|Talk About Teaching and Learning
January 22, 2013,
Volume 59, No. 18
In spite of my being a computer scientist, technology has had remarkably little impact on my teaching. A time traveller from the ’60s might notice I have replaced overhead transparencies with PowerPoint slides, but with no effect on teaching or learning. Last semester, however, inspired by the surge of press about online courses (“they will revolutionize teaching”) and by high profile articles in Science and Nature claiming that active learning—where students help each other solve problems in class—significantly improves learning outcomes, I changed how I taught. I was surprised by much of what I discovered.
I made three major changes in my Artificial Intelligence class of roughly 80 seniors and masters students.
- I used clickers to get in-class feedback.
- I made class more like a recitation, having students solve problems in small groups.
- I recorded videos for students to watch outside of class.
These changes are all tightly coupled. I always wanted three hours per week of lecture and three hours per week of recitation, but couldn’t impose that much time on the students, so the recitation was short-changed. Putting parts of the lectures onto video opened up time for more in-class discussion and problem solving.
My goals for the Artificial Intelligence (AI) course fall in two categories: I want students to understand and apply specific technical algorithms, but I also want them to appreciate the bigger picture: the goals of AI, its recent progress (e.g. self-driving cars and IBM’s Jeopardy-playing Watson), and why developing AI is so hard. Pushing lectures to the web seemed a good way to free up time for class discussion and active learning.
Let me share what I learned about clickers and videos.
Using “clickers” (more formally, the “TurningPoint student response system”) was a revelation to me. Although I have taught this course many times, I was surprised at what the students knew or did not know, and with how they interpreted my questions. I asked dozens of short questions every class. Having instant feedback was great. Some questions everyone got right, and I moved on. Others had 60/40 splits between the right and wrong answer, and led either to clarification of what the question meant, or of the concepts behind it. Sometimes I used the clickers to demonstrate that different students had different approaches to a problem.
Could I have achieved the same effect by having students just give a show of hands? Not if they can see who else has their hands up. (One colleague said he has students close their eyes and raise their hands; I can’t bring myself to do that.) After most of the students had selected an answer, I displayed a bar chart showing how many students had picked each answer. Within seconds, the number of majority answers grew while the minority answers shrank. Students all wanted to give the majority answer, even though they were not graded on their answers.
Using clickers had a secondary benefit; I have always made lectures optional. Now with outside videos reducing the amount of “tutorial background” information being given in class, and clicker use showing attendance, I graded (a token amount) for participation. Showing up is a good thing, especially for more open-ended discussions that go beyond the textbook. Students also did small think-pair-share in-class problem assignments, thinking about a question, discussing it with another student and reporting back. An old technology, but a good one.
I also had students watch a number of pre-recorded videos, some taken from the Stanford AI course, and some that I recorded myself. All featured multiple short segments on specific mini-topics. I wasn’t sure how students would like the videos. I don’t much like watching instructional videos; I like the control of skimming or reading slowly that I get with a textbook. But my students, mostly, really liked the videos. In particular, students complain that the textbooks don’t have enough worked examples. It is extremely easy for me to work an example on a video—talking is much faster than writing—and so I put up a bunch of worked examples for those students who wanted more help. They loved it.
More generally, I had implicitly assumed that I needed to cover in class all the material that students need to know. Producing the videos gave me the (in retrospect) blindingly obvious insight that I just need to be sure the students have the resources to learn the material I want them to know; I don’t need to teach everything. I could thus spend more time in class giving fun, big picture insights, and relegate much more of the technical detail to homework—either reading the textbook or watching videos.
I did try for one class moving the entire lecture to video, and using the whole class for clickers and class participation, following the model suggested for some online courses. Students hated it. Perhaps I just didn’t have engaging enough class exercises (too many short-answer questions are tedious), but the students vociferously requested a mix of class lecture and discussion.
At the end of the semester, I surveyed the students to see how they liked the course, and got an overwhelming positive response to both the use of clickers and videos. Everyone said to keep them next year.
How will I teach the course next year? I will again punctuate my lectures with clicker questions and think-pair-share exercises that get the students to understand both how methods work, and why they need to work that way. For each lecture, I’ll make a clearer specification of what students need to know before the class; I’ll provide a reading or video containing the prerequisite material, and a few self-test questions to make sure the students know the material. I’ll also make sure there are plenty of videos working problem examples for use after class. And I’ll indulge myself in spending more time on the bigger, philosophical questions about human and computer intelligence.
My initial goal had been to better understand online learning, but in the end I learned a different lesson. Using new technologies drove me to explore different ways of teaching, allowed me to interact more deeply with my students and, I hope, enabled them to learn more. Sometimes even old professors can learn new tricks.
Lyle H. Ungar is an Associate Professor of Computer and Information Science at Penn Engineering and
Associate Director of the Penn Center for Bioinformatics (PCBI).
This essay continues the series by the Center for Teaching & Learning that began in the fall of 1994 as the joint creation of the
College of Arts & Sciences and the Lindback Society for Distinguished Teaching.
See www.upenn.edu/almanac/teach/teachall.html for the previous essays.