In Defense of MOOCs, Discipline in the Attention Economy, and ChatGPT
Last time I shared about my experiences doing deep dives in technical areas for my work at Actuate Innovation and Department of Energy ARPA-E. But my journey with learning online actually started earlier in life, with Massive Open Online Courses (MOOC’s). These unfortunate enterprises have been much maligned publically, for their low completion rate, perpetuating low-quality education via certification, and lack of a business model. But for me, they’ve had a large role in shaping my trajectory and how I learn. I’ll share about my experience with them growing up, how I still use them today, and some reflections on why they haven’t taken off.
When I started taking chemistry in high school, I knew chemistry was a field I was interested in pursuing, so I wanted to do well in the class and actually understand the material. I looked for online intro chemistry college classes and stumbled across MIT’s OpenCourseWare. Anyone that wants to understand the business model of a university needs to account for the fact that places like MIT put all of their lectures, homeworks, and notes online for free and people still pay money to go there.
The first class I took on MIT OCW was “Intro to Solid State Chemistry” by Donald Sadoway. It speaks to the power and reach of digital connectivity that I can claim a MIT professor as one of the most influential teachers in my high school career - his class notes and lectures, which always ended with a historical story that utilized the principles of the lecture, where what sparked my love in material science (and he was an electrochemist before it was cool)!
I was also interested in solar as a high schooler, so I found Tonio Buonassisi’s course “Fundamentals of Photovoltaics” on MIT OCW as well. At the time, I was trying to brute-force learn my way through research papers on perovskite solar cells, and while it is possible to slowly grasp foreign concepts through contextualized correlations, it was much faster and simpler to have organizing frameworks and different spectroscopies explained from first principles.
Looking for a good online course quickly became a first-step initialization whenever I had to learn something that I knew I wanted to invest in and deeply understand. When I took linear algebra in high school, I supplemented it with Gilbert Strang’s Linear Algebra class from MIT OCW the summer before. When I started learning programming, I bought a book (which wasn’t very helpful) and used Coursera’s Python for Everybody and Web Design for Everybody (which also weren’t terribly helpful, but did get me comfortable with seeing code and the different languages used e.g. Python, HTML, CSS).
Throughout high school, I was able to supplement my high school education with these freely available online courses, which were often of a higher quality and richer in knowledge than what I could receive from high school alone. These online classes were one of the major factors why I did well in high school academically and in many ways, laid the foundation for many parts of my knowledge base. I was able to develop familiarity in technical topics that prepared me well for college and are still useful in shaping how I do first-principles thinking today.
Of course, as an EECS major at UC Berkeley, every class I took was basically a MOOC. The lectures, the discussion sheets, and the homework were all on course websites. Some of the best hits include Denero’s CS61A, Hug’s CS61B, and CS188 (which has really cute lecture slides).
Though I was busier in college, I did take Andrew Ng’s canonical Deep Learning Specialization on Coursera during my freshman year summer, which provided me the basic knowledge to start a career in AI (though in hindsight, I probably specialized too early).
The only things that changed from high school educationally at least were that I had TA’s and classmates to do things with, and I actually got a piece of paper at the end of the journey that other people cared about.
Reflections on MOOCs and the Discipline Gap
MOOCs and online university courses are aggregated pools of knowledge, usually presented in video format. This is in comparison to the hidden, often murky and unorganized knowledge embedded in text format (usually in PDF’s) that I focused on previously. Here’s how I think about MOOC’s in the context of an online learning resouce:
risks: sometimes they’re poorly run or, like any academic institution, have particular biases and focuses on what they teach
cost: a fair amount of up-front commitment, but (mostly) free and openly accessible
value: builds a foundational knowledge base before diving into deep technical subjects. Having content presented in a well-organized, contextualized manner is also pretty rare outside of school and can be quite valuable (and save a lot of time). Solid source for introductory overview of technical fields.
form: I honestly prefer video lectures over reading PDF’s
Of course, some of you might think, “But Charles, even if they worked for you, MOOCs clearly don’t work for most people - they have terrible completion rates! We need [insert some ed-tech AI startup idea]”. And sure, I can pretty easily be convinced that MOOCs and the traditional hour-long university lecture are a suboptimal medium for learning.
But that actually misses the larger systemic issue with why MOOCs haven’t taken off. Namely, that in the attention economy, capital-driven forces strive to capture attention. Tiktok and Reels are the latest in a decade-long evolution towards ever more addicting and captivating content forms to maximize user attention. And this is what MOOCs and online education are always in competition with.
Many thought leaders in the 2000’s liked to talk about how the internet was a force for equality. Access to education and economically valuable skills, which used to come with class and wealth, are now evenly accessible to all. But in reality, there is a new divide, a new imbalance created by the attention economy, one thats invisible, but certainly still present: what I’m terming the “discipline gap”. There are those who have the discipline to pull themselves away from the honeypots of social media or engage with them in a healthy manner and instead utilize the vast pools of digital content out there to leverage themselves up into higher skill domains, and those who don’t.
(I don’t want to imply the “discipline gap” is a new libertarian schtick that uses the internet as a vehicle to place the sole burden of responsibility for success or failure on the individual.One’s utilization of MOOCs and online education is a strong function of broadband access, household stability, and implicit local community practices around digital media. The ability to apply those skills and credentials is also contingent on the opportunities available, opportunities which I was uniquely fortunate to be afforded.
It’s a common saying that financial wisdom is an intergenerational asset e.g. parents telling you about S&P 500 annualized returns, maxing out 401k, and how to manage credit cards and debt. I’m convinced that familial practices around social media and “digital wisdom” e.g. not getting a smartphone early on, managing screen time, etc are going to become similar intergenerational inheritances, if not quite as easily measurable as financial wealth.)
ChatGPT and the Erasure of Instrumental Ends in Learning
Considering the perspective of discipline with respect to technology provides a framework to incorporate advances like ChatGPT as well. In today’s education system, we use test-driven education to incentivize “learning” i.e. “be able to pass this test or else”, reducing learning to an instrumental end. This is opposed to a system where we inculcate a desire for learning, where students strive and desire to learn for its intrinsic value.
The issue is that the “internet” i.e. wikipedia, wolfram alpha, and now ChatGPT are erasing these instrumentally created ends. Why learn derivatives when wolfram can do it? Why write argumentative essays about anything when ChatGPT can do it? The tension between test-driven learning and the reality that doing well on these tests as an end is irrelevant is being stretched with every new AI model.
The reward horizon of learning, where you actually use the skills you learn, is being pushed farther and farther away, until to most kids, all of secondary education is pointless. The end-goal and utility of education is still the same (we learn mutiplication and derivatives because they are useful skills later on), but the journey there is filled with far more temptations than before. Technology has brought the collective weight of human learning into the palm of our hand but it has not solved the problem of the human condition — sloth, anger, addiction — or does it contain the solution — self-control, discipline, and conviction. For those, we must look elsewhere.
Yes this means I was on MIT OCW before the website updated. Honestly, kinda miss the janky 2000’s HTML website, I’ve started to associate bad website design with hidden knowledge.
This was also the point where I started exploring MOOCs, like Coursera’s Organic Solar Cell class hosted by Technical University of Denmark, another great course, if somewhat out of date by now.
This is also somewhat analogous to how LLM AI’s learn compared to symbolic AI. LLM are learning probabilistic correlations in language patterns, which is essentially what I was trying to do by reading a ton of papers. But my learning was much more data-efficient if I had an organizing framework and was able to define the concepts first.
Someone who has read more Foucault than me should tell me if there’s some connection here between discipline in attention economies and power.
This is similar to Thi Ngyuen’s distinction between instrumental ends and striving play in the context of board games. I can’t recommend his book “Games: Agency as Art” (and Matthew Jordan’s review of it) enough!