September 14th, 4:12pm. I’m sitting on the lawn outside of the Stata Center at MIT. It’s a balmy 84 degrees. I’ve got an iPad in one hand. A Mountain Dew in the other. And I’m at lecture. Sort of.
I’m listening to a lecture on the right half of my iPad, taking notes with a stylus on the other half. It's a video lecture (obviously), and this is MIT's first ever completely-online-course for MIT students. Watching the lectures and taking notes on my iPad made me feel like the child in Alan Kay's Dynabook photo. It felt like peeking into the future. Not without flaws, but closer than anything I've seen before. Here, I describe the class, it's strengths, weakness, and why more classes should adopt this model.
I took Circuits and Electronics this semester, but instead of the standard version (6.002, in MIT talk), I took 6.s064. It was MIT's first on-campus, online class for credit, organized by Professor Sussman, Professor Agarwal, and Bonnie Lam.
The entire class was hosted on the EdX platform, and the structure of the class was quite straightforward: instead of in-person lectures, there were weekly online video sequences. Instead of written problem sets to turn in, there were online questions that needed to be completed each week, along with a online "lab". There were 3 exams: two midterms and a final, all of which were proctored, but completed online. I'll break down my thoughts on each component.
Online Video Lectures
Each week, there were two video sequences. These sequences were made up of shorter videos, each about 5 minutes. I haven't added up the time, but it felt similar to the standard 3 hours/week of content that the standard class contains. The advantages for everyone here are obvious.
For students: you are never late to class, you never oversleep. If you don't understand a particular section, you can simply re-watch it. The answer to "can we have class outside today?" is always yes. There are also mini-questions throughout the sequence to make sure you understand what's happening. Unfortunately many of the questions were simply plug-and-chug style questions. I often skipped them, since it was obvious that you just had to plug in the numbers in the question to the formula that was just explained. In the future, it would be great for these questions to be more conceptual, less algebra.
For professors: you don't have to re-teach the same material over and over again. You simply record it once, or many times and pick the best ones. Once you do it once, it's done for years. Since the sequences are very modular, individual lectures can be easily updated when the field changes, or if an error is caught.
The lectures in 6.s064 were easy to understand and very thorough. However, I wanted to take this class because I saw Professor Sussman fly through a circuit diagram with a very deep and intuitive understanding of the system (see starting at 19:55 in this video), and I wanted to learn to understand circuit systems like he did. Unfortunately Sussman didn't make the lecture videos, and Agarwal's style is more formula based.
In fact, Sussman gave a review session right before the final exam, where he systematically explained all the "powerful ideas" of the course. Agarwal does note all the powerful ideas ("aha moments", etc), but I wish a larger portion of the class was dedicated to that. During the review sessions Sussman said something like "formulas are a good way to find specific answers, but a weak form of understanding". I agree, and I think a future version of this class could decrease the focus on formulas and increase the focus on developing an intuition.
There's also one minor technical complaint: I love watching videos on the iPad, especially in split screen where I can watch the video on one half and take notes on the other, but you can't speed up the video on an iOS device†.†June 2020 Update: Since originally writing this, you can watch YouTube videos on an iPad at 2x speed now. I can comfortably watch videos at 2x speed, so I often had to revert back to a desktop to watch the videos. This should be a straightforward software enhancement. It seems trivial, but being able "to walk around with a lecture" is amazingly liberating. You are so much less constrained when you are feeling tired midway through a lecture and have the ability to simply get up, stretch, or get a cup of coffee while still holding the "lecture" (iPad) in your hand.
Online Problem Sets
One of the biggest problems with most standard classes is the massive delay in the problem-solving feedback loop. Students spend hours and hours on a problem set, then they have to turn it in a wait a week to find out whether they understand the material.
It's very clear that instant feedback is the best way to learn something. Imagine if you were learning tennis, but you had to wait 10 minutes after you hit a shot to see if it went in. You'd never learn. I think the unbelievably slow feedback loop in standard problem sets is a huge hinderance to learning.
EdX's online problem sets fix this problem pretty easily. The class allowed 25 submissions per problem, so you can get the question wrong 24 times before you lose credit for the problem. But even more importantly, you know right away whether or not you know how to solve the problem. Then, as soon as you get the question correct, you can see the step-by-step solution, to see if you did the problem the most efficient way.
Again, the problems are very formula/numerical focused, and the system relies on a boolean result, effectively: "yes you got the answer exactly right" or "no, that's not right". In future versions of the software, it would be useful for the system to provide more feedback about incorrect solutions (are you in the correct order of magnitude? is the sign right? etc). This is a more complex software challenge, but it's definitely doable.
There's a much easier way, however, to improve the online problem sets for this course: have more problems! The online, instant feedback nature of these problems means they are actually much faster to complete. I didn't have to spend any time double checking my work like I normally do (looking for missing negative signs or incorrectly multiplying), I simply submitted the answer. If it was wrong, then I went back and looked over my work. The net result of this is that solving problem is much faster. This saved time could be used to get a broader array of questions by simply increasing the question bank for each problem set.
The online labs, like the problem sets, were due weekly. There were guided instructions to completing a more open ended problem. I found these to be of little use. Simultaneously with this class, I took the 3 unit 6.002 extension: 6.169 Theory and Application of Circuits and Electronics. I found in the in-physical-labs to be much more helpful than the digital versions. For off-campus learners, online labs might be a passable solution to substitute for hands on experience, but for on-campus students, the real lab is much more useful. I think online learning classes and labs should be two distinct things. See my concluding thoughts below for more details.
The lab software is simply not good enough yet. This not a fault of the course, but a fault of the entire computing field. Seymour Papert coined the term "mathland", a place where math was part of the culture, and therefore easily absorbed (equivalent to learning French by going to France). Papert thought computers could be that mathland. But computers can also be circuitsland. Unfortunately, current software is quite far from circuitsland. More on this in a future post….
Both midterms and the final exam took place on the EdX platform, but in a proctored setting. All students came into a single classroom, and had two or three hours to complete the exam online. The exams were open-book, open-note, and you had 4 submissions per question.
I thought this was a great way to do exams. It's a great compromise between the honor-system online at-home exams, and written tests. It was great to see the result right away, but there is one major issue: there's no partial credit.
I knew how to solve almost all of the questions on the exams, but with no partial credit, there's no way to show that you know how to do the problem if you make any mistakes. Flipping a negative sign should not be equivalent to not knowing how to do the problem at all. I'm not sure what the solution to this is... the software system would have to be fairly advanced to start assigning partial credit. Maybe the instructors could collect student's work, and then assign points to incorrect questions based on their work. Of course, this removes the fully-automated nature of the class, so maybe this is an intractable downside for the foreseeable future.
The instructors held office hours (OHs), class meet-ups, and review sessions in-person. They also answered questions through the Wiki/QA system built into the EdX software. I didn't attend OHs because I had a pretty busy semester, and I was able to figure out most of the questions that I had. I would have loved to attend the class meet-ups, but I had a conflicting class during the scheduled period. In the future, varying the times of the meet-ups could be helpful. I did attend the final exam review session, and (as mentioned above), it was extremely helpful for understanding the big-picture concepts of the class. Other students likely got more out of OHs than I did.
MIT's motto is "Mens Et Manus", Mind and Hand. In my vision of the future of the undergraduate experience, there are two types of classes: "mind classes" and "hand classes".
Mind classes, which are mostly online, will be very similar to this experimental class, with video lectures and online problem solving. These classes will be the place to learn structured content.
Hand classes, meanwhile, are lab based. These classes will be where students make discoveries, collaborate with peers, and put their mind-classes to use. Currently, the MIT EECS curriculum requires one lab class. I claim that about 50% of classes should be lab classes. When people talk about digital learning, the question of "why be on campus" always seems to come up. Lab classes are the obvious answer: a place to interact with peers, and a place to work on things that can't be easily simulated online.
I've taken a few classes now where the online and in-class components have been blended — sort of like the flipped classroom model, but in a lecture hall with over 100 students. Flipped classrooms should simply kill off the lecture, and leave in recitations. But watching lectures online, and then still coming into lecture has not seemed to work. Classes should either be entirely online, or online + in-person problem solving. Trying to simply add online components without removing another aspect of the class should be avoided.
This past semester's experimental, online circuits class is a great start towards creating "mind classes". Small changes to the software and structure of various parts of the course could go along way, but the class is already great. Taking this class was glimpsing into the future. Other classes should adopt this structure as fast as possible. Soon, MIT truly will be mind and hand.
Update: The MIT News article about this class quotes me from this blog post.