Small evaluation, big impact.

One of the problems with any model that builds in more feedback is that we incur both the time required to produce the feedback and we also have an implicit requirement to allow students enough time to assimilate and make use of it. This second requirement is still there even if we don’t have subsequent attempts at work, as we want to build upon existing knowledge. The requirement for good feedback makes no sense without a requirement that it be useful.

But let me reiterate that pretty much all evaluation and feedback can be very valuable, no matter how small or quick, if we know what we are trying to achieve. (I’ll get to more complicated systems in later posts.)

Novice programmers often struggle with programming and this early stage of development is often going to influence if they start off thinking that they can program or not. Given that automated evaluation only really provides useful feedback once the student has got something working, novice programming classes are an ideal place to put human markers. If we can make students think “Yes, I can do this” early on, this is the emotion that they will remember. We need to get to big problems quickly, turn them into manageable issues that can be overcome, and then let motivation and curiosity take the rest.


That first computer experience can stay with you your whole life. (Mine was actually punch cards but they don’t blink.)

There’s an excellent summary paper on computer programming visualisation systems aimed at novice programmers, which discusses some of the key problems novices face on their path to mastery:

  1. Novices can see some concepts as code rather than the components of a dynamic process. For example, they might see objects as simply a way of containing things rather than modelling objects and their behaviours. These static perceptions prevent the students from understanding that they are designing behaviours, not just writing magic formulas.
  2. There can be significant difficulties in understanding the computer, seeing the notional machine that is the abstraction, forming a basis upon which knowledge of one language or platform could be used elsewhere.
  3. Misunderstanding fundamental concepts is common and such misconceptions can easily cause weak understanding, leaving the students in the liminal state, unable to assimilate a threshold concept and move on.
  4. Students struggle to trace programs and work out what state the program should be in. In my own community, Raymond Lister, Donna Teague, Simon, and others have clearly shown that many students struggle with the tracing of even simple programs.

If we have put human markers (E1 or E2) into a programming class and identified that these are the problems we’re looking for, we can provide immediate targeted evaluation that is also immediate constructive feedback. On the day, in response to actual issues, authentic demonstration of a solution process that students can model. This is the tightest feedback and reward loop we can offer. How does this work?

  • Program doesn’t work because of one of the key problem areas.
  • Human evaluator intervenes with student and addresses the issue, encouraging discovery inside the problem area.
  • Student tries to identify problem and explains it to evaluator in context, modelling evaluator and based on existing knowledge.
  • Evaluator provides more guidance and feedback.
  • Student continues to work on problem.
  • We hope that the student will come across the solution (or think towards it) but we may have to restart this loop.

Note that we’re not necessarily giving the solution here but we can consider leading towards this if the student is getting visibly frustrated. I’d suggest never telling a student what to type as it doesn’t address any of the problems, it just makes the student dependent upon being told the answer. Not desirable. (There’s an argument here for rich development environments that I’ll expand on later.)

Evaluation like this is formative, immediate and rich. We can even streamline it with guidelines to help the evaluators although much of this will amount to supporting students as they learn to read their own code and understand the key concepts. We should develop students simple to complex, concrete to abstract, so some problems with abstraction are to be expected, especially if we are playing near any threshold concepts.

But this is where learning designers have to be ready to say “this may cause trouble” and properly brief the evaluators who will be on the ground. If we want our evaluators to work efficiently and effectively, we have to brief them on what to expect, what to do, and how to follow up.

If you’ve missed it so far, one of our big responsibilities is training our evaluation team. It’s only by doing this that we can make sure that our evaluators aren’t getting bogged down in side issues or spending too much time with one student and doing the work for them. This training should include active scenario-based training to allow the evaluators to practise with the oversight of the educators and designers.

We have finite resources. If we want to support a room full of novices, we have to prepare for the possibility of all of them having problems at once and the only way to support that at scale is to have an excellent design and train for it.

“Guest” post by Buckminster Fuller


Biosphère Montréal

I’m about to start a new thread of discussion, once I’ve completed the assessment posts, and this seemed to be good priming for thinking ahead.

“The true business of people should be to go back to school and think about whatever it was they were thinking about before somebody came along and told them they had to earn a living.”

Buckminster Fuller, reference.


Equity is the principal educational aesthetic

I’ve laid out some honest and effective approaches to the evaluation of student work that avoid late penalties and still provide high levels of feedback, genuine motivation and a scalable structure.


But these approaches have to fit into the realities of time that we have in our courses. This brings me to the discussion of mastery learning (Bloom). An early commenter noted how much my approach was heading towards mastery goals, where we use personalised feedback and targeted intervention to ensure that students have successfully mastered certain tiers of knowledge, before we move on to those that depend upon them.

A simple concept: pre-requisites must be mastered before moving on. It’s what much of our degree structure is based upon and is what determines the flow of students through courses, leading towards graduation. One passes 101 in order to go on to courses that assume such knowledge.

Within an individual course, we quickly realise that too many mastery goals starts to leave us in a precarious position. As I noted from my earlier posts, having enough time to do your job as designer or evaluator requires you to plan what you’re doing and keep careful track of your commitments. The issue that arises with mastery goals is that, if a student can’t demonstrate mastery, we offer remedial work and re-training with an eye to another opportunity to demonstrate that knowledge.

This can immediately lead to a backlog of work that must be completed prior to the student being considered to have mastered an area, and thus being ready to move on. If student A has completed three mastery goals while B is struggling with the first, where do we pitch our teaching materials, in anything approximating a common class activity, to ensure that everyone is receiving both what they need and what they are prepared for? (Bergmann and Sams’ Flipped Mastery is one such approach, where flipping and time-shifting are placed in a mastery focus – in their book “Flip Your Classroom”)

But even if we can handle a multi-speed environment (and we have to be careful because we know that streaming is a self-fulfilling prophecy) how do we handle the situation where a student has barely completed any mastery goals and the end of semester is approaching?

Mastery learning is a sound approach. It’s both ethically and philosophically pitched to prevent the easy out for a teacher of saying “oh, I’m going to fit the students I have to an ideal normal curve” or, worse, “these are just bad students”. A mastery learning approach tends to produces good results, although it can be labour intensive as we’ve noted. To me, Bloom’s approach is embodying one of my critical principles in teaching: because of the variable level of student preparation, prior experience and unrelated level of privilege, we have to adjust our effort and approach to ensure that all students can be brought to the same level wherever possible.

Equity is one of my principle educational aesthetics and I hope it’s one of yours. But now we have to mutter to ourselves that we have to think about limiting how many mastery goals there are because of administrative constraints. We cannot load up some poor student who is already struggling and pretend that we are doing anything other than delaying their ultimate failure to complete.

At the same time, we would be on shaky ground to construct a course where we could turn around at week 3 of 12 and say “You haven’t completed enough mastery goals and, because of the structure, this means that you have already failed. Stop trying.”

The core of a mastery-based approach is the ability to receive feedback, assimilate it, change your approach and then be reassessed. But, if this is to be honest, this dependency upon achievement of pre-requisites should have a near guarantee of good preparation for all courses that come afterwards. I believe that we can all name pre-requisite and dependency patterns where this is not true, whether it is courses where the pre-requisite course is never really used or dependencies where you really needed to have achieved a good pass in the pre-req to advance.

Competency-based approaches focus on competency and part of this is the ability to use the skill or knowledge where it is required, whether today or tomorrow. Many of our current approaches to knowledge and skill are very short-term-focussed, encouraging cramming or cheating in order to tick a box and move on. Mastering a skill for a week is not the intent but, unless we keep requiring students to know or use that information, that’s the message we send. This is honesty: you must master this because we’re going to keep using it and build on it! But trying to combine mastery and grades raises unnecessary tension, to the student’s detriment.

As Bloom notes:

Mastery and recognition of mastery under the present relative grading system is unattainable for the majority of students – but this is the result of the way in which we have “rigged” the educational system.

Bloom, Learning for Mastery, UCLA CSEIP Evaluation Comment, 1, 2, 1968.

Mastery learning is part and parcel of any competency based approach but, without being honest about the time constraints that are warping it, even this good approach is diminished.

The upshot of this is that any beautiful model of education adhering to the equity aesthetic has to think in a frame that is longer than a semester and in a context greater than any one course. We often talk about doing this but detailed alignment frequently escapes us, unless it is to put up our University-required ‘graduate attributes’ to tell the world how good our product will be.

We have to accept that part of our job is asking a student to do something and then acknowledging that they have done it, while continuing to build systems where what they have done is useful, provides a foundation to further learning and, in key cases, is something that they could do again in the future to the approximate level of achievement.

We have to, again, ask not only why we grade but also why we grade in such strangely synchronous containers. Why is it that a degree for almost any subject is three to five years long? How is that, despite there being nearly thirty years between the computing knowledge in the degree that I did and the one that I teach, they are still the same length? How are we able to have such similarity when we know how much knowledge is changing?

A better model of education is not one that starts from the assumption of the structures that we have. We know a lot of things that work. Why are we constraining them so heavily?

Being honest about driverless cars

I have been following the discussion about the ethics of the driverless car with some interest. This is close to a contemporary restatement of the infamous trolley problem but here we are instructing a trolley in a difficult decision: if I can save more lives by taking lives, should I do it? In the case of a driverless car, should the car take action that could kill the driver if, in doing so, it is far more likely to save more lives than would be lost?

While I find the discussion interesting, I worry that such discussion makes people unduly worried about driverless cars, potentially to a point that will delay adoption. Let’s look into why I think that. (I’m not going to go into whether cars, themselves, are a good or bad thing.)

Many times, the reason for a driverless car having to make such a (difficult) decision is that “a person leaps out from the kerb” or “driving conditions are bad” and “it would be impossible to stop in time.”

As noted in CACM:

The driverless cars of the future are likely to be able to outperform most humans during routine driving tasks, since they will have greater perceptive abilities, better reaction times, and will not suffer from distractions (from eating or texting, drowsiness, or physical emergencies such as a driver having a heart attack or a stroke).

In every situation where a driverless car could encounter a situation that would require such an ethical dilemma be resolved, we are already well within the period at which a human driver would, on average, be useless. When I presented the trolley problem, with driverless cars, to my students, their immediate question was why a dangerous situation had arisen in the first place? If the car was driving in a way that it couldn’t stop in time, there’s more likely to be a fault in environmental awareness or stopping-distance estimation.

If a driverless car is safe in varied weather conditions, then it has no need to be travelling at the speed limit merely because the speed limit is set. We all know the mantra of driving: drive to the conditions. In a driverless car scenario, the sensory awareness of the car is far greater than our own (and we should demand that it was) and thus we will eliminate any number of accidents before we arrived at an ethical problem.

Millions of people are killed in car accidents every year because of drink driving and speeding. In Victoria, Australia, close to 40% of accidents are tied to long distance driving and fatigue. We would eliminate most, if not all, of these deaths immediately with driverless technology adopted en masse.

What about people leaping out in front of the car? In my home city, Adelaide, South Australia, the average speed across the city is just under 30 kilometres per hour, despite the speed limit being 50 (traffic lights and congestion has a lot to do with this). The average human driver takes about 1.5 seconds to react (source), then braking deceleration is about 7 metres per second per second, less effectively in the wet. From that source, the actual stopping part of the braking, if we’re going 30km/h, is going to be less than 9 metres if it’s dry, 13 metres if wet. Other sources note that, with human reactions, the minimum overall braking is about 12 metres, 6 of which are braking. The good news is that 30km/h is already the speed at which only 10% of pedestrians are killed and, given how quickly an actively sensing car could react and safely coordinate braking without skidding, the driverless car is incredibly unlikely to be travelling fast enough to kill someone in an urban environment and still be able to provide the same average speed as we had.

The driverless car, without any ethics beyond “brake to avoid collisions”, will be causing a far lower level of injury and death. They don’t drink. They don’t sleep. They don’t speed. They will react faster than humans.

(That low urban speed thing isn’t isolated. Transport for London estimate the average London major road speed to be around 31 km/h, around 15km/h for Central London. Central Berlin is about 24 km/h, Warsaw is 26. Paris is 31 km/h and has a fraction of London’s population, about twice the size of my own city.)

Human life is valuable. Rather than focus on the impact on lives that we can see, as the Trolley Problem does, taking a longer view and looking at the overall benefits of the driverless car quickly indicates that, even if driverless cars are dumb and just slam on the brakes, the net benefit is going to exceed any decisions made because of the Trolley Problem model. Every year that goes by without being able to use this additional layer of safety in road vehicles is costing us millions of lives and millions of injuries. As noted in CACM, we already have some driverless car technologies and these are starting to make a difference but we do have a way to go.

And I want this interesting discussion of ethics to continue but I don’t want it to be a reason not to go ahead, because it’s not an honest comparison and saying that it’s important just because there’s no human in the car is hypocrisy.

I wish to apply the beauty lens to this. When we look at a new approach, we often find things that are not right with it and, given that we have something that works already, we may not adopt a new approach because we are unsure of it or there are problems. The aesthetics of such a comparison, the characteristics we wish to maximise, are the fair consideration of evidence, that the comparison be to the same standard, and a commitment to change our course if the evidence dictates that it be so. We want a better outcome and we wish to make sure that any such changes made support this outcome. We have to be honest about our technology: some things that are working now and that we are familiar with are not actually that good or they are solving a problem that we might no longer need to solve.

Human drivers do not stand up to many of the arguments presented as problems to be faced by driverless cars. The reason that the trolley problems exists in so many different forms, and the fact that it continues to be debated, shows that this is not a problem that we have moved on from. You would also have to be highly optimistic in your assessment of the average driver to think that a decision such as “am I more valuable than that evil man standing on the road” is going through anyone’s head; instead, people jam on the brakes. We are holding driverless cars to a higher standard than we accept for driving when it’s humans. We posit ‘difficult problems’ that we apparently ignore every time we drive in the rain because, if we did not, none of us would drive!

Humans are capable of complex ethical reasoning. This does not mean that they employ it successfully in the 1.5 seconds of reaction time before slamming on the brakes.

We are not being fair in this assessment. This does not diminish the value of machine ethics debate but it is misleading to focus on it here as if it really matters to the long term impact of driverless cars. Truck crashes are increasing in number in the US, with over 100,000 people injured each year, and over 4,000 killed. Trucks follow established routes. They don’t go off-road. This makes them easier to bring into an automated model, even with current technology. They travel long distances and the fatigue and inattention effects upon human drivers kill people. Automating truck fleets will save over a million lives in the US alone in the first decade, reducing fleet costs due to insurance payouts, lost time, and all of those things.

We have a long way to go before we have the kind of vehicles that can replace what we have but let’s focus on what is important. Getting a reliable sensory rig that works better than a human and can brake faster is the immediate point at which any form of adoption will start saving lives. Then costs come down. Then adoption goes up. Then millions of people live happier lives because they weren’t killed or maimed by cars. That’s being fair. That’s being honest. That will lead to good.

Your driverless car doesn’t need to be prepared to kill you in order to save lives.

A google driverless car on a stretch of route 66, in the desert. The car is stationary and facing the camera in a posed shot.

And you may even still be able to get your kicks.

$6.9M Federal Funding for CSER Digital Technologies @cseradelaide @UniofAdelaide @birmo @cpyne @sallyannw

Our research group, the Computer Science Education Research Group, has been working to support teachers involved in digital technologies for some time. The initial project was a collaboration between Google and the University of Adelaide, with amazing work from Sally-Ann Williams of Google to support us, to produce a support course that was free, open and recognised as professional development for teachers who were coming to terms with the new Digital Technologies (draft) curriculum. Today we are amazed and proud to announce $6.9 million dollars in Federal Funding over the next four years to take this project … well … just about everywhere.

You can read about what we’ve been doing here

I’ll now share Katrina’s message, slightly edited, to the rest of the school.

Today we hosted a visit from Ministers Birmingham and Pyne to announce a new funding agreement to support a national support program for Australian teachers within the Digital Technologies space.

Ministers Birmingham and Pyne confirmed that the Australian Government is providing $6.9 million over four years to the Computer Science Education Research Group at the University of Adelaide to support the roll out, on a national basis, of the teacher professional learning Massive Open Online Course (MOOC) supporting Australian primary and junior secondary teachers in developing skills in implementing the Australian Curriculum: Digital Technologies.

The CSER MOOC program provides free professional development for Australian teachers in the area of Computer Science, and supports research into the learning and teaching of Computer Science in the K-12 space. As part of this new program, we will be able to support teachers in disadvantaged schools and Indigenous schools across Australia in accessing the CSER MOOCs. We will also be able to establish a national lending library program to provide access to the most recent and best digital technologies education equipment to every school.

The Ministers, along with our Executive Dean and the Vice-Chancellor accompanied us to visit a coding outreach event for children run this morning as part of the University’s Bright Sparks STEM holiday program.

Here’s the ministerial announcement.

Bright Sparks 21 Jan 2015 VC

Senator Birmingham, Minister Pyne, Professor Bebbington (VC of the University Adelaide) and A/Prof Katrina Falkner with one of the Bright Spark participants.

The hand of an expert is visible in design

In yesterday’s post, I laid out an evaluation scheme that allocated the work of evaluation based on the way that we tend to teach and the availability, and expertise, of those who will be evaluating the work. My “top” (arbitrary word) tier of evaluators, the E1s, were the teaching staff who had the subject matter expertise and the pedagogical knowledge to create all of the other evaluation materials. Despite the production of all of these materials and designs already being time-consuming, in many cases we push all evaluation to this person as well. Teachers around the world know exactly what I’m talking about here.

Our problem is time. We move through it, tick after tick, in one direction and we can neither go backwards nor decrease the number of seconds it takes to perform what has to take a minute. If we ask educators to undertake good learning design, have engaging and interesting assignments, work on assessment levels well up in the taxonomies and we then ask them to spend day after day standing in front of a class and add marking on top?

Forget it. We know that we are going to sacrifice the number of tasks, the quality of the tasks or our own quality of life. (I’ve written a lot about time before, you can search my blog for time or read this, which is a good summary.) If our design was good, then sacrificing the number of tasks or their quality is going to compromise our design. If we stop getting sleep or seeing our families, our work is going to suffer and now our design is compromised by our inability to perform to our actual level of expertise!

When Henry Ford refused to work his assembly line workers beyond 40 hours because of the increased costs of mistakes in what were simple, mechanical, tasks, why do we keep insisting that complex, delicate, fragile and overwhelmingly cognitive activities benefit from us being tired, caffeine-propped, short-tempered zombies?

We’re not being honest. And thus we are not meeting our requirement for truth. A design that gets mangled for operational reasons without good redesign won’t achieve our outcomes. That’s not going to achieve our results – so that’s not good. But what of beauty?

A panel from the Morris Snakeshead textile showing flowers with interwoven branches and leaves, from the Arts and Crafts movement.

William Morris: Snakeshead Textile

What are the aesthetics of good work? In Petts’ essay on the Arts and Crafts movement, he speaks of William Morris, Dewey and Marx (it’s a delightful essay) and ties the notion of good work to work that is authentic, where such work has aesthetic consequences (unsurprisingly given that we were aiming for beauty), and that good (beautiful) work can be the result of human design if not directly the human hand. Petts makes an interesting statement, which I’m not sure Morris would let pass un-challenged. (But, of course, I like it.)

It is not only the work of the human hand that is visible in art but of human design. In beautiful machine-made objects we still can see the work of the “abstract artist”: such an individual controls his labor and tools as much as the handicraftsman beloved of Ruskin.

Jeffrey Petts, Good Work and Aesthetic Education: William Morris, the Arts and Crafts Movement, and Beyond, The Journal of Aesthetic Education, Vol. 42, No. 1 (Spring, 2008), page 36

Petts notes that it is interesting that Dewey’s own reflection on art does not acknowledge Morris especially when the Arts and Crafts’ focus on authenticity, necessary work and a dedication to vision seems to be a very suitable framework. As well, the Arts and Crafts movement focused on the rejection of the industrial and a return to traditional crafting techniques, including social reform, which should have resonated deeply with Dewey and his peers in the Pragmatists. However, Morris’ contribution as a Pragmatist aesthetic philosopher does not seem to be recognised and, to me, this speaks volumes of the unnecessary separation between cloister and loom, when theory can live in the pragmatic world and forms of practice can be well integrated into the notional abstract. (Through an Arts and Crafts lens, I would argue that there is are large differences between industrialised education and the provision, support and development of education using the advantages of technology but that is, very much, another long series of posts, involving both David Bowie and Gary Numan.)

But here is beauty. The educational designer who carries out good design and manages to hold on to enough of her time resources to execute the design well is more aesthetically pleasing in terms of any notion of creative good works. By going through a development process to stage evaluations, based on our assessment and learning environment plans, we have created “made objects” that reflect our intention and, if authentic, then they must be beautiful.

We now have a strong motivating factor to consider both the often over-looked design role of the educator as well as the (easier to perceive) roles of evaluation and intervention.


I’ve revisited the diagram from yesterday’s post to show the different roles during the execution of the course. Now you can clearly see that the course lecturer maintains involvement and, from our discussion above, is still actively contributing to the overall beauty of the course and, we would hope, it’s success as a learning activity. What I haven’t shown is the role of the E1 as designer prior to the course itself – but that’s another post.

Even where we are using mechanical or scripted human markers, the hand of the designer is still firmly on the tiller and it is that control that allows us to take a less active role in direct evaluation, while still achieving our goals.

Do I need to personally look at each of the many works all of my first years produce? In our biggest years, we had over 400 students! It is beyond the scale of one person and, much as I’d love to have 40 expert academics for that course, a surplus of E1 teaching staff is unlikely anytime soon. However, if I design the course correctly and I continue to monitor and evaluate the course, then the monster of scale that I have can be defeated, if I can make a successful argument that the E2 to E4 marker tiers are going to provide the levels of feedback, encouragement and detailed evaluation that are required at these large-scale years.

Tomorrow, we look at the details of this as it applies to a first-year programming course in the Processing language, using a media computation approach.

Four-tier assessment

We’ve looked at a classification of evaluators that matches our understanding of the complexity of the assessment tasks we could ask students to perform. If we want to look at this from an aesthetic framing then, as Dewey notes:

“By common consent, the Parthenon is a great work of art. Yet it has aesthetic standing only as the work becomes an experience for a human being.”

John Dewey, Art as Experience, Chapter 1, The Live Creature.

Having a classification of evaluators cannot be appreciated aesthetically unless we provide a way for it to be experienced. Our aesthetic framing demands an implementation that makes use of such an evaluator classification, applies to a problem where we can apply a pedagogical lens and then, finally, we can start to ask how aesthetically pleasing it is.

And this is what brings us to beauty.

A systematic allocation of tasks to these different evaluators should provide valid and reliable marking, assuming we’ve carried out our design phase correctly. But what about fairness, motivation or relevancy, the three points that we did not address previously? To be able to satisfy these aesthetic constraints, and to confirm the others, it now matters how we handle these evaluation phases because it’s not enough to be aware that some things are going to need different approaches, we have to create a learning environment to provide fairness, motivation and relevancy.

I’ve already argued that arbitrary deadlines are unfair, that extrinsic motivational factors are grossly inferior to those found within, and, in even earlier articles, that we too insist on the relevancy of the measurements that we have, rather than designing for relevancy and insisting on the measurements that we need.

To achieve all of this and to provide a framework that we can use to develop a sense of aesthetic satisfaction (and hence beauty), here is a brief description of a four-tier, penalty free, assessment.

Let’s say that, as part of our course design, we develop an assessment item, A1, that is one of the elements to provide evaluation coverage of one of the knowledge areas. (Thus, we can assume that A1 is not required to be achieved by itself to show mastery but I will come back to this in a later post.)

Recall that the marking groups are: E1, expert human markers; E2, trained or guided human markers; E3, complex automated marking; and E4, simple and mechanical automated marking.

A1 has four, inbuilt, course deadlines but rather than these being arbitrary reductions of mark, these reflect the availability of evaluation resource, a real limitation as we’ve already discussed. When the teacher sets these courses up, she develops an evaluation scheme for the most advanced aspects (E1, which is her in this case), an evaluation scheme that could be used by other markers or her (E2), an E3 acceptance test suite and some E4 tests for simplicity. She matches the aspects of the assignment to these evaluation groups, building from simple to complex, concrete to abstract, definite to ambiguous.

The overall assessment of work consists of the evaluation of four separate areas, associated with each of the evaluators. Individual components of the assessment build up towards the most complex but, for example, a student should usually have had to complete at least some of E4-evaluated work to be able to attempt E3.

Here’s a diagram of the overall pattern for evaluation and assessment.


The first deadline for the assignment is where all evaluation is available. If students provide their work by this time, the E1 will look at the work, after executing the automated mechanisms, first E4 then E3, and applying the E2 rubrics. If the student has actually answered some E1-level items, then the “top tier” E1 evaluator will look at that work and evaluate it. Regardless of whether there is E1 work or not, human-written feedback from the lecturer on everything will be provided if students get their work in at that point. This includes things that would be of help for all other levels. This is the richest form of feedback, it is the most useful to the students and, if we are going to use measures of performance, this is the point at which the most opportunities to demonstrate performance can occur.

This feedback will be provided in enough time that the students can modify their work to meet the next deadline, which is the availability of E2 markers. Now TAs or casuals are marking instead or the lecturer is now doing easier evaluation from a simpler rubric. These human markers still start by running the automated scripts, E4 then E3, to make sure that they can mark something in E2. They also provide feedback on everything in E2 to E4, sent out in time for students to make changes for the next deadline.

Now note carefully what’s going on here. Students will get useful feedback, which is great, but because we have these staggered deadlines, we can pass on important messages as we identify problems. If the class is struggling with key complex or more abstract elements, harder to fix and requiring more thought, we know about it quickly because we have front-loaded our labour.

Once we move down to the fully automated systems, we’re losing opportunities for rich and human feedback to students who have not yet submitted. However, we have a list of students who haven’t submitted, which is where we can allocate human labour, and we can encourage them to get work in, in time for the E3 “complicated” script. This E3 marking script remains open for the rest of the semester, to encourage students to do the work sometime ahead of the exam. At this point, the discretionary allocation of labour for feedback is possible, because the lecturer has done most of the hard work in E1 and E2 and should, with any luck, have far fewer evaluation activities for this particular assignment. (Other things may intrude, including other assignments, but we have time bounds on this one, which is better than we often have!)

Finally, at the end of the teaching time (in our parlance, a semester’s teaching will end then we will move to exams), we move the assessment to E4 marking only, giving students the ability (if required) to test their work to meet any “minimum performance” requirements you may have for their eligibility to sit the exam. Eventually, the requirement to enter a record of student performance in this course forces us to declare the assessment item closed.

This is totally transparent and it’s based on real resource limitations. Our restrictions have been put in place to improve student feedback opportunities and give them more guidance. We have also improved our own ability to predict our workload and to guide our resource requests, as well as allowing us to reuse some elements of automated scripts between assignments, without forcing us to regurgitate entire assignments. These deadlines are not arbitrary. They are not punitive. We have improved feedback and provided supportive approaches to encourage more work on assignments. We are able to get better insight into what our students are achieving, against our design, in a timely fashion. We can now see fairness, intrinsic motivation and relevance.

I’m not saying this is beautiful yet (I think I have more to prove to you) but I think this is much closer than many solutions that we are currently using. It’s not hiding anything, so it’s true. It does many things we know are great for students so it looks pretty good.

Tomorrow, we’ll look at whether such a complicated system is necessary for early years and, spoilers, I’ll explain a system for first year that uses peer assessment to provide a similar, but easier to scale, solution.

Four tiers of evaluators

We know that we can, and do, assess different levels of skill and knowledge. We know that we can, and do, often resort to testing memorisation, simple understanding and, sometimes, the application of the knowledge that we teach. We also know that the best evaluation of work tends to come from the teachers who know the most about the course and have the most experience, but we also know that these teachers have many demands on their time.

The principles of good assessment can be argued but we can probably agree upon a set much like this:

  1. Valid, based on the content. We should be evaluating things that we’ve taught.
  2. Reliable, in that our evaluations are consistent and return similar results for different evaluators, that re-evaluating would give the same result, that we’re not unintentionally raising or lowering difficulty.
  3. Fair.
  4. Motivating, in that we know how much influence feedback and encouragement have on students, so we should be maximising the motivation and, we hope, this should drive engagement.
  5. Finally, we want our assessment to be as relevant to us, in terms of being able to use the knowledge gained to improve or modify our courses, as it is to our student. Better things should come from having run this assessment.

Notice that nothing here says “We have to mark or give a grade”, yet we can all agree on these principles, and any scheme that adheres to them, as being a good set of characteristics to build upon. Let me label these as aesthetics of assessment, now let’s see if I can make something beautiful. Let me put together my shopping list.

  • Feedback is essential. We can see that. Let’s have lots of feedback and let’s put it in places where it can be the most help.
  • Contextual relevance is essential. We’re going to need good design and work out what we want to evaluate and then make sure we locate our assessment in the right place.
  • We want to encourage students. This means focusing on intrinsics and support, as well as well-articulated pathways to improvement.
  • We want to be fair and honest.
  • We don’t want to overload either the students or ourselves.
  • We want to allow enough time for reliable and fair evaluation of the work.

What are the resources we have?

  • Course syllabus
  • Course timetable
  • The teacher’s available time
  • TA or casual evaluation time, if available
  • Student time (for group work or individual work, including peer review)
  • Rubrics for evaluation.
  • Computerised/automated evaluation systems, to varying degree.

Wait, am I suggesting automated marking belongs in a beautiful marking system? Why, yes, I think it has a place, if we are going to look at those things we can measure mechanistically. Checking to see if someone has ticked the right box for a Bloom’s “remembering” level activity? Machine task. Checking to see if an essay has a lot of syntax or grammatical errors? Machine task. But we can build on that. We can use human markers and machine markers, in conjunction, to the best of their strengths and to overcome each other’s weaknesses.

Some cast-iron wheels and gears, connected with a bicycle chain.

We’ve come a long, in terms of machine-based evaluation. It doesn’t have to be steam-driven.

If we think about it, we really have four separate tiers of evaluators to draw upon, who have different levels of ability. These are:

  1. E1: The course designers and subject matter experts who have a deep understanding of the course and could, possibly with training, evaluate work and provide rich feedback.
  2. E2: Human evaluators who have received training or are following a rubric provided by the E1 evaluators. They are still human-level reasoners but are constrained in terms of breadth of interpretation. (It’s worth noting that peer assessment could fit in here, as well.)
  3. E3: High-level machine evaluation includes machine-based evaluation of work, which could include structural, sentiment or topic analysis, as well as running complicated acceptance tests that look for specific results, coverage of topics or, in the case of programming tasks, certain output in response to given input. The E3 evaluation mechanisms will require some work to set up but can provide evaluation of large classes in hours, rather than days.
  4. E4: Low-level machine evaluation, checking for conformity in terms of length of assignment, names, type of work submitted, plagiarism detection. In the case of programming assignments, E4 would check that the filenames were correct, that the code compiled and also may run some very basic acceptance tests. E4 evaluation mechanisms should be quick to set up and very quick to execute.

This separation clearly shows us a graded increase of expertise that corresponds to an increase of time spent and, unfortunately, a decrease in time available. E4 evaluation is very easy to set up and carry out but it’s not fantastic for detailed feedback or higher Bloom’s level. Yet we have an almost infinite amount of this marking time available. E1 markers will (we hope) give the best feedback but they take a long time and this immediately reduces the amount of time to be spent on other things. How do we handle this and select the best mix?

While we’re thinking about that, let’s see if we are meeting the aesthetics.

  1. Valid? Yes. We’ve looked at our design (we appear to have a design!) and we’ve specifically set up evaluation into different areas while thinking about outcomes, levels and areas that we care about.
  2. Reliable? Looks like it. E3 and E4 are automated and E2 has a defined marking rubric. E1 should also have guidelines but, if we’ve done our work properly in design, the majority of marks, if not all of them, are going to be assigned reliably.
  3. Fair? We’ve got multiple stages of evaluation but we haven’t yet said how we’re going to use this so we don’t have this one yet.
  4. Motivating? Hmm, we have the potential for a lot of feedback but we haven’t said how we’re using that, either. Don’t have this one either.
  5. Relevant to us and the students. No, for the same reasons as 3 and 4, we haven’t yet shown how this can be useful to us.

It looks like we’re half-way there. Tomorrow, we finish the job.


Not just videos!


Just a quick note that on-line learning is not just videos! I am a very strong advocate of active learning in my face-to-face practice and am working to compose on-line systems that will be as close to this as possible: learning and doing and building and thinking are all essential parts of the process.

Please, once again, check out Mark’s CACM blog on the 10 myths of teaching computer science. There’s great stuff here that extends everything I’m talking about with short video sequences and attention spans. I wrote something ages ago about not turning ‘chalk and talk’ into ‘watch and scratch (your head)’. It’s a little dated but I include it for completeness.

Collaboration and community are beautiful

There are many lessons to be learned from what is going on in the MOOC sector. The first is that we have a lot to learn, even for those of us who are committed to doing it ‘properly’ whatever that means. I’m not trying to convince you of “MOOC yes” or “MOOC no”. We can have that argument some other time. I’m talking about we already know from using these tools.

We’ve learned (again) that producing a broadcast video set of boring people reading the book at you in a monotone is, amazingly, not effective, no matter how fancy the platform. We know that MOOCs are predominantly taken by people who have already ‘succeeded’ at learning, often despite our educational system, and are thus not as likely to have an impact in traditionally disadvantaged areas, especially without an existing learning community and culture. (No references, you can Google all of this easily.)

We know that online communities can and do form. Ok, it’s not the same as twenty people in a room with you but our own work in this space confirms that you can have students experiencing a genuine feeling of belonging, facilitated through course design and forum interaction.

“Really?” you ask.

In a MOOC we ran with over 25,000 students, a student wrote a thank you note to us at the top of his code, for the final assignment. He had moved from non-coder to coder with us and had created some beautiful things. He left a note in his code because he thought that someone would read it. And we did. There is evidence of this everywhere in the forums and their code. No, we don’t have a face-to-face relationship. But we made them feel something and, from what we’ve seen so far, it doesn’t appear to be a bad something.

But we, as in the wider on-line community, have learned something else that is very important. Students in MOOCs often set their own expectations of achievement. They come in, find what they’re after, and leave, much like they are asking a question on Quora or StackExchange. Much like you check out reviews on-line before you start watching a show or you download one or two episodes to check it out. You know, 21st Century life.

Once you see that self-defined achievement and engagement, a lot of things about MOOCs, including drop rates and strange progression, suddenly make sense. As does the realisation that this is a total change from what we have accepted for centuries as desirable behaviour. This is something that we are going to have a lot of trouble fitting into our existing system. It also indicates how much work we’re going to have to do in order to bring in traditionally disadvantaged communities, first-in-family and any other under-represented group. Because they may still believe that we’re offering Perry’s nightmare in on-line form: serried ranks with computers screaming facts at you.

We offer our students a lot of choice but, as Universities, we mostly work on the idea of ‘follow this program to achieve this qualification’. Despite notionally being in the business of knowledge for the sake of knowledge, our non-award and ‘not for credit’ courses are dwarfed in enrolments by the ‘follow the track, get a prize’ streams. And that, of course, is where the diminishing bags of dollars come from. That’s why retention is such a hot-button issue at Universities because even 1% more retained students is worth millions to most Universities. A hunt and peck community? We don’t even know what retention looks like in that context.

Pretending that this isn’t happening is ignoring evidence. It’s self-deceptive, disingenuous, hypocritical (for we are supposed to be the evidence junkies) and, once again, we have a failure of educational aesthetics. Giving people what they don’t want isn’t good. Pretending that they just don’t know what’s good for them is really not being truthful. That’s three Socratic strikes: you’re out.

The Eiffel Tower, Paris, at night being struck at the apex by three bolts of lightning simultaneously.

We had better be ready to redirect that energy or explode.

We have a message from our learning community. They want some control. We have to be aware that, if we really want them to do something, they have to feel that it’s necessary. (So much research supports this.) By letting them run around in the MOOC space, artificial and heavily instrumented, we can finally see what they’re up to without having to follow them around with clipboards. We see them on the massive scale, individuals and aggregates. Remember, on average these are graduates; these are students who have already been through our machine and come out. These are the last people, if we’ve convinced them of the rightness of our structure, who should be rocking the boat and wanting to try something different. Unless, of course, we haven’t quite been meeting their true needs all these years.

I often say that the problem we have with MOOC enrolments is that we can see all of them. There is no ‘peeking around the door’ in a MOOC. You’re in or you’re out, in order to be signed up for access or updates.

If we were collaborating with all of our students to produce learning materials and structures, not just the subset who go into MOOC, I wonder what we would end up turning out? We still need to apply our knowledge of pedagogy and psychology, of course, to temper desire with what works but I suspect that we should be collaborating with our learner community in a far more open way. Everywhere else, technology is changing the relationship between supplier and consumer. Name any other industry and we can probably find a new model where consumers get more choice, more knowledge and more power.

No-one (sensible) is saying we should raze the Universities overnight. I keep being told that allowing more student control is going to lead to terrible things but, frankly, I don’t believe it and I don’t think we have enough evidence to stop us from at least exploring this path. I think it’s scary, yes. I think it’s going to challenge how we think about tertiary education, absolutely. I also think that we need to work out how we can bring together the best of face-to-face with the best of on-line, for the most people, in the most educationally beautiful way. Because anything else just isn’t that beautiful.