What that online course can't offer

“Happy learning!” Andrew Ng, the founder of the online learning platform Coursera and the AI learning platform deeplearning.ai, ends his weekly newsletter, The Batch, with this cheery sign-off.

Ng has long been a learning enthusiast. Through Coursera and deeplearning.ai, he has focused on making the learning experience scalable and accessible. What motivated him, he once said, was the frustration he felt as a professor at Stanford, when he would deliver the same lectures, complete with the same jokes, year after year. He felt that his time was better spent building relationships with students.

In August 2011, he released the first set of lecture videos that formed the basis for one of the world’s first Massive Open Online Courses (MOOCs). By today’s standards, the technology was rudimentary: Ng, filmed live in a lecture hall, speaks to a Logitech webcam while wearing a simple lapel microphone. The lectures remained online after the course, free for others to view. Lecture 1 now has 1.1m YouTube views. It’s safe to say that Ng has never had to give the same lectures again.

An entire education industry has emerged around the idea of offering learning experiences to anyone with an internet connection, often at low prices (or for free). Even more appealing is another value proposition: better career prospects. That’s what got me interested.

I was about to complete my Bachelor’s degree in the UK. The previous few years of study had improved my thinking and research skills; they had given me an appreciation for well-articulated ideas. But the job descriptions I had seen worried me. Companies weren’t after “good thinking skills”, let alone “appreciation for knowledge”. Instead, they would list bullet points like “create Excel macros with VBA”, or “experience with C++”, and, in extreme cases, “Ninja coding skills”.

I most certainly did not have these skills. As a graduate from the life sciences, I knew how to break down an academic paper and craft a research question, but those abilities paled in comparison to something concrete like coding a website or managing a database. The most relevant class I had taken was a short course for scientists in statistics and R programming, a language for statistical computing and graphics. That did not end well.

Our professor, bless him, had tried to explain some basics by making us count M&Ms.  “Statistics will be the most transferable skill from your entire degree!” the lecturer flung helplessly at us as we stomped out of the lecture hall, grabbing handfuls of M&Ms as we left but with no grasp of statistics.

A year later, his parting words rang true. There I was, browsing Coursera and trying to learn more about this field called statistical data analysis and this thing called programming because they seemed to be something tangible that might help me get a job.

I wasn’t a stranger to self-reliant learning, or to bypassing what structured educational institutions offered. In school, memorising scientific facts and diagrams was tedious for me. But my dad passed on an old microscope and we spent afternoons working the magnification knobs, preparing slides with drops of aquarium water, and watching wriggling micro-organisms from the eyepiece. I kept a pet ladybird I observed daily. I made friends with the toads nesting in the gnarled and ancient trees at school. I believed knowledge and skills could be cultivated anywhere, not necessarily in a classroom.

Still, as I was about to find out, self-reliance, curiosity and taking one or two online courses are necessary, but not enough, to be hired. One needs community, deliberate practice and, most of all, consistency, to build the hard skills valued by the job market.

“Every Saturday, you will have a choice between staying at home and reading research papers versus watching TV. If you spend all Saturday working, there probably won't be any short-term reward, and your current boss won't even know or say "nice work". Also, after that Saturday of hard work, you're not actually that much better at machine learning. But here's the secret: if you do this not just for one weekend, but instead study consistently for a year, then you will become very good.

But the most important thing is to keep on learning. Not just for a few months, but for years.”

The hard part about dieting is not losing weight. The hard part is changing behaviours that keep the weight off, consistently, for a long time. Learning is similar. Spending a month on one course is easy. To take another course and then another, and then build a portfolio of projects, and then join a study group, and then finally score a job interview that opens the door to a better career— that’s harder.

Nevertheless, continuous learning is becoming a non-negotiable practice. In a suitably titled report, “Careers and learning: real time, all the time”, Deloitte, a consultancy, estimates the “half-life of a learned skill” as approximately five years. SkillsFuture, Singapore’s continual learning unit, cites an estimate that a worker with a 30-year career can expect to reinvent themselves up to six times. Workers can no longer rely on a single degree to equip them for a lifetime of employment.

This presents both a problem and an opportunity. The World Economic Forum (WEF) estimates that automation may displace 85m current jobs but might create 97m new roles, which could require reskilling and upskilling. Career “pivots”, or transitions into a “wholly new role occupation” will become commonplace. WEF separately found that 72 percent of transitions into the Data and AI industry originate from a completely different job family (e.g. former economists or political scientists). Professions in Data and AI are new and still evolving, and thus have fewer fixed requirements.

Against this backdrop, MOOCs occupy a special niche. They can deliver upskilling content, at scale and low cost, to a wide audience. Over the years, I’ve discussed the MOOC experience with a diverse set of people, especially in relation to their commitments like childcare or full-time work.

We had signed up for courses like Introduction to Data Analysis or Introduction to HTML aspiring to improve our career prospects. Some wanted a complete career restart, others a skills refresher. Some were simply curious, using their Skills Futures credits to explore new concepts like “Data Analysis” and “Artificial Intelligence”.

We were united by a tacit recognition of the reality that Ng described—consistency is hard.

We each had our own study hacks. Some snuck in three hours on a Sunday afternoon to finish an online module, others preferred to stay up late. Some got up early, while others bought generous mobile data plans so they could watch lectures while commuting.

What worked for me was a very specific, almost ritualistic, weekend morning routine that began at 6:30am when the meowing “alarm” rang. I would feed my cats, tidy up my desk, brew coffee, and sit down to work for a few hours before my brain had time to protest. Starting early was good: I could get something done and still take a few hours later in the day to meet friends or go for a nice lunch. I didn’t feel too deprived.

After cost and accessibility, time and motivation are the biggest barriers to a student successfully finishing a course. Signing up for a certified four-week course only takes a few clicks and S$66 a month. But grinding through course material after a long day at work is exhausting. Weekends may offer larger chunks of free time, but in reality, a certain rebelliousness usually kicks in (I’ve worked so hard this week, surely I deserve some rest). Sunday evening comes around and the course completion progress bar remains at zero percent.

In theory, one can learn everything from rhetoric to public health to Java programming. But in reality, one is only a human with finite amounts of attention and energy.

“Just-in-time” learning delivers material in engaging and bite-sized modules. None are too complex; all are available immediately when the need arises. With this model, knowledge does not have to be swallowed, python-like, through a comprehensive two-hour, in-person lecture.

Instead, it can be chunked into a series of 10-minute videos, each addressing a discrete learning outcome. Students can pick courses based on their needs, login at a suitable time, and then learn just enough to solve an immediate problem. It’s a model more like snacking than a 10-course dinner. It recognises certain realities of the modern workplace—an employee gets paid to solve problems, most of which do not require deep background knowledge.

Short courses first allowed me to quickly explore the wide range of options (data visualisation, data science, software engineering) available in the very broad field that is Data. Later, when my work evolved away from data analysis towards engineering and computer science, short courses familiarised me with other programming languages and tools.

This form of just-in-time learning felt like relying on an index to look up information when I needed it, rather than being programmed in one go with an entire library’s worth of information. I loved it.

But crossing the cliff between beginner and practitioner is not straightforward. Learners have to navigate a mix of books, self-driven projects, and more courses. In response to this learning cliff, some edu-tech platforms' response is: more technology. At DataCamp, for example, insights from research into skills mastery—working just beyond your current level of competency, receiving tailored feedback—are incorporated as yet more platform features.

For example, data is collected about common learning mistakes. On questions where enough data is collected, errors can come included with helpful hints, prompting a user to rethink where they went wrong and eventually, help them finish their assignments. One of my pre-Covid commute routines was casually swiping through exercises on the bus. Those quick bursts of repetition did more to train me in programming concepts than the hours spent reading conceptual descriptions from a textbook.

This way of learning was beneficial to me because there was so much about computing that I didn’t know. Things like algorithms, which may have been intuitive to a graduate in mathematics or engineering, were new to me. So too other skills, like monitoring the resources your computer is using, which would have been familiar to a gamer.

It was like appearing in an alternate universe where beings communicated by waving their trunks instead of by talking—assimilating meant bending my brain to fit an entirely new mental model. By taking in new ideas slowly, and in small chunks, I was able to gradually build a foundation that would serve me in the long term.

Indeed, this blend of research-based pedagogy and thoughtful interface design is a promising way of developing skills in a digital setting. The individualised, learn-at-your-own pace style is kinder and gentler, especially when compared to the traditional classroom. There, a student who falls behind usually stays behind, unable to take the time they need to work through misconceptions and misunderstandings.

Nevertheless, such metric-driven methods lend themselves better to quantitative, technical skills compared to softer skills such as those involved in business negotiation or leadership. There can be negative implications to focusing too much on the measurable.

Specifically, to only focus on in-demand technical skills is to miss out on broader work trends. Burning Glass Technologies, a labour market analytics firm, describes the “Blended Digital Professional”: a worker who combines human skills (communication, creativity, relationship-building) with digital skills (data analysis, computer programming) and domain experience (exposure to a particular industry such as manufacturing or healthcare).

“Just a few years ago, no one would have thought advertising and marketing managers would need to configure software, or that software developers would need the skills to build rapport with customers. But advancing technology is creating both new opportunities for workers and new challenges for employers trying to fill roles,” explains David Jones, senior managing director of Robert Half Asia Pacific, a human resource consultancy.

“Learning is a process,” says David Kolb in Experiential Learning. “Each act of understanding is the result of a process of continuous construction and invention.” In Kolb’s framework, a learner moves through a four-stage process: living through a concrete experience, reflecting on that experience, learning from it by conceptualising abstract ideas from the concrete, and finally actively experimenting with what they have learnt.

That’s indeed how my most effective learning experiences have progressed. During my first coding tutorial, I downloaded a free code editor and could immediately get started with a hands-on exercise. Although I promptly got stuck for the next two hours (my script was missing two barely noticeable brackets), there were enough internet resources to eventually explain my errors. When my script was finally running, I could then reflect on my mistakes and build a mental model for how computers worked. Finally, I could rewrite my code, run it again, see a different error, and do another rewrite.

It was this cycle, repeated again and again that helped me make a crucial transition—moving from being hand-held through an online learning module, to being self-sufficient as a coder.

Another important factor was a peer group. Early on, I got in touch with a group of other women also learning to code. We met weekly for picnics and study sessions. We complained to each other and exchanged encouraging words (and cake and chocolate).

Some were librarians looking to better manage their information systems, some were marketing analysts trying to be more quantitative, others were physics PhDs who wanted to move into the technology industry. We were united in our diverse backgrounds. We gave each other a different perspective on technical concepts, job hunting, and being a minority in the industry.

Josh Waitzkin is an American chess prodigy who later became world champion in competitive Taichi. In The Art of Learning, he recognises the people who helped him in his journey: “In the two years before the 2004 Taiwan tournament, Dan and I basically lived on the mats together. Some nights we were drilling techniques...in other sessions we were refining footwork.”

Separately he recalls learning to play chess from veterans at Washington Square Park: “The guys took me under their wings, showed me their tricks, taught me how to generate devastating attacks and get into the head of my opponent.”

As recognised by social learning theory, we don’t learn only by reading a book or watching a video on our own. Learning also occurs through interacting with others and observing them.

Sometimes, when the digital learning debate focuses on concrete concerns such as on-demand skills and job security, the social and emotive aspects of learning can be lost.

A week after I completed some short course software tests, I got pulled into a work call. During the call, I was hit with a question. I gave a passable answer straight from the course material, but a follow-up question pushed the limits of my knowledge. I promised to do more research. After the meeting, I scrolled through our internal knowledge base and ran my questions by a few trusted colleagues. I ran a few experiments myself, and then added what I wrote to a presentation I shared with my co-workers.

This continuous cycle of challenging and reworking my mental model is what really solidifies the technical ideas that I’ve previously had a surface-level knowledge of. More often than not, this experience happens not through an instructional video, but in the trenches of day-to-day problem solving. It’s a messy journey. It doesn’t fall neatly into a list of course outcomes with green ticks next to them or culminate in a badge I can pin to my LinkedIn Profile.

But the encounter does leave me with a bone-deep sense of victory. It’s a feeling that keeps me wanting to learn more, to keep up with the latest knowledge, and to stay happy learning.

Rachel Lee is the pen name of a female data scientist working in Singapore. Jom respects her decision to use a pen name because of certain professional sensitivities.