Guest Contribution by Rob Rambusch
Picture by Ressaure (CC licensed)
Me: Test Subject
Rob Rambusch is a Project Manager for software development and implementation based in New York City. This was his first exposure to any class on Artificial Intelligence. He can be contacted at robrambusch [AT] gmail.com, or through Google+.
"A bold experiment in distributed education, "Introduction to Artificial Intelligence" will be offered free and online to students worldwide from October 10th to December 18th 2011. The course will include feedback on progress and a statement of accomplishment. Taught by Sebastian Thrun and Peter Norvig, the curriculum draws from that used in Stanford's introductory Artificial Intelligence course. The instructors will offer similar materials, assignments, and exams."
"Peter Norvig is Director of Research at Google Inc. He is also a Fellow of the American Association for Artificial Intelligence and the Association for Computing Machinery. Norvig is co-author of the popular textbook Artificial Intelligence: A Modern Approach. Prior to joining Google he was the head of the Computation Sciences Division at NASA Ames Research Center."
"Sebastian Thrun is a Research Professor of Computer Science at Stanford University, a Google Fellow, a member of the National Academy of Engineering and the German Academy of Sciences. Thrun is best known for his research in robotics and machine learning."
One Month Before: Preparation
I started working on the prerequisites in the month before the course began. I prepared for the class by learning Probability and Linear Algebra from Salman Khan at the Khan Academy website. I followed up by watching lectures on Linear Algebra by Gilbert Strang at the MIT OCW website. So even before the class began I had exposure to two common online teaching styles, the filmed lecture and the video tutorial.
Things started out easy with an introduction to AI followed by an overview of different uses for it.
Sebastian Thrun wrote:
PURPOSE OF THIS CLASS
- TO TEACH YOU THE BASIS OF ARTIFICIAL INTELLIGENCE
- TO EXCITE YOU
The second unit for this week covered search. It was all new but not too demanding.
The format of the class, short videos with embedded quizzes, had an easy rhythm to it. There was an immediate contrast with the Khan videos which usually ran 8 to 12 minutes. The Unit 1 videos ranged in size from 12 seconds to 6 minutes 28 seconds.
The subjects flowed easily from one into another and the short lengths made it possible to easily structure study time. You were always within a few minutes of a convenient break point. This was in marked contrast to the filmed lectures I had seen during the previous month. There, one had the sense of viewing an artefact of an event that one was not involved in. Here there was a feeling of immediacy. The professors were either looking directly at a camera with no one else in the picture or drawing/writing on a piece of paper that filled the camera frame. Peter and Sebastian - I'll refer to them by first name because while I was taking the course there was no one else in the room - clearly were talking to me.
October 17th - November 17th
Class work got harder not in an ascending curve but in fits and starts according to subject. It was still possible to get a good result on the homework but it took more time to do so.
The short video format was very successful. The quizzes made re-watching the videos worthwhile as they gave you the chance to re-test your understanding of the subject and know whether you had really absorbed the information being presented. One of the students came up with a website that leveraged the captioning associated with the lectures. It combined the YouTube videos with written transcripts of them. It turned out to be very useful in preparing for homeworks and exams.
After each homework was put up, it was followed by a flurry of questions about perceived ambiguities or need for clarifications. Some of this was probably down to the large number of students taking the course in what was for them a second language. Some of it was sheer contrariness or fishing for answers.
Sebastian and Peter instituted "Office Hours". During the week students would suggest questions at http://www.youtube.com/eduatgoogle and vote on them. The most popular questions were answered by Peter and Sebastian at the end of the week as a YouTube video.
November 19th - November 21st - Midterm
We were given 72 hours in which to do the midterm. Some of this was probably due to server load issues and some due to the experience we had in the homeworks of emerging "clarifications" of the questions. Since it was an open book exam, reviewing the lectures and transcripts was a valid and successful strategy. The average score was 83%.
November 22nd - December 15th
Class work got harder still. The grades on quizzes and homeworks were harder to come by.
It helped to be able to discuss questions and concepts on the two most widely used forums - Aiqus and Reddit:
- Aiqus - The Q&A format limited discussion but seemed to help participants focus on course-related topics better. I fell into becoming a forum moderator as a result of being willing to devote large amounts of my free time to it. (In the early weeks of the class I was one of the more active posters. One day I noticed that I and two or three others had a diamond next to our names. That meant that we had a slightly increased privilege level. It was all pretty organic, with a few posts on the forum explaining the mechanics of the forum by the original moderator, who had two diamonds next to his name.)
- Reddit - Seemed less useful because of the larger number of participants and the unstructured nature of the way posts and replies were organized.
December 16th - December 18th - Final
Again we were given 72 hours and the format was open-book. I don't know what the class averaged on this exam.
There were a few of the usual grumblings on the forum about "ambiguities" and scoring methods. There was also a sense that this was a really good course and many expressed an interest in continuing on further in AI. So based upon the "Purpose of this Class" - the experiment was a success.
In trying to explain the appeal of the class in a post on Aiqus, this was the closest I could come: -
"This class felt like sitting in a bar with a really smart friend who is explaining something you haven't yet grasped but are about to."
The whole drawn-on-a-napkin feel of the class was responsible for much of its charm. The napkin was visible to 160,000 people but that didn't detract from the personal nature of the learning experience. The website had a few bumps but still managed to handle a large load of students on at all times using multiple OS/browser combinations. So the course succeeded in creating a platform that can be extended and used for future classes. We won't know much about the data until we hear from Peter Norvig when he speaks at TED in February. Circle that on your calendar.
The thing that impressed me the most about the class wasn't the size of the student body, the delivery method, or the ridiculous student-to-teacher ratio. It was the quality of instruction. I always chose college classes by three criteria. Was the professor an excellent teacher? Did he/she have a deep knowledge of the subject matter? Was he/she passionate about their field? Any class that had two out of those three would be memorable. This is the first time I've ever seen all three questions answered affirmatively for a class, and by both teachers.
It turns out that it's not the camera, it's the actor.
Rob Rambusch, 20/12/2011
* Quotes in the "Experiment" and "Experimenters" section are taken from the class website - https://www.ai-class.com/ .