University of Memphis tutor Marco is not a normal teacher.
A student's garbled answer to a question elicits a puzzled
look from Marco, much like it would from most professors.
But that's about the only normal thing about Marco. For this
teacher has no body.
Marco, designed by U of M researchers, "talks" to students
and helps them find answers to their questions through
individualized tutoring sessions
Marco, an innovative computer "tutor" designed by U of M
researchers, has the ability to express emotion, give feedback
and carry on a conversation with students. Dubbed AutoTutor,
this "talking head" simulates the conversation of a human
tutor, facilitating interaction between the learner and the
tutor. Studies show it could eventually have a profound effect
"This is the future of human computer interaction," says
Dr. Arthur C. Graesser, co-director of the Institute for Intelligent
Systems and director of the Center for Applied Psychological
Research at The U of M. "If you project out five or 10 years
from now, we are not going to have keyboards. We are going
to be having conversations with computers, just like people
have face-to-face conversations."
How AutoTutor works is quite simple. The "talking head" produces
a question on a screen. Students answer the question, and
the computer gestures and talks in response to the student's
typed answers. Interaction between the computer and the student
progresses accordingly with Marco cajoling the student into
the proper line of inquiry.
Graesser says AutoTutor will eventually improve and transform
the educational system since it can provide more detailed
and tailored tutoring than humans can sometimes provide.
"One of the problems with our educational system is that
the lectures in classrooms are skewed more toward shallow
knowledge rather than deep knowledge," says Graesser. "This
talking head or tutoring system will assist students at deeper
levels. When you get into deeper levels of knowledge, you
need to have more conversation to clarify things. But this
can be very tedious for a tutor or a human being to do. Certainly
in a classroom a teacher can't be answering hundreds of questions.
You need a computer to do that. That is what AutoTutor can
do. It can give a student more tailored tutoring and deeper
"AutoTutor provides more 'conversational scaffolding' for
deeper levels of knowledge," says Graesser. "For instance,
when your car breaks down, if you have a smart person like
a tutor around who can explain things to you at critical points,
you would do much better at repairing it. The same thing with
AutoTutor. A teacher can't be holding 30 conversations at
the same time, whereas this computer can."
Within the academic world, AutoTutor has other applications.
It can evaluate students' paragraph-length answers to questions
"almost as good as graduate assistants can," says Graesser.
What sets the program apart from other educational software
is its ability to think and hold a conversation. "AutoTutor
asks questions and comprehends the answers students type into
the keyboard, then it evaluates the quality of the students
answers," Graesser says. He adds that U of M researchers are
currently "teaching" AutoTutor to learn from the learning
process it engages.
believes that computers such as AutoTutor will free up
teachers to handle more creative and difficult material
that requires human expertise
of a Brainchild
The AutoTutor program is the brainchild of an interdisciplinary
team effort led by Graesser. It is nearing the end of its
development under a $900,000 National Science Foundation (NSF)
grant begun in 1997 and concluding this August. It began after
The U of M Psychology Department received one of 23 grants
(out of 384 proposals) for funding. Only six institutions
are doing research in this area.
"There's a small club of academic people doing the conversational
agents," explains Graesser, principal investigator for the
proposal. "The MIT Media Lab is one, North Carolina State,
University of Southern California and Northwestern University
are working on it, and then there are people in industry."
Graesser has secured more than $5 million in grants for The
U of M and was the 1999 Eminent Faculty Award winner.
For the past three years, more than 30 U of M people - 10
faculty, 15 graduate students and five undergraduates in psychology,
computer science and education - have worked to design and
build AutoTutor. The Tutoring Research Group includes Drs.
Dipankar Dasgupta, Stan Franklin (BS '59), Max Garzon, Barry
Gholson, Xiangen Hu, Robert Kosma, Roger Kreuz, Bill Marks
and Phillip Wolff, all of The U of M, and Natalie Person (MS
'91, PhD '94) of Rhodes College. Co-director of the institute,
Franklin is considered a leading authority on artificial intelligence
and is also a past Eminent Faculty award winner.
While AutoTutor has been used successfully in the classroom,
it has other uses too. The project has moved into the realm
of conceptual physics as a tutor named WHY2. "We have a $4
million grant from the Office of Naval Research partnering
with the University of Pittsburgh in building this conversation-based
tutoring system that can be used for training," says Graesser.
The five-year WHY2 project started last May and will be finished
Another partnership spurred by AutoTutor has been one formed
with the Institute for Defense Analyses, which includes ThoughtWare
Technologies, a Memphis-based company addressing corporate
knowledge-management needs. Graesser says this partnership
will build a Web facility where military people explore the
ethics of using human subjects in research.
The research has commercial applications as well. Graesser
and U of M researchers have teamed with the Memphis-based
software firm Challenger Corp. to develop an Appendicitis
Diagnostic Assistant to help in the educational and training
needs of more than 340,000 practicing American primary care
and hospital-based physicians. "That agent performed very
well - it exceeded the typical accuracy of doctors in several
formal clinical studies," says Graesser.
AutoTutor and similar programs may have a profound effect
on the educational system. "Future computers will be replacing
teachers who simply promote shallow learning," Graesser says.
"But teachers need not worry about being replaced. I believe
that in the future computers such as AutoTutor will free up
teachers to handle much more creative and difficult material
that requires human expertise. Deep understanding will always
require a human teacher. But conversational computers such
as AutoTutor will make a big difference in the educational
system of the future."