As I see it, the critical problem of education today remains the same as it has always been: access to expert-level teachers. These are individuals with expertise in both the discipline under study and the discipline of teaching. I’ll go into greater depth on a definition of expert teacher at the end (go to definition).
I’ve thought carefully about the role of technology in education. At present, it is suitable only as an aid in teaching and learning — and primarily to collect and aggregate practice and performance data for display to the learner and teacher. In its present form, it simply cannot displace expert teachers. It is possible that with present technologies, using more and better practice and performance data, we may be able to train unsupervised learning algorithms which can then make recommendations to the learner on par with an expert teacher (though that is not their only function).
Reinforcement learning algorithms already teach themselves to reach superhuman performance levels (most recently and notably, in adversarial games such as Go, DotA2, and Heads-up, No-Limit Poker). However, as Hod Lipson at Columbia pointed out, the missing bridge is that the machine must still be taught how to teach a human. And that is no simple matter.
Then there is basically the same teaching experience, just widely distributed via video & text: MOOCs, Youtube videos, Tutorials, etc. Clearly these are very good for proliferating information at the level of technique, for a particular set of learners with a given set of existing knowledge, skills, and habits of mind (not the least of which being self-discipline). I can find endless MOOCs, Articles, Videos, GIFs, Homework samples, etc… which will detail a relatively consistent list of standard CSci data structures and algorithms (i.e. Arrays, Heaps, … Binary search, Bubble sort, etc.)
However, there are 3 big problems:
1. As of now, there is no standard, scalable assessment of the learner’s capabilities and understanding feeding back into the courseware. The Force Concept Inventory (FCI), Bennett Test (for mechanical intuition), et. al are reasonable approximations to assessment of understanding for their particular domains (though narrow). But the question remains… what do you serve up as a recommendation based on the results of the FCI? There are a LOT more data points needed on the learner’s existing knowledge and skills… and the content, sequence, and quality of presentation of the available courses, videos, articles, etc. I.e. We need to matchmake between people and learning materials, but we don’t appear to have the models for doing so. The result is that we simply presume a fairly uniform student body… and that produces the conditions of problem #2 — Learning-as-job.
Aside: A couple quick hypotheses re: this apparent dearth in data…
(1a) The institutions which are potentially capable of capturing the necessary data simply have elected not to implement assessment of understanding and haven’t been able to apply the data from their existing, almost entirely “quiz-style” approaches into practical use (Udemy, Coursera, Khanacademy, Udacity, and of course Knewton, etc.) — Remarkably, even the wide spread of mobile apps for language learning: DuoLingo, MemRise, Drops, BeeLingua… seem to generate zero insight into what you understand, only which quiz questions you’ve gotten right or wrong and thus which ones you need to continue working on. You are left to draw your own conclusions as to why — and whether it matters.
(1b) The data collection approaches are simply too complex to conceive of using present computing technologies. I.e. A valid assessment of understanding for learners of infinitely-varied (and largely unknown) backgrounds requires a lot of “hand-crafting”, rendering it an exercise in futility. A lot of work for little reward, and thus apparently easier to simply train humans for the task, particularly given we already have an existing infrastructure for employing humans in the task. There’s some evidence of this in the records of DoD contracts for development of a computer-based tutor for Navy IT personnel. Despite the reports of apparent superhuman success, the exercise took several multi-million dollar contracts to develop even one such tutor program for a particular class of learner, and a highly-specific domain area. And it’s also not clear if the technology is competitive in a more longitudinal study, particularly in comparison to expert teachers.
2. Learning-as-job (i.e. Pay attention because this will be on the test) is fundamentally ineffective and will naturally give way to entertainment, mischief, and (in the best of cases) more pleasurable learning activities. Achievement goes up with the square of learner’s Attentiveness. I.e. With even reasonably high levels of attention (incl. note-taking) during a lecture, you may still accomplish very little — pick up a few prominent facts and prominent storyline elements (hence why story is a particularly effective channel for conveying ideas). Insight, however, only comes through reflection, examination, and the effortful attempt to make well-founded assertions and ask the poignant questions. The PhD dissertation is the culmination of such effortful learning activity in a given domain.
This is an issue with children’s MOOCs especially, like Renaissance’s Achieve3000 (which produces a daily set of Current Event and subject matter articles, written at multiple Reading Levels for differentiated instruction based on learner’s present reading level). They are destined to occupy the low-attentiveness regime for most learners — except those who are already capable of self-governance and can likely identify which reading materials are suitable for them.
Machines as yet can only engage the human mind through carefully produced storylines (crafted, of course, by humans) in movies and games. And there are shining lights in both genres. Euclidea, for example, is a mobile app for uncovering geometric principles from their elementary origins following a sequence of progressively advancing puzzles. However, again, this app was not built for children… It is VERY difficult to build a universally-accessible learning instrument. At a very minimum, you will presently require a human to guide you between the various learning technologies — though they will likely need to fill in other gaps along the way. And movies, of course, are not interactive, though they can move a person to reflection, if done properly — and if the viewer’s cognitive & social environment supports it. VR has the potential to produce its own share of shining lights — but again, I do not expect it will “revolutionize” learning due to the same scaling limitations as with the other genres. Screen-based video-gaming is sufficiently immersive to keep people playing for 12+hrs without break.. I don’t think the immersiveness is the problem.
3. The economic pathway and social and political will for development of better teaching technologies is, in my view, troubled. Public education as an investment by the nation into the capabilities of its next generation likely sounds, to modern ears, like an antiquated value and probably one made up in a 1984-esque “rewriting history” exercise — “back in the day, we used to be better people…” And it’s even less practical to a generation of anti-aging enthusiasts (hey, I’m one of them — I agree with Harari in his intro to Homo Deus… aging is a technical problem — complex, but solvable with better models and technologies) with a historically low fertility rate. Why should we make our lives harder on behalf of this upcoming generation? Given we haven’t stopped death by aging, we can all plan to be gone in roughly 100 years, so it’s a crazy question to be asking. But obviously we have a history of mortgaging the future on behalf of the present.
The parameters of the best-case solution for my kids are as follows:
1. Expert Teacher with the objective to cultivate, from the outset, domain expertise — Not to entertain or to occupy the child with activities. Rather, this would be a carefully-designed course of study, intended to result in mastery of the domain, beginning with first exposure to the domain at a young age. The child must, from the very beginning of a dedicated learning effort, understand that he or she is doing something very important — different from other things. Kids can readily begin to make that distinction if they are trained to do so early, and they will then have the capacity to employ their highest attention to the learning task. And under high-engagement conditions, children learn quickly and almost effortlessly. Many parents make a critical error of omission early in the child’s development by not cultivating a ritual for activating a child’s command of their own attention. Montessori, of course, used the 2’x3′ rectangular mat as a standard cue.
NOTE: I’ve included a few indicators for recognizing an expert teacher in contrast to an average teacher in the bullets below.
2. Non-school-building learning environment: Learners should not be subjected to extended isolation from the subject matter “in the wild”, particularly in the company of largely disengaged peers. Would you study flower anatomy without a garden? Would you find it easier to seek deep insight if the other members of your social fabric (peers are an unavoidably significant recipient of the brain’s allocation of attention) were chiefly concerned with the flower’s utility as an aid in a game of “He loves me not”? Our ongoing attachment to the practice of shuttling people in and out of the confines of a school building for a daily dose of “regression to the mean” suggests the architects of the system must prize above all else (including development of expertise) the logistical benefits of an aggregated population. Co-location would be limited only to those circumstances necessitating physical propinquity, and simulated environments would be used as a precursor to an analogous activity in the wild. Other activities can be done remotely, for which data-transmission technologies are already reasonably well-suited.
3. From youth, develop an expanding skill set to provide services to an expanding circle of influence. The best way to consolidate understanding of concepts is to attempt to deploy them within a design. Children who rush out to newly-fallen snow to build snow forts and snowmen are often flummoxed when the snow simply won’t hold together. “How does anyone ever make a snowman?” the child wonders.
Naturally, playing outdoors affords endless opportunities to “try and see”, and it’s a critical component of human development from start to finish. I.e. You never stop benefiting from playing outdoors.
However, while “try and see” will not produce expertise. With that objective, the only way a child will be afforded the opportunity to instantiate a design on a real-world system is under the conditions of a labor exchange. I.e. The child removes some part of the total labor for a particular job so the expert practitioner can allocate time and energy to training you and providing feedback. Participation with real people in a task of real importance creates an inarguable need for more complete conceptual models, better-honed skills, and more effective execution. This, of course, is the standard apprenticeship model.
This is harder and harder to do as the jobs which previously would have been performed by the novices are now performed by machines. That means the child must have more (but not inordinately so, depending on the context) background capabilities to provide useful service in the real-world. Reading, Writing, and basic Arithmetic are certainly in the set of prerequisites, as communication of instructions will often be in text form.
4. Cultivate a circle of peers from early on — and build friendships with work capacity. An “immune system” of buddies who can help you out when you need it. Preferably, each member of the team would be cultivating a unique specialization. The peer group then becomes yet another venue for consolidation of understanding, in which the teacher is absent and the child must employ his or her knowledge and skills under different, more difficult circumstances — but not entirely alone.
5. Hybridization as the answer to Breadth vs Depth. The idea is to learn to pair skill sets both within a single individual and between individuals. This is actually one area about which I haven’t seen much written. There’s plenty on the concept of transfer and analogy, of course, but less on the intentional design of a hybridized expertise. Instead, any hybrid becomes its own thing. Like the Biologist who also studies and manipulates computational models and data. That’s now called Computational Biology. The question, of course, is whether either domain is receiving sufficient attention. A hybridized learning model would sequence learning operations with transfer from one domain to the other as the pedagogical objective — rather than treading the bisecting line between the two vectors.
Hybridization is the holy grail for project-based learning schools — The whole idea of the project is that it has to check many boxes to be sufficiently satisfying to all stakeholders that the child won’t be hindered from ivy league aspirations by this alternative track. Again… presently, we rely heavily on the attributes of the Expert Teacher.
As promised in #1 above, here are a few indicators of an Expert Teacher…
1. They tend to (at times frustratingly) emphasize things that initially seem irrelevant or unnecessary — but which will shortly prove essential. And they de-emphasize those things which appear critical or have risen to popular prominence but are, in fact, irrelevant or even harmful.
In seconds to minutes, an expert teacher can assess the developmental stage of an individual in a particular skill domain and impart to them a key to unlock new insight and growth. Alone, or under lesser tutelage, the beginner might have sought this key indefinitely, without substantive progress — the beginner may see the key right in front of him but will not know its worth nor how to fit it to the lock.
In contrast, average teachers tend to follow an all-inclusive course of study — and just as well, as deviations from that course tend to decrease clarity.
This average teacher may be an enthusiastic youth (new undergrad or grad student), still seeing the many exciting elements of their domain of study for the first time, ready to spill out their insights en masse. These individuals will typically narrow in on the more resonant pupils, who are intrigued by, and able to follow, the scattershot of concepts and ideas. This is a useful and invigorating relationship for a more advanced pupil to cultivate. However, you would not rely on the “enthusiastic youth” to instruct a child from their early stages. The enthusiasm will soon wear off in the face of the much slower pace of an early learner.
Another average teacher is the long-time practitioner who has limited their excursions into the either one or both of their required domain areas. Typically because they themselves did not receive adequate tutelage in the subject matter and they never adapted an autodidactic approach. Average teachers of this sort will typically de-emphasize the cultivation of expertise, as it is a perpetually unsettling enterprise. They will instead promote the social aspects of the “teacher-pupil” relationship, as a proxy metric for success in the teaching profession. E.g. “I love to see smiling faces.” These individuals do have the potential to do their pupils good as a “caring adult”, but they will likely also do harm by operating the course “by the book” with minimal engagement and minimal insight — possibly forever casting for those students an impression of the domain as dull and lifeless.
The enthusiastic youth produces outcomes of high-variance — i.e. Some students are highly engaged while others are entirely put off. The long-time non-expert technician produces outcomes of low-variance, but low-accomplishment, and possibly attenuated ongoing interest. The expert, on the other hand, will see most students out of the course with a set of well-consolidated foundational concepts — and there will often be a few miracle students: Either very high achievers or previously low-achievers with remarkable leaps into high-achievement.
2. Expert-level teachers typically think and teach at the method-level vs technique-level. You’ll hear them continuously relating new concepts to old via accessible analogies, illustrating concepts with insightful imagery, and turning concepts over to look at them entirely differently than a beginner could imagine.
In contrast, average teachers typically have yet to consolidate the system of ideas that compose a discipline and navigate it ponderously and haltingly. The human mind requires incubation time to consolidate ideas, and typically the average teachers have not had sufficient incubation time. Either because they are young or because they have primarily been technicians of the teaching trade, purveying roughly the same courses for the extant of their career, without ongoing development of their teaching practice and the domain under study.
3. Expert-level teachers take the long view and are typically unmoved by fads. This can be their undoing, of course — fads are not always, in fact, fads. However, they will make far fewer missteps by avoiding fads than they will by embracing them. Thus, you will typically find the expert referencing older ideas and referring to the lessons learned from their own mentors. This approach is very good for the arts and humanities, as their heydays are typically in the past. This is also true for the older maths. Geometry still looks back to Archimedes for its most compelling storylines. In areas like physics, we still begin with Newtonian laws, as they are suitable for simple applications. And even Einstein’s relativity is looking relatively old. 🙂
Expert-level teachers will face a great challenge in operating solo in the younger disciplines, where their expertise is both a guide and a hindrance. This is where a cooperative teaching effort is best — pairing the enthusiasm, insatiable curiosity, and supple thinking of the young with the wisdom and pattern-recognition of the expert. And as with all cooperative efforts, there will be a standard forming, storming, norming, and performing sequence — thus, they can’t expect to do much profitable work together in the first 2-3 years. However, in their 4th year, they’ll likely be prepared to generate a remarkable contribution to their field.