Teaching Online

If you use robots to teach, you teach people to be robots.

— variously attributed

Technology has changed teaching and learning many times. Before blackboards were introduced into schools in the early 1800s, for example, there was no way for teachers to share an improvised example, diagram, or exercise with an entire class at once. Combining low cost, low maintenance, reliability, ease of use, and flexibility, blackboards enabled teachers to do things quickly and at scale that they had only been able to do slowly and piecemeal before. Similarly, the hand-held video camera revolutionized athletics training, just as the tape recorder revolutionized music instruction a decade earlier.

Many of the people pushing the Internet into classrooms don’t know this history, and don’t realize that it is just the latest in a long series of attempts to use machines to teach Watt2014. From the printing press through radio and television to desktop computers and mobile devices, every new way to share knowledge has produced a wave of aggressive optimists who believe that education is broken and that technology can fix it. However, ed tech’s strongest advocates have often known less about “ed” than they do about “tech”, and have often been driven more by the prospect of profit than by the desire to improve learning.

Today’s debate is often muddied by the fact that “online” and “automated” don’t have to be the same thing. Live online teaching can be a lot like leading a small-group discussion. Conversely, the only way to teach several hundred people at a time is to standardize and automate assessment; the learner’s experience is largely the same from whether the automation uses software or a squad of teaching assistants working to a tightly-defined rubric.

This chapter therefore looks at how the Internet can and should be used to deliver automated instruction, i.e., to teach with recorded videos and assess via automatically-graded exercises. The next chapter will then explore ways of combining automated instruction with live teaching delivered either online or in person.


The highest-profile effort to reinvent education using the Internet is the Massive Open Online Course, or MOOC. The term was invented by David Cormier in 2008 to describe a course organized by George Siemens and Stephen Downes. That course was based on a connectivist view of learning, which holds that knowledge is distributed and learning is the process of finding, creating, and pruning connections.

The term “MOOC” was quickly co-opted by creators of courses that kept more closely to the hub-and-spoke model of a traditional classroom, with the instructor at the center defining goals and the learners seen as recipients or replicators of knowledge. Classes that use the original connectivist model are now sometimes referred to as “cMOOCs”, while classes that centralize control are called “xMOOCs”. (The latter kind of course is also sometimes called a “MESS”, for Massively Enhanced Sage on the Stage.)

Two strengths of the MOOC model are that learners can work when it’s convenient for them, and that they have access to a wider range of courses, both because the Internet brings them all next door and because online courses typically have lower direct and indirect costs than in-person courses. Five years ago, you couldn’t cross the street on a major university campus without hearing some talking about how MOOCs would revolutionize education, destroy it, or possibly both.

But MOOCs haven’t been nearly as effective as their more enthusiastic proponents claimed they would be Ubel2017. One reason is that recorded content is ineffective for many novices because it cannot clear up their individual misconceptions (s:models): if they don’t understand an explanation the first time around, there usually isn’t a different one on offer. Another is that the automated assessment necessary in order to put the “massive” in MOOC only works well at the lower levels of Bloom’s Taxonomy. It’s also now clear that learners have to shoulder much more of the burden of staying focused in a MOOC, and that the impersonality of working online can demotivate people and encourage uncivil behavior.

Marg2015 examined 76 MOOCs on various subjects, and found that the quality of lesson design was poor, though organization and presentation of material was good. Closer to home, Kim2017 studied 30 popular online coding tutorials, and found that they largely teach the same content the same way: bottom-up, starting with low-level programming concepts and building up to high-level goals. Most require learners to write programs, and provide some form of immediate feedback, but this feedback is typically very shallow. Few explain when and why concepts are useful (i.e., they don’t show how to transfer knowledge) or provide guidance for common errors, and other than rudimentary age-based differentiation, none personalize lessons based on prior coding experience or learner goals.

Personalized Learning

Few terms have been used and abused in as many ways as personalized learning. To most ed tech proponents, it means dynamically adjusting the pace or focus of lessons based on learner performance, which in practice means that if someone answers several questions in a row correctly, the computer will skip some of the subsequent questions.

Doing this can produce modest improvements in outcomes, but better is possible. For example, if many learners find a particular topic difficult, the teacher can prepare multiple alternative explanations of that point—essentially, multiple paths forward through the lesson rather than accelerating a single path—so that if one explanation doesn’t resonate, others are available. However, this requires a lot more design work on the teacher’s part, which may be why it’s a less popular approach with the tech crowd.

And even if it does work, the effects are likely to be much less than some of its advocates believe. A good teacher makes a difference of 0.1–0.15 standard deviations in end-of-year performance in grade school Chet2014 (see this article for a brief summary). It’s simply unrealistic to believe that any kind of automation can outdo this any time soon.

So how should the web be used in teaching and learning tech skills? From an educational point of view, its pros and cons are:

Learners can access more lessons, more quickly, than ever before.
Provided, of course, that a search engine considers those lessons worth indexing, that their internet service provider and government don’t block it, that the truth isn’t drowned in a sea of attention-sapping disinformation.
Learners can access better lessons than ever before,
unless they are being steered toward second-rate material in order to redistribute wealth from the have-nots to the haves McMi2017. (It’s worth remembering that scarcity increases perceived value, so as online education becomes cheaper, it will be seen as being worth less.)
Learners can access far more people than ever before as well.
But only if those learners actually have access to the required technology, can afford to use it, and aren’t driven offline by harassment or marginalized because they don’t conform to the social norms of whichever group is talking loudest. In practice, most MOOC users come from secure, affluent backgrounds Hansen2015.
Teachers can get far more detailed insight into how learners work.
So long as learners are doing things that are amenable to large-scale automated analysis and either don’t object to the use of ubiquitous surveillance in the classroom, or aren’t powerful enough for their objections to matter.

Marg2015,Mill2016a,Nils2017 describe ways to accentuate the positives in the list above while avoiding the negatives:

Make deadlines frequent and well-publicized,
and enforce them, so that learners will get into a work rhythm.
Keep synchronous all-class activities like live lectures to a minimum
so that people don’t miss things because of scheduling conflicts.
Have learners contribute to collective knowledge,
e.g., take notes together (s:classroom-notetaking), serve as classroom scribes, or contribute problems to shared problem sets (s:individual-peer).
Encourage or require learners to do some of their work in small groups
that do have synchronous online activities such as a weekly online discussion to help learners stay engaged and motivated without creating too many scheduling headaches. (See s:meetings for some tips on how to make these discussions fair and productive.)
Create, publicize, and enforce a code of conduct
so that everyone can actually (as opposed to theoretically) take part in online discussions (s:intro-code-of-conduct).
Use lots of short lesson episodes rather than a handful of lecture-length chunks
in order to minimize cognitive load and provide lots of opportunities for formative assessment. This also helps with maintenance: if all of your videos are short, you can simply re-record any that need maintenance, which is often cheaper than trying to patch longer ones.
Use video to engage rather than instruct,
since, disabilities aside, learners can read faster than you can talk. The exception to this rule is that video is actually the best way to teach people verbs (actions): short screencasts that show people how to use an editor, step through code in a debugger, and so on are more effective than screenshots with text.
Identify and clear up misconceptions early
(s:models). If data shows that learners are struggling with some parts of a lesson, create alternative explanations of those points and extra exercises for them to practice on.

All of this has to be implemented somehow, which means that you need some kind of teaching platform. You can either use an all-in-one learning management system like Moodle or Sakai, or assemble something yourself using Slack or Zulip for chat, Google Hangouts or appear.in for video conversations, and WordPress, Google Docs, or any number of wiki systems for collaborative authoring. If you are just starting out, then use whatever requires the least installation and administration on your side, and the least extra learning effort on your learners’ side. (I once ran a half-day class using group text messages because that was the only tool everyone was already familiar with.)

The most important thing when choosing technology is to ask your learners what they are already using. Normal people don’t use IRC, and find its arcane conventions and interface offputting. Similarly, while this book lives in a GitHub repository, requiring non-experts to submit pull requests has been an unmitigated disaster, even with GitHub’s online editing tools. As a teacher, you’re asking people to learn a lot; the least you can do in return is learn how to use the tools they prefer.

Points for Improvement

One way to demonstrate to learners that they are learning with you, not just from you, is to allow them to edit your course notes. In live courses, we recommend that you enable them to do this as you lecture (s:classroom-notetaking); in online courses, you can put your notes into a wiki, a Google Doc, or anything else that allows you to review and comment on changes. Giving people credit for fixing mistakes, clarifying explanations, adding new examples, and writing new exercises doesn’t reduce your workload, but increases engagement and the lesson’s lifetime (s:process-maintainability).

A major concern with any online community, learning or otherwise, is how to actually make it a community. Hundreds of books and presentations discuss this, but most are based on their authors’ personal experiences. Krau2016 is a welcome exception: while it predates the accelerating descent of Twitter and Facebook into weaponized abuse and misinformation, most of what was true then is true now. Foge2005 is also full of useful tips for the community of practice that learners may hope to join.

Freedom To and Freedom From

Isaiah Berlin’s 1958 essay “Two Concepts of Liberty” made a distinction between positive liberty, which is the ability to actually do something, and negative liberty, which is the absence of rules saying that you can’t do it. Unchecked, online discussions usually offer negative liberty (nobody’s stopping you from saying what you think) but not positive liberty (many people can’t actually be heard). One way to address this is to introduce some kind of throttling, such as only allowing each learner to contribute one message per discussion thread per day. Doing this gives those with something to say a chance to say it, while clearing space for others to say things as well.

One other concern people have about teaching online is cheating. Day-to-day dishonesty is no more common in online classes than in face-to-face settings Beck2014, but the temptation to have someone else write the final exam, and the difficulty of checking whether this happened, is one of the reasons educational institutions have been reluctant to offer credit for pure online classes. Remote exam proctoring is possible, usually by using a webcam to watch the learner take the exam. Before investing in this, read Lang2013, which explores why and how learners cheat, and how courses can be structured to avoid giving them a reason to do so.


A core element of cMOOCs is their reliance on recorded video lectures. As mentioned in s:performance, a teaching technique called Direct Instruction that is based on precise delivery of a well-designed script has repeatedly been shown to be effective Stoc2018, so recorded videos can in principle be effective. However, scripts for direct instruction have to be designed, tested, and refined very carefully, which is an investment that many MOOC authors have been unwilling or unable to make. Making a small change to a web page or a slide deck only takes a few minutes; making even a small change to a short video takes an hour or more, so the cost to the teacher of acting on feedback can be unsupportable. And even when they’re well made, videos have to be combined with activities to be beneficial: Koed2015 estimated, “…the learning benefit from extra doing…to be more than six times that of extra watching or reading.”

If you are teaching programming, you may use screencasts instead of slides, since they offer some of the same advantages as live coding (s:performance-live). Chen2009 offers useful tips for creating and critiquing screencasts and other videos; f:online-screencasting reproduces the patterns that paper presents and the relationships between them, and is also a good example of a concept map (s:memory-concept-maps).

Patterns for Screencasting ([Chen2009](#BIB))

Guo2014 measured engagement by looking at how long learners watched MOOC videos. Some of its key findings were:

  • Shorter videos are much more engaging—videos should be no more than six minutes long.

  • A talking head superimposed on slides is more engaging than voice over slides alone.

  • Videos that felt personal could be more engaging than high-quality studio recordings, so filming in informal settings could work better than professional studio work for lower cost.

  • Drawing on a tablet is more engaging than PowerPoint slides or code screencasts, though it’s not clear whether this is because of the motion and informality, or because it reduces the amount of text on the screen.

  • It’s OK for teachers to speak fairly fast as long as they are enthusiastic.

One thing Guo2014 didn’t address is the chicken-and-egg problem: do learners find a certain kind of video engaging because they’re used to it, so producing more videos of that kind will increase engagement simply because of a feedback loop? Or do these recommendations reflect some deeper cognitive processes? Another thing this paper didn’t look at is learning outcomes: we know that learner evaluations of courses don’t correlate with learning Star2014,Uttl2017, and while it’s plausible that learners won’t learn from things they don’t watch, it remains to be proven that they do learn from things they do watch.

I’m a Little Uncomfortable

Guo2014’s research was approved by a university research ethics board, the learners whose viewing habits were monitored almost certainly clicked “agree” on a terms of service agreement at some point, and I’m glad to have these insights. On the other hand, I attended the conference at which this paper was published, and the word “privacy” didn’t appear in the title or abstract of any of the dozens of papers or posters presented. Given a choice, I’d rather not know how engaged learners are than see privacy become obsolete.

There are many different ways to record video lessons; to find out which are most effective, Mull2007a assigned 364 first-year physics learners to online multimedia treatments of Newton’s First and Second Laws in one of four styles:

concise lecture-style presentation.
Extended Exposition:
as above with additional interesting information.
Exposition with common misconceptions explicitly stated and refuted.
Learner-tutor discussion of the same material as in the Refutation.

Refutation and Dialogue produced the greatest learning gains compared to Exposition; learners with low prior knowledge benefited most, and those with high prior knowledge were not disadvantaged.

Flipped Classrooms

Fully automated teaching is one way to use the web in teaching; in practice, almost all learning in affluent societies has an online component: sometimes officially, and if not, through peer-to-peer back channels and surreptitious searches for answers to homework questions. Combining live and automated instruction allows instructors to use the strengths of both. In a classroom, the instructor can answer questions immediately, but it takes time for learners to get feedback on their coding exercises (sometimes days or weeks). Online, it can take a long time to get an answer, but learners can get immediate feedback on their coding (at least for those kinds of exercises we can auto-grade).

Similarly, online exercises have to be more detailed because they’re anticipating questions. I find that in-person lessons start with the intersection of what everyone needs to know and expands on demand, while online lessons have to include the union of what everyone needs to know because you aren’t there to do the expanding.

The most popular hybrid teaching strategy today is the flipped classroom, in which learners watch recorded lessons on their own, and class time is used for discussion and to work through problem sets. Originally proposed in King1993, the idea was popularized as part of peer instruction (s:classroom-peer), and has been studied intensively over the past decade. For example, Camp2016 compared students who chose to take a CS1 class online with those who took it in person in a flipped classroom. Completion of (unmarked) practice exercises correlated with exam scores for both, but the completion rate of rehearsal exercises by online students was significantly lower than lecture attendance rates for in-person students. Looking at what did affect the grade, they found that the students’ perception of the material’s intrinsic value was only a factor for the flipped section (and only once results were controlled for prior programming experience). Conversely, test anxiety and self-efficacy were factors only for the online section; the authors recommend trying to improve self-efficacy by increasing instructor presence online.

But are lectures worth attending at all? Or should we just provide recordings? Nord2017 examined the impact of recordings on both lecture attendance and students’ performance at different levels. In most cases the study found no negative consequences of making recordings available; in particular, students don’t skip lectures when recordings are available (at least, not any more than they usually do). The benefits of providing recordings are greatest for students early in their careers, but diminish as students become more mature.

Life Online

Nuth2007 found that there are three overlapping worlds in every classroom: the public (what the teacher is saying and doing), the social (peer-to-peer interactions between learners), and the private (inside each learner’s head). Of these, the most important is usually the social: learners pick up as much via cues from their peers as they do from formal instruction.

The key to making any form of online teaching effective is therefore to facilitate peer-to-peer interactions. To aid this, courses almost always have some kind of discussion forum. Vell2017 analyzes discussion forum posts from 395 CS2 students at two universities by dividing them into four categories:

request for help that does not display reasoning and doesn’t display what the student has already tried or already knows.
reflect students’ reasoning or attempts to construct a solution to the problem.
course policies, schedules, assignment submission, etc.
Content clarification:
request for additional information that doesn’t reveal the student’s own thinking.

They found that constructive and logistical questions dominated, and that constructive questions correlated with grades. They also found that students rarely ask more than one active question in a course, and that these don’t correlate with grades. While this is disappointing, knowing it helps set instructors’ expectations: while we might all want our courses to have lively online communities, most won’t.

Learners use forums in very different ways, and with very different results. Mill2016a observed that, “…procrastinators are particularly unlikely to participate in online discussion forums, and this reduced participation, in turn, is correlated with worse grades. A possible explanation for this correlation is that procrastinators are especially hesitant to join in once the discussion is under way, perhaps because they worry about being perceived as newcomers in an established conversation. This aversion to jump in late causes them to miss out on the important learning and motivation benefits of peer-to-peer interaction.”


Gull2004 describes an innovative online coding contest that combines collaboration and competition. The contest starts when a problem description is posted along with a correct, but inefficient, solution. When it ends, the winner is the person who has made the greatest overall contribution to improving the performance of the overall solution. All submissions are in the open, so that participants can see one another’s work and borrow ideas from each other; as the paper shows, the final solution is almost always a hybrid borrowing ideas from many people.

Batt2018 described a small-scale variation of this used in an introductory computing class. In stage one, each student submitted a programming project individually. In stage two, students were paired to create an improved solution to the same problem. The assessment indicates that two-stage projects tend to improve students’ understanding, and that they enjoyed the process.

Discussion isn’t the only way to get students to work together online. Pare2008 and Kulk2013 report experiments in which learners grade each other’s work, and the grades they assign are then compared with grades given by graduate-level teaching assistants or other experts. Both found that student-assigned grades agreed with expert-assigned grades as often as the experts’ grades agreed with each other, and that a few simple steps (such as filtering out obviously unconsidered responses or structuring rubrics) decreased disagreement even further. And as discussed in s:individual-peer, collusion and bias are not significant factors in peer grading.

Cumm2011 looked at the use of shareable feedback tags on homework; students could attach tags to specific locations in coding assignments (like code review) so that there’s no navigational cost for the reader, and they controlled whether to share their work and feedback anonymously. Students found tag clouds of feedback on their own work useful, but that the tags were really only meaningful in context. This is unsurprising: the greater the separation between action and feedback, the greater the cognitive load. What wasn’t expected was that the best and worst students were more likely to share than middling students.

Trust, but Educate

The most common way to measure the validity of feedback is to compare students’ grades to experts’ grades, but calibrated peer review (s:individual-peer) can be equally effective. Before asking learners to grade each others’ work, they are asked to grade samples and compare their results with the grades assigned by the teacher. Once the two align, the learner is allowed to start giving grades to peers. Given that critical reading is an effective way to learn, this result may point to a future in which learners use technology to make judgments, rather than being judged by technology.

One technique we will definitely see more of in coming years is online streaming of live coding sessions Haar2017. This has most of the benefits discussed in s:performance-live, and when combined with collaborative note-taking (s:classroom-notetaking) it can come pretty close to approximating an in-class experience.

Looking even further ahead, Ijss2000 identified four levels of online presence, from realism (we can’t tell the difference) through immersion (we forget the difference) and involvement (we’re engaged but aware of the difference) to suspension of disbelief (we are doing most of the work). Crucially, they distinguish physical presence, which is the sense of actually being somewhere, and social presence, which is the sense of being with others. In most learning situations, the latter is more important, and one way to foster it is to bring the technology learners use every day into the classroom. For example, Deb2018 found that doing in-class exercises with realtime feedback using mobile devices improved concept retention and student engagement while reducing failure rates.

Hybrid Presence

Combining online and in-person instruction can be more effective than either on its own. I have delivered very successful classes using real-time remote instruction, in which the learners are co-located at 2–6 sites, with helpers present, while I taught via streaming video (s:joining-using). This scales well, saves on travel costs, and is less disruptive for learners (particularly those with family responsibilities). What doesn’t work is having one group in person and one or more groups remotely: with the best will in the world, the local participants get far more attention.

Online teaching is still in its infancy: Luxt2009 surveyed peer assessment tools that could be useful in computing education, and Broo2016 describes many other ways groups can discuss things, but only a handful of these ideas are widely known or used.

I think that our grandchildren will probably regard the distinction we make between what we call the real world and what they think of as simply the world as the quaintest and most incomprehensible thing about us.

— William Gibson


Give Feedback on a Bad Screencast (whole class/20)

Watch this screencast as a group and give feedback on it. Organize feedback along two axes: positive vs. negative and content vs. presentation. When you are done, have each person in the class add one point to a 2x2 grid on a whiteboard (or in the shared notes) without duplicating any points that are already up there. What did other people see that you missed? What did they think that you strongly agree or disagree with? (You can compare your answers with the checklist in s:teacheval.)

Two-Way Video (pairs/10)

Record a 2–3 minute video of yourself doing something, then swap machines with a partner so that each of you can watch the other’s video at 4X speed. How easy is it to follow what’s going on? What if anything did you miss?

Viewpoints (individual/10)

According to Irib2009, different disciplines focus on different factors affecting the success or otherwise of online communities:

customer loyalty, brand management, extrinsic motivation.
sense of community, intrinsic motivation.
group identity, physical community, social capital, collective action.
Computer Science:
technological implementation.

Which of these perspectives most closely corresponds to your own? Which are you least aligned with?

Helping or Harming (small groups/30)

Susan Dynarski’s article in the New York Times explains how and why schools are putting students who fail in-person courses into online courses, and how this sets them up for even further failure.

  1. Working in small groups, read the article, come up with 2–3 things that schools could do to compensate for these negative effects, and create rough estimates of their per-student costs.

  2. Compare your suggestions and costs with those of other groups. How many full-time teaching positions do you think would have to be cut in order to free up resources to implement the most popular ideas for 100 students?

  3. As a class, do you think that would be a net benefit for the students or not?

Budgeting exercises like this are a good way to tell who’s serious about educational change. Everyone can think of things they’d like to do; far fewer are willing to talk about the tradeoffs needed to make change happen.