently, my colleagues Xiangen Hu and Zhiqiang Cai have opened my eyes to China.