10 | How Education Works

Revisited: People Must Be Made to Learn

Why do we continue to use extrinsic motivation as the primary driver in education, even though we know with some certainty that it is harmful to intrinsic motivation? The answer lies in the nature of the technologies upon which our education systems are based and the rich layers of path dependencies that they embody.

As we have seen, the parts of all technologies, including learning technologies, must be assembled in ways that work, governed by hierarchical constraints in which the bounded are always strongly affected by the limits of their boundaries. Our education systems were developed within a set of clear boundaries, of time, space, purpose, theory, and so on. They started with a limited range of resources and constraints: there were more students than teachers, more readers than books, and so on. These and other phenomena, including many path dependencies, beliefs about human nature, available funding sources, and so on, determined the adjacent possible. This is how the lecture was born: it might not be ideal, but it was the best, most economical solution to scarcity of resources required for learning at a time when no other alternatives were suitable. There had always been alternatives—much softer apprenticeship models, in particular, were and remain powerful educational forms that can be incredibly effective, but being softer they were not as scalable, consistent, reliable, or cheap. The uses to which resources were put also played a large role in determining education systems’ forms, especially regarding their later roles in society as filters for employers, babysitters for children, and producers of knowledge of value outside their walls.

To a large extent, much of the purpose of at least the seminal higher education systems of Paris, Oxford, and Bologna, was (at first) to transfer particular doctrines, notably those of specific religions but also philosophical, mathematical, and practical knowledge. They therefore sought hardness in their educational machines, with a clearly defined body of knowledge and clearly defined measures of competence. Given the available phenomena and the intention to replicate doctrine, the orchestration developed—primarily in the form of lectures, seminars, and tutorials—was hard, and it made best use of what could be assembled. It needed scheduled times (timetables) so that students could gather together, lectures so that rare books and wise words could reach the largest number, and classrooms to limit distraction and make the words more audible to more people. This led in turn to a need for a successive procession of counter-technologies (Dubos, 1969) that slowly began to define the process. For example, such classrooms needed rules of behaviour, expectations of conduct, and an innate power structure that favoured the teacher as controller of what went on in the classroom, entailed by the need for lecturers to be heard (lecterns, pulpits, and other paraphernalia of academia reinforced this pattern).

The process needed terms and semesters to cater to intermittent student availability. It needed libraries to house and allocate access to the rare books, and processes for loaning them, so that resources could be fairly allocated. As time went by, universities needed processes to manage both their increasing size and the ever-increasing diversity of lecturers and, in the absence of modern ICTs, hierarchical forms of management were a good solution. Because the intention was to transfer hard doctrines, they needed means to distinguish those who had successfully learned them from those who had not, at first in relatively soft ways (e.g., an oral defence to professors and peers) but, as time wore on, in increasingly hard ways (from written exams starting in the late 18th century to “objective” tests in the late 19th century and early 20th century). Increasingly, they needed specialists, who clustered into disciplines and subject areas and thence into departments, schools, and faculties, which fitted neatly into a hierarchical system of management. The need for comparison with competitors and a “product” to sell led to increasingly standardized curriculums. The list goes on. What this adds up to is a relentless though not inevitable path toward the traditional education system that we still recognize today. However, it is worth remembering that it emerged in the first place as a solution to the one central problem of how to allocate scarce resources (teachers, books, time, etc.) in physical spaces. All the rest built upon that foundation, solving problems that it created. Having solved those problems, most of the rest were counter-technologies designed to limit the side effects and take advantage of the adjacent possibles that the initial technologies opened up.

These solutions to the problem of indoctrination were sensible, in the context of the technologies available in medieval times, and, thanks to the inevitable increase in complexity and refinements introduced over centuries, remain largely compatible with one another and our broader and more open teaching needs to this day, at least in the context of in-person learning. The system has evolved, much like bows and arrows, to be efficient and effective, layering counter-technologies and (not always conscious) exploitation of fortunate happenstances until a system has emerged that educates reasonably well. Many of those serendipitous adjacent possibles have become exaptations that form critical and often overlooked parts of the distributed collective that teaches. Libraries, for instance, provide not only books but also ways of organizing and therefore understanding them, opportunities to bump into different ideas, to learn from librarians, and so on. Corridors do not just connect classrooms but also offer chances to connect with others, to prepare to learn, to make the simple act of getting to a classroom more effortful and therefore more salient. Smokers’ areas and dormitory kitchens are incredibly rich environments that connect diverse learners, who can talk about what they are learning and cross-fertilize one another with new ideas. In-person universities, not just the teachers they employ, play a significant teaching role.

The consequences of the designs of education systems are many, but one stands out as more intractable than all the rest: education systems of this kind are systematically antagonistic toward intrinsic motivation (Dron, 2016). As many have shown (e.g., Deci & Ryan, 2008; Reeve et al., 2004; Ryan & Deci, 2000a), intrinsic motivation—through which we do things simply because we love doing them—demands support for learners to experience autonomy (to be in control), gain competence (achieve mastery), and feel relatedness (feel that what they do has value in a social context). Taking support away from any of these fundamental aspects of the process results in diminished or, more often, non-existent intrinsic motivation. The education systems that developed a thousand or more years ago were innately supportive of relatedness, both within and beyond the classroom. However, they inevitably took away autonomy (people had to attend at a place and time and learn what they were told to learn in ways determined by others) and support for competence (the need to cater to the whole class meant that, at best, some would be underchallenged and bored, whereas others would be overchallenged and confused).

The primary and persistent consequence is that, since early teachers (at least) had a strong desire to impart a specific set of knowledge and skills fairly independent of the needs of learners, extrinsic motivation became the primary means to achieve the goal of teaching. We know, and our ancestors knew, that rewards and punishments are highly effective as means of making someone comply with externally regulated requirements (Kohn, 1999), so this made sense in the context of indoctrination. As our education systems evolved, rewards and punishments came to be the technologies of choice, whether in the form of grades, praise, censure for failure to attend, or whatever seemed (in the short term) to work in the context of a classroom. The opportunity for accreditation for learning in the form of degrees, certificates, diplomas, and other proxies for competence was a particularly powerful lure that came to play an increasingly significant role as the centuries rolled by, as education systems began to occupy a more important place in secular society, and as systems of employment began to intertwine with those of education, forming a broader educational ecosystem with needs different from those of learning alone. It was probably also influenced by the regular adoption of such techniques in schools, which faced similar constraints and sought to pass on knowledge and skills at times and places that might not have suited children.

Unfortunately, as decades of research have shown, externally regulated extrinsic motivation can enforce compliance, but it does not add to already high intrinsic motivation. In fact, always and unfailingly, it substantially reduces and, more often than not, totally extinguishes it (Ariely, 2009; Deci, 1972; Deci & Ryan, 2008; Gneezy & Rustichini, 2000; Hidi, 2000; Kohn, 1999; Lin et al., 2003; Ryan & Deci, 2000b). Extrinsic motivation crowds out intrinsic motivation, replacing pleasure or satisfaction in performing the task with the goal of gaining a reward or avoiding a punishment or both. In brief, extrinsic motivation—especially when driven by someone other than the affected individual—is highly antagonistic to intrinsic motivation, mainly because it diverts attention from the task at hand to something external to it, sending a powerful message that the task itself is unenjoyable or insufficient to sustain motivation.

Perhaps the most central design challenge of formal education systems is therefore to overcome or compensate for the loss of intrinsic motivation that its design constraints tend to impose. Although intrinsic motivation is always more desirable than extrinsic motivation, assuming that we are concerned about learners more than doctrines, Ryan and Deci (2017) show that forms of extrinsic motivation which are internally regulated are mostly less harmful, and that higher forms may be positively beneficial. Although doing things because we fear consequences is little better than external regulation, doing things because they help us to achieve goals that matter to us or, better still, because they are part and parcel of our sense of self and identity, may come close to intrinsic motivation in value (Vansteenkiste et al., 2004). Teachers might find it difficult to avoid imposing extrinsic targets altogether, but it is often possible to help learners find those higher forms of extrinsic motivation within themselves through good pedagogical methods and caring support.

Many of our most cherished and deeply embedded pedagogies are in fact designed to motivate, enthuse, challenge, and captivate in order to trigger these higher, self-imposed forms of extrinsic motivation, from scene-setting, attention-drawing introductions to lectures to “hey, wow!” science demonstrations. Good teachers tend to leverage the social aspects of their teaching to emphasize relatedness, with group work, open-ended questions, memorization of students’ names and interests, and so on. Many learning designs are based upon ways to increase time on task, for instance through gamified designs or activities intended to enthrall. Others are designed to hold attention or support teacher authority in a classroom. Yet others are intended to sustain enthusiasm or offer meaningful challenges. In fact, the closer one looks at what actually happens in classrooms, the more it becomes apparent that (at least in conventional classrooms) most of the pedagogical methods applied are simply counter-technologies invented to keep people engaged who otherwise would be disengaged. From roll calls to classroom layouts to assessment regimes to chunking of lectures or spacing of learning activities, our education systems are machines made to compensate for the lost intrinsic motivation that is a fundamental consequence of their design.

This is a highly evolved machine, the result of many inventions and inspirations stretching back into prehistory, so despite such problems the system works reasonably well most of the time. It is possible—indeed, I would argue, necessary—for educators to diminish the risks of extrinsic motivation by supporting needs for competence, autonomy, and relatedness. However, as long as the design of education systems systematically diminishes both autonomy and the need for the development of competence for at least some learners, and as long as extrinsic drivers—notably in the form of accreditation—remain the sine qua non of education systems, it will always be an uphill struggle. It is perhaps the central problem that in-person pedagogies must solve. Much of educational design and good practice is necessary only because the conditions under which education is “delivered” (a terrible term that reveals much about attitudes toward it) militate against effective learning.

Online learning, without the application of much ingenuity and effort, is the intrinsic motivational inverse of in-person learning. With provisos and exceptions, the relatedness aspect of online learning tends to be weaker in most online teaching, but inherent support for autonomy and competence tends to be greater (illustrated in Figure 3).

Online teachers can never exert the second-by-second control of their in-person counterparts, even in synchronous sessions, because a learner is always inhabiting at least one other significant environment in which teachers have no control: their own environment. What is described as a “virtual learning environment” is actually nothing of the kind: it is just a technology operating in the context of the learner’s environment. The ability to take detours, or alternative paths, and to make use of alternative resources beyond the constraints of working together in a classroom offers more choice to online and distance learners than to their in-person counterparts. In fact, one of the fundamental (potential and often actual) benefits of online learning is support for autonomy because, no matter how hard teachers try to control it, control rests more firmly in the hands of the learner at all stages of the learning journey. The internet is filled with teachers—labelled as such or not—whom learners can choose to guide as much or as little of their learning trajectories as they wish.

Building upon the work of Morten Paulsen (1993), Terry Anderson and I (Dron & Anderson, 2014a) identified at least 10 types of freedom in online learning that differ from those found in conventional in-person learning, including the common distinctions of time, place, and pace as well as degree of social interaction, choice of media, ability to delegate or assume control, choice of tools, content, and pedagogical method, and so on. I would now subsume some of them under a single category of “choice of technology,” but sometimes there is value in subcategorizing different kinds of technology. Our decision to limit the range to only 10 factors was largely pragmatic—to maintain manageable complexity—and based upon informal observation rather than patterns and themes that we had discovered through rigorous empirical research. There might be many other freedoms that emerge from separating learners and teachers in time and space.

Although tempered by the typically greater difficulties in direct communication, competence is generally better supported in online learning than in in-person learning because usually it is much easier for students to take things at their own pace, to explore alternative resources, and to adapt the pedagogies that they use to the need at hand, without having to follow the dictates of the teacher or the rest of the class in lockstep. That said, it can be difficult—at least in typical formal settings—to get timely help when things become complicated, so the benefits are not necessarily so great in practice. Unless there are others in their environments who can help, it can be harder for online students to know what to do next or to get immediate feedback on misconceptions or help with procedural difficulties. The inability to know what to do next to solve a learning problem can be very disempowering. However, at scale and outside a conventional teaching environment, the opposite can be true.

A question posed to a popular forum such as Reddit, Quora, or StackOverflow often can be answered within minutes, any hour of the day or night (though the quality of the answers can be difficult to judge). Similar benefits are often found in large-scale MOOCs. For instance, in one of the first MOOCs to exceed 100,000 students, the median response time to questions posed on a discussion forum was 22 minutes (Severance, 2012, p. 9). This advantage also helps to get more of a sense of relatedness, albeit seldom to the extent of the close-knit groups that are the norm for in-person learners.

What is particularly odd (though perfectly understandable from a technological perspective, given the need for interoperability with the rest of the education system, and the nature of technology evolution that builds upon existing pieces) is that distance and online learning systems, though not as constrained by the demands of physics as their in-person counterparts, have substantially replicated the pedagogical methods, notably including those that relate to motivation, designed for the different technological context and problems of the in-person classroom.

Many standard teaching patterns in online and distance learning—especially those that follow an objectivist pattern—are even harder than those usually found within in-person classrooms. Course developers work from an assumption of teacher control, of objectives and lessons set by the teacher, often even more rigidly proscribed than their in-person counterparts. This is especially true of self-paced courses that are designed in their entirety in advance and delivered as a single package because it would be highly disruptive to make more than minor changes to a course while students who have already explored its materials are taking it. Thus, materials tend to be designed with greater care, in more detail, and with more prescriptive tasks and activities than in most in-person or even paced online learning contexts.

Above all, the majority of online teachers use assessments and grades to drive motivation, whether purposefully or not, thereby attempting to claw back much of the teacher control that, inherently, is lost in online learning. In the absence of most of the in-person power structures that allow classroom teachers to control proceedings, grades often play a far more central role in online and distance learning, and there tend to be more of them. Paradoxically, this is even (and perhaps especially) true for self-paced learners who work in different ways from conventional in-person learners and who appear, by default, to have far greater freedom of choice. Self-paced courses, which usually can be taken at any time, without the need for interactions with other students, at a pace to suit the learner, and without a fixed schedule, do not need to be designed like in-person courses. Despite this, they still impose teacher control through grading as well as limits on time and work needed as defined by assumed study time in and content of conventional courses. There is thus a mismatch between the propensities of online learning and the in-person pedagogies that tend to be superimposed on them.

Unsurprisingly, given the consequent effects on motivation, many (though far from all) online courses experience lower completion rates than their in-person counterparts, despite (and perhaps partly because of) their attempts to keep students on track with extrinsic motivation. When grades and credentials are removed from the picture but everything else remains much the same, the results speak for themselves. Most credential-free MOOCs, for instance, are lucky to achieve completion rates of more than about 6%–7% (Jordan, 2016) If some kind of certification is offered then completion rates rise to an average of 15% or more. This still-low figure reflects only partial uptake of the usually paid opportunity to gain transferable credentials. Although not a problem for independent students who attend only part of a course for personal learning reasons, or dabbling, or simply visiting out of interest, this is a damning indictment of the methods and designs that they share in common with credential-bearing courses, when viewed from the teacher’s perspective. By way of comparison, my fully online and credential-awarding university (Athabasca University) reckons to achieve average completion rates of about 85%1 and, in many courses, much higher than that, though it is important to note that, following the almost universal practice by in-person universities of counting only students who turn up for a course in their completion statistics, this excludes 30% or more of registered students who never get as far as submitting their first assignments (our equivalent of identifying attendance in self-paced courses). This, too, is on par with traditional universities.

The differences are stark. Although some of that relatively high persistence rate can be attributable to loss aversion (students who have paid a significant sum of money, been sponsored by their organizations or governments with conditions and expectations attached, or require a course to complete a program that matters to them tend not to drop out as often as those for whom there is no significant commitment) and/or maintaining face with family members, friends, and colleagues who know that they are taking a course, the difference is vast enough to suggest that the traditional course format, without its defining motif of accreditation, is far from sufficiently motivating to learners to be useful or effective. Pedagogically, most MOOCs are at least as well designed as their for-purchase counterparts, and often more so, because they are usually produced by passionate teachers who wish to share their knowledge and expertise with wider audiences rather than as part of their daily duties. Designers and teachers of MOOCs are thus far more likely to be intrinsically motivated and therefore to put greater effort and care into course design and delivery. There is immense structural significance to grades in the overall assembly of online courses, but the consequences for motivation perhaps are even more dire than they are to in-person learners.

There are many ways to reduce such problems, even within a fairly conventional course structure. Getting rid of grades and replacing them with opportunities for discussion and feedback are a good start (Blum & Kohn, 2020; Sharp, 1997), at least until the end. Designing courses that support personal interests, allow students to choose topics, provide negotiable learning outcomes, and so on can be useful. Because much of the extrinsic motivation that drives education results from the need to acquire credentials, one simple solution—relevant to both in-person and online education—would make a large difference: to decouple learning from accreditation. There is also enormous promise in the use of all those myriad teachers found both locally and especially across the internet to support autonomous learning. I will discuss these issues further in the next section.

Revisited: Online Learning Dominates in-Person Learning (except in Formal Education)

At least among those with reliable and easily accessed internet connections, the ubiquity and effectiveness of their learning online show conclusively that, freed from its formal and extrinsically motivated fetters, online learning not only works but also dominates the learning landscape in the 21^st century. This is easily explained in co-participation terms. There are literally billions of potential teachers just a search term or two away, many of whom attempt to communicate knowledge, be it true or false, or skills, whether effectively or not. This is distributed teaching at its current apogee. Almost any learning orchestration is possible, and, with sufficient care and effective search strategies, almost any can be found. More is different (Anderson, 1972). With scale comes complexity, new interactions, new emergent adjacent possibles, and greater diversity. And the learning occurs precisely when it is needed, without coercion, so learner motivation is high.

Much of the learning accomplished via the internet is orchestrated (at a broader scale) by the learners themselves, pulling together harder (and sometimes softer) pieces such as written tutorials, YouTube videos, Wikipedia articles, MOOCs, email correspondence, and so on to provide unique learning solutions tailored to their needs. It is often possible to gain direct tuition from originators of ideas, research, and theories. This is possible because the internet is an incredibly soft technology, especially thanks to the ease with which its technologies can be assembled with others. For the most part, it uses open standards—the underlying technologies that make it work such as TCP/IP, HTTP, JavaScript, and XML—that allow a diverse range of uses, with a wide range of devices, over highly disparate network substrates, from wired ethernet networks to satellites to undersea cables to cellular networks. Even for the increasing number of commercial and even non-commercial systems that overlay proprietary, locked-in toolsets on top of this open infrastructure (following a design pattern of replacement in order to harden what is essentially soft), interfaces, APIs, gateways, and so on make it relatively simple to connect almost any service, device, or tool to a vast number of others. Increasingly, standards such as xAPI, LTI, Caliper, OpenBadges, and SCORM make it possible to integrate this cornucopia with more formal learning and records of it.

This is an archetypal example, however, of the problem that soft is hard. The internet consists of at least millions of hard technologies enacted in billions of ways every day. The more choices available, the more difficult it is to make them without at least some hardening. Hardening, of course, is exactly what the algorithms and interfaces powering search engines such as Google or those internal to specific sites or toolsets provide. The internet, infamously, is a massive swamp of stuff with a few isolated islands of well-organized, reliable information (Crawford, 1999). It is rich in deliberate falsehoods, distractions, trolls, and echo chambers, albeit perhaps not as disturbingly as is often portrayed in the popular press (Dubois & Blank, 2018). Technologies intended to help harden it to make it easier to find reliable sources for the knowledge that we seek, even when they work well and as advertised, often result in filter bubbles (Pariser, 2011). Unfortunately, without them, the extreme softness can be daunting or overwhelming, even for expert users. Too much choice is as bad as too little choice, especially for learners near the start of their learning journeys.

We need to be able to delegate some choices to others whom we perceive, rightly or wrongly, as sufficiently expert to help us. Unfortunately, the algorithms that filter or provide us with recommendations are seldom designed to support better learning. Although I and some others have designed systems for that purpose (none of which made it far beyond research systems), most are more focused on driving traffic to sites relevant to the subject matter. To make matters much worse, search engine optimization (SEO) strategies are often deliberately designed to subvert this, whether for commercial or propaganda purposes. The chances of finding resources useful to learning, consequently, are diminished.

Some have suggested that at least part of the solution to this problem is better education, especially in digital, network, and social media “literacies” (using the term, as previously discussed, in a fuzzy way). Proponents point to things as diverse as simple operational skills to those of design, legal, ethical, and social behaviours. However, though they definitely can do some good, the problem is only partially susceptible to direct teaching, because the internet, by definition, is a network of networks and, above all, a network of networked communities and cultures, each of which demands at least in part its own distinctive (and sometimes mutually exclusive) literacies.

Many of these networked cultures, not least because of the massive interconnections and scale of interactions supported, are evolving rapidly, so literacies acquired now might soon be out of date. Often, thanks to the structural softness of the most common technologies used (from sites to browsers to protocols), it might not even be clear that we are interacting with a specific culture. For instance, the uniformity of search results or news feeds deliberately masks significant differences in content, design, and other contextual indicators. In such a context, a search for “evolution,” say, might well lead to nonsense on intelligent design or radical religious sites, rather than to information about plausible theories, sometimes with little to allow a novice learner to distinguish between reliable and unreliable sources.

It is even more problematic for topics such as the climate crisis, in which deliberately misleading information can be provided by powerful interest groups in forms designed intentionally to deceive or cast doubt. Attempting to solve the problem with greater hardness may introduce new problems of its own, because one size seldom fits all. For instance, efforts by web browser manufacturers to protect their users from scams, privacy intrusions, and malevolent sites often render perfectly legitimate and harmless sites inoperable. Similarly, by taking away agency from those who do understand the problems, the heavy-handed hard filtering sometimes prevents or seriously constrains those who might find alternative, softer solutions from doing anything about it.

The Googlization problem also reveals another set of technological incompatibilities between the pedagogies (and especially the assessments) of traditional in-person learning and the capabilities of this vast, distributed teacher. From teachers who ban cellphones in their classrooms or the use of Wikipedia in assignments and homework to the lament that students just copy their work from internet sources or, worse, use ChatGPT or employ others at the low rates that a vast international network affords to do the work for them, the methods of teaching and assessment typically used in institutional learning have failed to keep pace with the changing ways that we go about learning in an internet-connected age. To a large extent, both problems are direct results of the motivational issues discussed in the previous section.

When we force people to learn things that they might not wish to learn, in ways that might not suit their needs or skills, in a disempowering context, and then we apply extrinsic motivation to make the focus of learning the achievement of grades and credentials, it is little wonder that our students take the quickest, most direct approach to achieving them, and it is hardly surprising that there are plentiful services available to meet this demand. It is unusual for intrinsically motivated learners without extrinsic drivers to take shortcuts or to cheat, because the only people whom they would be cheating would be themselves. Preschool children who learn through play often deeply resent being interrupted in their learning because it is fun. By “fun,” I do not mean that every moment makes them laugh for joy. Indeed, watching children at play, it is normal to see much seriousness, and even anger and frustration, as they struggle to overcome challenges of some game or toy, as they practise and struggle until they achieve success. What Csikszentmihalyi et al. (2014) describe as “flow” (a theory highly congruent with self-determination theory) is seldom a state of elation, though it can be highly satisfying and meaningful.

The problem cannot be solved effectively by smarter policing or greater control of the process. In fact, it is likely to make things far worse by further reducing agency in, amplifying the power imbalances, and disrupting the caring, supportive relationships that we strive to nurture between teachers and learners, as well as further emphasizing the importance of the grade, rather than the learning activity itself. Using automated technologies to uncover plagiarism, or tightening restrictions on the use of technologies, is simply to use counter-technologies to harden further a technology already incompatible with the reality of the world around it, making it more difficult to bring about real change. While it remains soft there is at least some potential for flexibility and accommodation of other ways of approaching the learning problems. The more that we harden it, the less compatible with the pedagogies that the distributed internet provides it becomes. The struggle to control the teaching and assessment process has always been an arms race between teachers seeking control and students driven primarily by those teachers’ demands to judge and accredit their learning.

When new technologies of cheating (or, more charitably, shortcuts to getting the required grades) evolve and spread ever faster, it can be only a losing battle for educators to continue to teach and assess using the same methods that predate widespread access to the internet or generative AIs. The technologies of cheating are much softer, more agile, and more numerous than the stolid institutional processes that they circumvent. Simply attempting to harden a broken system through counter-technologies can never succeed for more than a moment. Every time that one part is hardened, a mass of soft processes is assembled around it that renders it useless.

The solution to such problems is both simple and difficult because it requires teachers to develop (or accommodate) softer, more flexible, adaptable, and personal (not personalized in the sense of something done to students) ways of teaching and softer, less tightly coupled ways of demonstrating that learning has occurred. It demands changes in how we support learning and in how we judge whether it has occurred.

One of the most harmful consequences of the methods that emerged from formal education’s medieval origins is the conflation of formal, judgmental assessment and learning: they must be decoupled, or the entire process must be altered radically, if there is to be any hope of supporting the intrinsic motivation natural to all learners. Although feedback is hugely valuable, summative judgment—especially when linked with credentials—and teaching, for the most part, are largely incompatible technologies. It would make some difference, at least in some cases, were teaching and credentialing to be entirely separate, unaligned activities, performed by different people, with clearly separate purposes. Even when credentials are administered by people other than those teaching, however, it is hard to prevent the almost ubiquitous practice of teaching to the test, as the abysmal effects of SATs in schools clearly show (Baker, 2020; Locker & Cropley, 2004). Notwithstanding the difficulties of implementing such a policy, the general principle of divorcing learning and credentialing as much as possible should be applied whenever, to whatever extent, and however it can be applied. If that is not possible, then we must either use whatever means are at our disposal to reduce the dependencies between learning and credentials or rethink how credentials are given.

One way to get around the fundamental incompatibility of credentialing and effective learning is to make the credential an award (a means of recognizing what we have achieved) rather than a reward (a means of assessing compliance with a specified set of demands). Rather than or at least in addition to imposing “objective” and extrinsically imposed measures of learning, we need to support explicitly flexible, expansive, and open outcomes. We need to recognize the things that have been learned rather than punish learners for the things that they have not learned (yet). Such unplanned outcomes are natural consequences of how we learn, as the example of the lecture in an earthquake reveals. We might make use of portfolios of learning, for example, that are created as part of the learning process and that naturally record such learning and then, as an entirely separate process, provide an award for what has been learned as opposed to achievement of what we intended to teach. In many cases, it might be possible to discover outcomes that match more than one award or to combine the outcomes of different learning activities to match a single award.

To achieve this means a shift in focus from assessment as a reward or punishment to assessment as an award for whatever successes we accomplish, in the spirit of appreciative inquiry (Cooperrider & Whitney, 2011), or outcome harvesting (Wilson-Grau & Britt, 2012), celebrating achievement rather than punishing failure. Appreciative inquiry deliberately seeks the things that work, recognizing that measurement changes what is measured in potentially harmful ways, so (if improvement is sought) it is far better to focus on what worked than on what did not and to explore ways of improving and extending it rather than diagnosing failure. Similarly, outcome harvesting seeks not only to discover whether intended outcomes were achieved but also which other outcomes occurred along the way.

Since there are virtually always learning outcomes in addition to those intended by a teacher, including changes to ways of thinking as well as discoveries and improvements in skill, this offers a far kinder, more precise way to evaluate learning as well as great insights to teachers about the teaching process, not just their own but also across the teaching gestalt. If, at the same time, we also remove the requirement for every course to be some multiple of a predetermined length (as we have seen, an arbitrary quantity determined mainly by the timing of medieval Christian holidays) and instead allow each course to be the length required for the subject matter, then a single course might provide evidence of more than one credential, and multiple courses might be aggregated to provide evidence of a single credential. Rather than being proof of having met the demands of a teacher for a course, credentials can thereby offer evidence of achievement of outcomes, thus becoming more beneficial to employers as well as teachers and students. It would also allow students to assemble learning from beyond the institution, and perhaps from multiple institutions, giving greater personal control over the process.

There is a crucial place for diagnosing problems in teaching and learning. Feedback on success and failure, especially diagnostic feedback, particularly when given promptly or innate to the task, is an extremely important part of most learning journeys, one of the most consistently reliable pedagogical technologies that we have invented, and (along with its counterpart, feedback to the teacher on the teaching process) perhaps the most vital part of any effective educational assembly. However, it should be part of a process of continuous improvement rather than a summative wall that shuts down further learning. If particular learning must be achieved—a necessity in many roles in life, from driving to brain surgery—then multiple attempts should be allowed until success is achieved or until the learner chooses to give up. The only value of negative feedback is as a signal that more learning is needed, not that learning so far has failed.

Timetables and schedules can also be enemies of effective learning. It is a consequence of the constraints of traditional learning that, if it is not perfectly achieved within the time frame that we set for it, on the whole we simply offer a lower grade (including, sometimes, the bizarre and counter-educative notion of a “fail”). Apart from a few special cases, this makes no sense in traditional teaching, assuming that we believe education to be primarily concerned with learning, let alone in the open, flexible, pace-free environment of the internet. When learning without the constraints of timetables and schedules, we should be able to try until we succeed. There is no reason that able students who continue to work at a skill or the acquisition of some knowledge should ever achieve less than total success (however it might be measured) as long as they persist, they are well supported, and there are no learning disabilities or other constraints (e.g., economic or inherent timing sensitivity) to prevent it. Anything less is a failure of teaching, in all its distributed forms, and even when it succeeds teaching can (as the co-participation model implies) always be improved.

It makes a great deal of sense to extend the boundaries of a formal course to encompass the internet, which can be achieved easily through tools of aggregation to syndicate resources from outside, curated lists of resources, student-created social bookmarks, sharing through Twitter hashtags, and a host of other mechanisms. Extending the course boundaries to allow members of the public (or at least a broader university community) to see what students are doing can also be extremely effective, as long as students can choose not to share at least some of their work: control is essential. Such methods are firmly in the complexivist tradition of pedagogies, and they are thus native to an internet-connected world to a far greater extent than in the objectivist and subjectivist methods common in online teaching. They thus tend to be less path dependent.2

Revisited: There’s No Significant Difference in Learning Outcomes No Matter Which Media or Tools You Choose

The fact that medium or mode of delivery appears to make no difference, on average, is unsurprising given that most such studies provide few controls for method, which might matter much more than medium (Clark, 1983). However, even if methods are as consistent as they can be across media (not really possible thanks to countless dependencies and interactions between technologies used in the assembly), any soft technology can be used well or badly, so it is not remarkable that some online learning is better than some in-person learning, and vice versa, or that the differences tend to balance out. To suggest that all e-learning is comparable to all p-learning (place-based learning) makes no sense given that there are virtually infinite ways to do both and that described methods are only a fraction of what actually goes into any learning experience. Any individual part of the assembly might be better or worse implemented, and the assembly itself might be well or less well orchestrated. This is not to mention the enormous contributions of the pedagogies and working practices of the learners themselves.

It makes no more sense to suggest—without extraordinary evidence—that any one technology, pedagogy, tool, or method is better than another than to suggest that oil paintings are better than watercolours or that blues is better than chamber music. Indeed, typically, there are more differences between two courses using the same modality than between those using two different modalities. There is a world of difference, say, between an online, self-paced, objectivist course and an online, paced, social, subjectivist course or between a small, intimate, in-person tutorial and a large, in-person, impersonal lecture class. There are also many issues related to stakeholders’ perceptions of online learning, especially in in-person institutions, where attitudes of faculty, students, and leadership, on average, are negative, especially thanks (mainly in the United States) to their association with for-profit degree mills of dubious quality (Protopsaltis & Baum, 2019).

On a more cautious note, the measurements of effectiveness that we tend to use can conceal a number of important differences. For example, in a well-conducted study, Heller et al. (2019) compare two MOOCs delivered in a paced format (with tutor support) and a self-paced format (with no tutor support) and find no significant difference between the two when measured in terms of course outcomes. Although both versions included discussion forums, discussion on those with tutor mediation, unsurprisingly, was significantly higher, but, perhaps more surprisingly, completion rates for each group differed little. The authors tentatively (and with reservations) suggest that tutors therefore might have done little to improve learning outcome or retention and that other factors such as initial motivation played much larger roles. It is unclear, though, from the study exactly how tutors were expected to contribute and what kind of role they played, nor is there much indication whether there were other benefits such as reduced time on task because tutors clarified questions.

In a longitudinal study of repeat iterations of a similarly constructed MOOC performed at Athabasca University, early attempts to provide moderately intensive tutor support offered little obvious value, much as Heller et al. (2019) found. However, in repeat iterations, when tutors were more carefully coached to encourage students to interact with one another rather than with them, along with a deliberately higher ratio of students to tutors intended to reinforce the message that tutors were there to connect students rather than to answer questions, retention rates improved significantly, though measured outcomes were similar (Mishra et al., 2019). Regardless of the effects on retention, in both cases there were differences in learners’ experiences depending on levels of engagement, but in neither case were attempts made to measure the full extent of what the learners actually learned as a result: both studies simply looked at measures such as persistence and achievement of intended outcomes. Everything that we do and experience changes us, so it is virtually impossible for such different experiences to have led to identical learning outcomes. Perhaps students improved techniques for asking questions, or discovered things not assessed but nonetheless useful, or formed relationships that continued to offer learning benefits beyond the course. Perhaps nothing of the sort happened.

The problem is that, because only visible engagement, persistence, and performance to predetermined objectives were measured, we have no way to tell. As Protopsaltis and Baum (2019) argue, there are many aspects of an in-person learning experience that matter, few of which are ever observed let alone measured. This is even more true of online learners, whose presence is measured only by the deliberate traces that they leave in digital spaces but whose learning almost certainly extends into other parts of their lives and interactions with others. If we are to understand education at all, we need better approaches to discovering more about what is learned—and how—than the simple things that we normally choose to record. Again, it would be useful to harvest outcomes (Wilson-Grau & Britt, 2012) rather than measure only those that we expected to occur. It would also help if we were to design learning activities that deliberately made those hidden parts of the process visible, for example by the incorporation of reflective learning diaries that discuss the process in the outputs that we expect of our students.

A similar set of issues underpins Hattie’s (2013) findings that just about every method works. In fact, Hattie finds that the average effect size of the improvement reported across all interventions is 0.4 (p. 32). Assuming this as a baseline, he tries to identify an ordered list of effective methods and strategies. Such an ordering is at least partly possible because he restricts his study to a fairly hard and limited set of technologies. The measure of success that he uses is simply the grade achieved in the context of a conventional institutional classroom. Furthermore, Hattie is concerned only with what “works” in a circumscribed, largely in-person, institutional classroom context. He barely considers the myriad other ways of learning that are or could be used were that context to be changed or the many other outcomes, good and bad, that might occur as well.

The studies that Hattie’s (2013) work compiles almost all relate to one particular kind of technology (the school or college), with a particular set of purposes (though set in an educational context, far from all educational in nature), by an atypical set of practitioners (those enthusiastic and informed enough to research their practice). Thus, what constitutes success is not generalizable to all human learning, only to the specific subset of common but often inauthentic contexts found in typical educational institutions, mainly in English-speaking countries. To apply this to all education, including everything outside a formal school or college context, would be a little like identifying features such as cylinder size, carburetor efficiency, and number of spark plugs that affect performance in cars, then applying them to electric cars or bicycles.

Hattie (2013) himself is the first to recognize the limitations of his methodology and that no prescriptive list of this kind can ever cater to the specific needs of a given set of learners, so the main message of his book emerges from going beyond simple methods and instead looking for the general patterns that most successful interventions share. Based upon his findings, he observes that passion and artistry in a teacher are usually more important than method. Hattie argues that effective teaching occurs when teachers continuously learn, about what their learners are learning and about how they are teaching, in a responsive process that constantly changes as learners (including teachers) themselves change. He describes this as “visible learning” (discovering what and how students learn) and “visible teaching” (discovering what and how teachers teach). His most significant contribution is less telling us which methods (specifically) we should use to teach than identifying how (in general) to be a teacher. Although stemming from an analysis of empirical data rather than deriving from first principles, these methods are in exact accordance with a co-participation perspective. Having read Hattie’s work some years after first developing the ideas in this book, I find it gratifying to know that they stand up to empirical validation.

Revisited: The Best Ways to Teach Are Not the Best Ways to Teach

It is not at all surprising that the evidence to support broadly subjectivist approaches to teaching is patchy and inconsistent or that the average effects on grades are mediocre. Of all pedagogies, those that lie in the subjectivist spectrum are among the softest, so it is inevitable that huge variance will be seen in the results because they will always be instantiated significantly differently—in terms of skill and method—from one context to the next. Indeed, constructivism and other subjectivist models are theories of learning, not of teaching, so inevitably large amounts of detail and process must be filled in if a teaching method based upon the idea is to be put into practice, the details of which can vary enormously. Teachers have to instantiate those pedagogies with creativity and skill. Subjectivist models also implicitly acknowledge that learner pedagogies can and must play a large role in learning, so their skill matters too. Thus, the degree of expertise and the level of engagement with the process must play a dominant role in the effectiveness of the learning technologies.

Far more than in objectivist approaches, the soft technique—the passion, caring, and creativity of teachers (including learners themselves)—leads to success or failure, rarely just the easily described processes, methods, and tools that formally recognized teachers use to bring that about. Unfortunately, simple statistics imply that there are likely to be more average and below-average teachers than those above average, so the chances of this happening are lower, on average, than when using well-proven harder pedagogical methods. The matter is made worse since a casual reading of the literature suggests that minimal guidance is needed for such pedagogies, which is neither normally true nor claimed by most theorists. It is just that the learners themselves are expected to play a more significant role in determining goals, approaches, and methods, with the support of those expected to know more as well as the support of one another. If anything, this demands a more active role from all the teachers involved in the process. This need for skillful technique is not just a problem when considering complex interventions with many possible permutations such as active learning. Arguably, one of the simplest and most uncontroversially valuable pedagogical interventions of all, feedback, on average is a Really Good Thing, but reported results are among the most variable in their influences in educational interventions (Hattie & Gan, 2011, p. 249). There are many ways of giving feedback that can be great, harmful, or anything in between. Learners can be discouraged by poor feedback as easily as they can be inspired by great feedback. Grades as feedback (perhaps except for some hard, right-answer topics), for example, are almost universally harmful (Kohn, 2011).

Active learning methods demand not only hard skill in applying the method but also soft technique, compassion, and artistry by all teachers involved. Because most reports of successful uses of such methods are made by people with exactly those attributes, it is not surprising that they are reported as being successful or that they fail when enacted by people who are underprepared, have insufficient practice, or are simply not attuned to the methods. In such cases, it might indeed be better to employ a harder pedagogical method (though still sufficiently soft to allow two-way feedback), such as direct instruction or mastery learning, that offers greater assurance of success to teachers who lack the necessary time, energy, or skill to implement softer pedagogies. This speaks to the need for teachers to learn more about how to teach and to engage more actively in reflective practice, to which we turn next.

Knowing more about educational theory and practice can help one to become a better teacher up to a point. Tellingly, pedagogical knowledge gained through education can have some effect on teaching proficiency, but even when it does that effect is not large. Goldhaber (2002) and Goldhaber and Brewer (1999) found that the effect of a teacher on performance is only slightly greater for teachers with full certification than for those with provisional certification following minimal training. Similarly, Hattie (2013, p. 151) observes negligible differences between the effectiveness of teachers with four-year certification and those with alternative forms of training, or even brief emergency certification, though (on average) there are slightly larger improvements among teachers with a few years of experience, and teachers without any training whatsoever tend to fare badly. This seems to imply that some pedagogical/teaching-process knowledge is useful, but there are few gains to be had from being taught a lot about pedagogy. Like a paintbrush, it takes some work to learn the basic strokes, but the real improvement comes about through practice, reflection, and iteration, especially when there are opportunities for feedback and discussion.

For most, there are diminishing returns beyond a certain level of expertise, and it is not uncommon for great teachers to reach a peak relatively early in their careers. This is not unusual in creative occupations as a whole. It would be a brave critic who suggested that Shakespeare’s plays from the middle of his career are any worse than those later in his career, for instance, despite presumably being the results of greater knowledge and experience, and for every example of authors, artists, musicians, mathematicians, or scientists producing their greatest work at a venerable age there will be at least as many (I have a hunch more) who did their best work earlier in their careers. Expertise can be developed over time; however, notwithstanding the widespread belief popularized by Gladwell (2008) that expertise follows from 10,000 or more hours of practice, careful studies have revealed that practice in fact might account for as little as a quarter of the skill of an individual (Macnamara & Maitra, 2019), and far from all who excel in their fields put in that many hours, whereas many who do not excel put in more.

The depressing truth is that a lifetime of reflective practice typically leads to a fair level of proficiency and competence, but it will not turn someone with no talent into a genius. Indeed, given that boredom can play a notable role (many teachers, especially in schools, move on to other careers), too much experience within a relatively turgid education system with little scope for growth might be counterproductive, unless teachers constantly and successfully take on new and interesting challenges (Boyd et al., 2011). It is not the pedagogies that need to be better developed but the teachers themselves.

It is important to remember that soft technologies are most suited to orchestrating parts that are fundamentally human and not technological at all—love, passion, interest, excitement, and so on often are of considerably greater importance than the mechanical methods, tools, gadgets, and structures that we use to express them. If you are a teacher and your teaching, of yourself or others, does not inspire your passion, then it might be time to learn new methods, topics, ways of learning, and tools or to get out of the business altogether. In part, this is because passion and compassion are among the key phenomena that can and should be orchestrated in successful teaching, whether of ourselves or others. To a large extent, though, they are what drive us to learn and teach, the energy that makes the educational machine run, without which we might as well be soulless AIs. Passion can be simulated up to a point, but it is usually a bad idea because it is hard to maintain the illusion for more than a little while, and it is difficult to fool everyone all the time. If you find little or no excitement in the subject, in helping learners to learn, and in figuring out how to do that better, then something needs to change. As we have seen, some learners will succeed despite your lack of interest because you are not the only orchestrator of phenomena, but so much more would be possible if you could rekindle your passion.

One approach to doing so is to seek novelty. There is much to be said for learning new technologies of teaching, not because they will improve practice automatically (few if any will do that) but because they encourage reflection, invoke surprise, and maybe even inspire delight in their novelty.

There are also technological means to nurture and engender passion, techniques that can be developed and improved with practice, such as method acting (Stanislavski, 1989). Method acting is used by actors as a means to become entirely immersed in their roles so that they are not so much acting as behaving and feeling like the characters whom they play. It employs a range of soft but formal processes for developing techniques in identifying and replicating sensations, achieving focus (hyper-attention), and removing tension (using relaxation methods common in therapy), and it uses deliberate methods for drawing on relevant memories of past feelings. It can take a long time to learn these techniques well, but some of the basic tricks can be used by anyone: relaxing, focusing, and remembering feelings might be enough to start the process rolling. It is difficult for most teachers to become completely immersed in teaching, inasmuch as few of us love every aspect of our subjects as much as every other, and there is often a need to cater to diversity in both subject and audience, as we flit from one class to the next. However, finding what we do love about what we teach, hopefully, is not a major problem for any teacher.

If deep role immersion is too difficult, imagining ourselves in a slightly different light can make a significant difference to how we think and behave (Brown et al., 2019). Even something as simple as forced smiling can activate parts of the brain associated with pleasure, thus potentially increasing the probability that we will actually feel pleasure (Hennenlotter et al., 2009). Passion is an entirely non-technological phenomenon, but technologies—including music, visual arts, dance, and literature—can stimulate passion and make use of it as part of an orchestration for teaching.

Pedagogical methods can both take advantage of and help to kindle our passions. Telling stories, anecdotes, or snippets of information that reveal our own attitudes and excitement, no matter how small, can make a large difference to how we communicate our passion to others, whether directly or in prepared course materials. One of the innate advantages of in-person teaching is that the effects of such stories can be seen immediately, so feedback loops can drive greater passion among all participants. In an in-person or webinar context, body language can matter. Facial expressions (not faked) certainly do. Responses to questions really matter, and the speed of a response, especially online, matters even more because, especially online, it is a prime indicator of interest in both the subject and the learner (Richardson & Lowenthal, 2017). The infectious nature of passion is yet another reason that social pedagogies are a good idea. In an in-person context, normally, it takes little effort to communicate passion, but online it is important to provide sufficient channels so that learners can be aware of how other learners are experiencing it, not just the teacher.

Discussion forums, backchat channels in webinars, shared blogs, other shared artifacts of learning such as files or bookmarks, and so on can all contribute effectively to an infection model of learning as long as the technologies (including pedagogies) are sufficiently soft to allow free and open expression. Asking a leading question to which answers should be expressed in a largely proscribed format, for example, is rarely an effective way to support infectious passion, still less the sharing of assignment outputs for critical review, though both relatively hard approaches can have a place in the overall assembly. Open questions, questions that encourage learners to connect the current experience to things that they care about, and wicked problems, especially those that incite passionate debate, can be far more effective. Much more could be said on this subject, but the key thing is that development of skill in the soft technologies of teaching is critical to the success of softer approaches to teaching, and without it the results are likely to be disappointing.

The importance of passionate, caring teaching in enacting soft pedagogies suggests that, rather than researching the effectiveness of these particular teaching methods, we should be putting far more energy into researching the effectiveness of particular teachers. If the methods are soft, reliant more on an individual’s distinctive technique than on a specific set of processes, then what matters most is developing those techniques in teachers, not honing the methods. The better we can understand what makes a great teacher, the more likely it is that our teaching will be effective (to a point) no matter which methods we use. The emphasis of research on soft methods of teaching should therefore be on raising the bar so that an average teacher of the future is better than a good teacher of today.

Revisited: No One Has Solved the 2 Sigma Problem

From a co-participation perspective, the most interesting thing about one-to-one tutoring is that it is simply a condition of teaching, not a process, not a method, and certainly not a pedagogy. Because pedagogies are soft technologies, what we describe as a “method” will always leave gaps, to a greater or lesser extent, that need to be filled by technique. One-to-one teaching is nothing but a gap. It allows a teacher and learner complete freedom to apply any method. The reason that the 2 sigma challenge is so hard to meet is that it compares any and every pedagogical method, chosen exactly when needed, in full and open communication with the learner, with specific pedagogical methods. This is an unfair contest. One-to-one tutoring is not a pedagogy: it is a situation in which any and every pedagogy could be used. It allows tutors to ride the wave of the adjacent possible, knowing more about the learner than most (if not all) other forms of teaching and being able to adapt and change almost instantly as both teacher and student learn about themselves and each other. It would still be better even if a particular method met the 2 sigma challenge because tutors simply could adopt it too. Because teaching is a soft technology, such a method cannot be perfect, so there are always ways to improve it, and tutors will still remain ahead.

Revisited: Matching Teaching Style to Learning Style Offers No Significant Benefit

One problem with learning style theories is that it is extremely difficult to find reliable evidence to back their claims because pedagogies are soft. Every enactment of them is unique and deeply determined by context. It is rarely if ever possible to construct directly comparable learning experiences independent of the skill and artistry of their creators. The case is similar to that of the no-significant-difference phenomenon, inasmuch as researchers must attempt to control a single technological variable (whether medium, method, or other factor) without consideration of the usually vast number of other factors that affect and are affected by it. There are, and must ever be, internal consistencies and inconsistencies of the technologies to consider too. For instance, if one wished to create an experience for a learner with a “reading” style, then it would make no sense simply to remove the pictures from an existing text or, vice versa, to remove the text for a “visual” learner.

It becomes significantly more complex when multi-dimensional learning style theories are used, such as the Felder-Silverman model (Felder & Spurlin, 2005), that allow for a wide range of combinations and blends, the absence of any one of which might be harmful (if the theory is valid). There is therefore a wide range of interdependencies among and knock-on effects from all choices made, whether they are about media, structure, attitude, values, or whatever, all of which radically change the technological assembly, making comparison between two instances pointless. If just a word or two or even a verbal emphasis can make a difference to learning (and it can—try shouting an offensive expletive at the top of your voice in the midst of a lesson if you do not believe me), then the massive restructuring involved in choosing a different medium, or in changing the structure of an explanatory text, let alone in altering the underlying pedagogical methods, can never allow us to reliably identify a single aspect of the experience that had the desired effect.

The situation is made worse since, if our preferred model is accurate, then presumably it is also accurate for the person designing the experience, who most likely has more or less skill in designing learning experiences for learners with different learning styles. It would do no good to our research study to assign different learning designers to different learning designs for different learning styles because the skill and artistry of the designer would play an even more prominent role. As a result, all that we can reliably compare is one (highly situated and unrepeatable) assembly with another.

The fact that even enthusiasts agree that all learning style theories appear to work more or less as well as one another (Felder, 1996; Kanadli, 2016) supports this view. Perhaps they just represent facets of a larger whole, but equally any effects that they appear to have might result from something else entirely. Hattie (2013, pp. 246–250) observes that, if learners have a particular learning preference, then it might mean that they enjoy the process more and thus tend to achieve more than they would otherwise; however, if it is enjoyment rather than learning style that matters, then almost certainly there will be better ways to make learning more enjoyable than to match teaching to a learned preference. Ability can make a significant difference across many dimensions, including simple hard literacy concerns such as reading proficiency, typing speed, or skill in operating image manipulation or multimedia editing tools. For instance, Nguyen (2016, p. 33) relates the following quotations from students interviewed about their learning styles:

I prefer visual, auditory and reflective style and online learning contains those. However, I still haven’t got used to this because my computer skill is weak. I can only type slowly.

When the teacher asks me to upload a video presentation or do an audio reflection, I don’t know how to do it. I have to get a lot of help from my friends. Anyway, I will try to improve my computer skill after this course.

If we lack the hard skills needed to cope with whatever a particular theory claims to be our style, then we are more likely to choose another style when offered the option. But hard skills are not innate, and weaknesses do not need to be persistent, for any lack of skill can be corrected through learning. It is also highly context sensitive. If, in our main occupations or activities, we need to exercise different skillsets, then it is highly likely that we will consistently apply appropriate styles in those contexts, regardless of any presumed innate tendencies, and thus we will get better at using them. Larger-scale context can make a big difference too. For example, Carr (2013) notes that there are large cultural differences among learning styles/preferences that exist at least at a national level, making findings that appear to be valid in one culture spurious in another.

Despite all this, were one determined to persist with a certain learning style model, the innate softness of all learning technologies means that there would be little point in doing so. It would still not allow one reliably to predict future improvements/failures to improve because there is always an indefinitely large number of other ways in which one might orchestrate the phenomena differently to achieve a better or worse outcome that we (and everyone else) never thought of and countless ways in which, in the future, we might do it better or worse. Just as in the no-significant-difference phenomenon, so bad methods can be enacted well, and great methods can be enacted badly. It all depends on the assembly, not just on the individual parts of it, and technique (how it is done) often matters more than method (what is done).

Perhaps the most damning indictment of the learning styles concept, though, is one that even dyed-in-the-wool learning style theorists would have to admit: it is extremely likely that, if they exist at all, persistent learning styles are not innate but learned. There is a high probability that individuals were taught or discovered sets of methods and tools that worked for them in the past—most likely in early childhood—and that they have honed their techniques more effectively than other approaches that might be even more effective if they took the time to learn them. Like any technique, practice increases competence. The reasons for initial preferences might be many and varied, from being associated with subjects/skills that we enjoy to liking a teacher. Such effects probably start early in life and certainly are reinforced by schooling. This leads to a self-reinforcing path dependency, a hardness that determines much that follows. The simplest explanation of any learning style—if such a thing exists—is that it is a being-taught habit. It is just a set of pedagogical techniques that we have learned. Because we rarely have a choice in the real world, outside formal education, about how the things that we need to learn are presented to us, it would be foolish to continue to reinforce that habit rather than learn to learn in other ways.

Whether or not learning style theories are valid, they do provide us with stories that help us to make sense of the world in different ways, which can help to catalyze new ideas and imaginative ways of learning and teaching. It is undoubtably a good thing for designers of learning to be aware of a diversity of learning strategies and techniques, and to use them, especially when teaching others. No doubt every learner is different, and all can benefit from approaching a topic or skill in different ways at different times in different contexts. In some ways, learning style models can be seen as coarse, caricatured, but still potentially useful ways of building personas (Pruitt & Grudin, 2003), or parts of them, which have been shown repeatedly to be effective design tools that make it easier for us to imagine the needs, interests, skills, and desires of our target audience. Given that education, fundamentally, is a design/performance discipline, this might be as much as we can hope for in a theory or model.

For example, despite finding the Kolb learning style model extremely problematic, I frequently make use of Kolb’s learning cycle—actually, as Kolb and Kolb (2005) acknowledge, the invention of Kurt Lewin—because I know it well, and it serves as a useful reminder of four of the most significant kinds of activity that should be considered in most learning trajectories: abstract conceptualization, reflective observation, active experimentation, and concrete experience. It is not always a cycle, and the order can vary because, being a technology, the best order in any given situation depends on other design choices. These are high-level learning strategies, not styles, and knowing about different strategies can be helpful when faced with a blank screen and a need to develop a learning intervention. Similarly, I often use Pask’s (1976b) serialist/holist model of learning strategies to remind myself that there are always different ways to learn the same things and to reduce the likelihood of falling into the trap of teaching the way that I would like to be taught. Such models can be useful intuition pumps whether we believe in them or not. It is almost always possible to distinguish learning or teaching strategies from learning style models, and it can be well worth doing so. Technologies do not have to rely on accurate models of the world in order to be useful. They do not need to be applied science.

Revisited: Experimental Educational Research Appears Not to Work Well

Reductive research methods, following those of the physical sciences, are mostly intended to seek, confirm, or deny underlying simple causal relationships: if x happens, then y will occur. In most cases, whether the discipline is softer or harder, the main purpose of performing reductive research is thus usually to test some generalizable law or principle, normally by attempting to disprove a hypothesis in a manner that could be replicated. In harder disciplines such as physics or chemistry, studies normally attempt either to replicate both methods and context or to apply different methods to the same context, the assumption being that natural phenomena should behave in the same way no matter where they happen in the universe. In softer disciplines such as education, replications are seldom as direct as those in harder disciplines. Given the situated nature of education and the inevitable softness of the technologies concerned, the context differs every time.

Distributed participation in all learning, with the learner always playing a role in orchestrating the final assembly, means that the same conditions can never hold twice. Although obviously and trivially true when looking at different individuals, the problem is even worse when we attempt to repeat an intervention for a single individual. As Smedslund (2016) argues, once some mental event (e.g., learning) has occurred, it must irreversibly change the person for whom it has happened, so no experiment can ever be repeated on the same person. Even if it could, Smedslund notes, the attempt would be scuppered by a combination of infinite possible contexts (situations that can and do never repeat) and social interactivity (in which there are and always must be many more influences than those controlled by the experimenter). Smedslund argues that, in any intervention that relates to psychological states (especially including learning), it is thus impossible to predict specific outcomes based upon prior observed behaviours and averages.

Following from this, given the complexity and variability of context, in most meaningful educational interventions apart (perhaps) from the hardest and most invariant, it would make little or no sense for researchers to follow identical procedures, because each educational intervention must have unique aspects that need to be controlled for differently and examined differently. As we have seen, virtually all interventions demand soft technique and creativity from their participants, which almost invariably makes all the difference between success and failure. Educational researchers therefore must attempt to follow conceptually similar procedures in necessarily different contexts. Conceptual similarity, however, is a vague concept in itself. These methods are technologies like any other, and there is softness at their heart, and thus small differences in technique that are virtually impossible to identify in their entirety and that are as reliant on skillful use as the phenomena that they seek to examine, can have significant impacts on the results.

When objectivist and reductive methods are applied to human-created technologies, such as pedagogies or tools for teaching, it is unlikely that researchers will find universals that resemble the laws of nature, because technologies are inventions that exist in complex and ever-shifting relationships with one another, continually evolve, rely on technique for their enactment, and seldom occur in similar enough assemblies to reliably infer causal mechanisms. Softer technologies might never occur in the same way twice. In assembly, they might orchestrate dozens, hundreds, or thousands of different phenomena, each of which can affect the whole (and one another) in unpredictable ways. Furthermore, minute differences across many qualitative dimensions (unpredictably) can have large effects, and the vast range of possible combinations can result in emergent phenomena that could not be predicted from simple cause-effect relationships, even if all were known individually.

These phenomena (intentionally or not) can uncover or confirm more atomic phenomena that are parts of the orchestration, such as (in education) ways that people learn or feel about things in general or (in engineering) the load-bearing properties of materials. However, this is not research on the ways and means of teaching and learning. It is simply using the technology in question as part of a different technology, designed to discover natural laws. Unfortunately, as discussed in Chapter 4, in an education system, the laws so discovered might not always be as “natural” as they seem because of the effects of our prior inventions on subjects’ behaviour. For instance, it is possible to discover “laws” such as the intermittent punishment effect (Parke et al., 1970) that appear to indicate motivational benefits of rewards and punishments, but this is the case only in a system that, through rewards and punishments, has killed students’ intrinsic motivation in the first place (Kohn, 1999).

Unfortunately, though the natural laws that we might uncover can provide (when well researched) a useful set of phenomena to orchestrate, even when we do find consistent causal laws of learning (and almost certainly there are such things), it might not help us much in designing real-life education because it is the assembly, in all its rich and deeply intertwingled complexity, that matters: many causes compete with or enhance one another, and emergent behaviours occur all the way down the line. From a design perspective, it is useful to understand how individual design ingredients tend to work but knowing how people learn (say) does not predict successful teaching any more than knowing how the engines of cars work predicts successful driving or the bristles of a paintbrush predict successful painting. It can help to explain phenomena that occur but not to predict reliably in advance what skillful practitioners can do with those phenomena, in artful assembly with countless others, many of which will never be repeated in the same way again. It could be argued that any knowledge is better than none, but there are complex interactions between these simple parts that make their effects inherently unpredictable. For example, we know that properly spaced learning provides a more effective way to remember than unspaced (or poorly spaced) learning, so it can and often does make sense to use this fact to help remember something, but if what is being remembered is boringly presented, personally irrelevant, or overly traumatic then the results might not be those anticipated. This is a simple example: most real-life situations are far more complex and intertwingled.

All that said, if the technologies that we examine are sufficiently hard and invariant, then we might discover consistent and useful facts about how they behave individually, but on the whole this is mainly valuable in the same way as knowing that steel is harder than paper. It provides us with a better range and understanding of components that we might use to build our technologies, but it does not determine or predict whether those technologies will work how we wish them to work when we put them together.

There are a few useful ways to apply reductive research to technological phenomena, most notably when the assembly itself is very hard—technologies such as SATs, self-paced online courses, and so on—and likely to be repeated with little or no variation many times. Based upon findings, we can make adjustments to those (and only those) hard technologies, adding or changing parts of the assembly, we can observe their effects, and we can explore differences among contexts to gain a fair idea of what works and what does not. Such methods—sometimes described as A/B testing (Dixon et al., 2011)—are common in technology design and often lead to improved inventions.

However, though it can be useful to the creator of that specific hard technology to know whether it works as intended or not, that in itself does not tell us anything that we can generalize reliably for a different tool, method, or process or whether the technology itself makes any sense in the first place. It can tell us little about whether the same technology applied in a different context would work the same way or whether parts of the assembly will behave in similar ways in different configurations. It can help us to understand the pieces, and others might use those pieces in their own assemblies, but it tells us little of value about what happens when they combine in even slightly different ways. This is extremely important if we wish to improve our practices.

Unless we can identify causal relationships, we have no way of knowing what it was about the system we used that did or did not work, and we have no reason to repeat or remove it. But causal relationships in educational contexts are parts of a complex adaptive system that makes it fundamentally impervious to study using methods that reduce complex phenomena to their component parts, as Kauffman (2008, 2009, 2016, 2019) persuasively argues. From those parts, we can never anticipate the boundary conditions of what can emerge in advance, and we can never pre-state what their functions will be by looking at their constituent elements. This means, in a real sense, that the behaviour of the parts does not predict the behaviour of the emergent whole, even though they may cause that behaviour to occur. The problem is not that the smaller-scale phenomena do not mechanistically combine to cause the larger-scale phenomena—of course they do. The problem is that, in a system of any complexity, emergent behaviours are meaningful only in the contexts of the systems to which they belong and cannot be understood by reduction to their component parts. We might know everything worth knowing, say, about cells in a body, but that would not help us to predict or explain the role of a heart in a circulatory system (Kauffman, 2019).

It is tempting to seek reductive empirical knowledge of educational practices because clearly there are causal relationships between what we do and how we learn. We can easily see that some teachers are more consistently successful than others, and it is tempting to ascribe that success to whatever pedagogies and other tools they use: to abstract what they do from how it is done. However, what makes them successful rarely has much to do with any specific method in isolation. Instead, good teachers tend to adapt to the learners and the surrounding contexts as needed. Just as I am at least as reliable a predictor of my local weather as the best meteorologists with the biggest and fastest computers if all I have to do is predict what it will be in five minutes, so too it is possible for a teacher, with sufficient indicators about a learner, to adapt a way of teaching to that individual’s needs when it matters. As Hattie (2013, p. 17) puts it, “the art of teaching, and its major successes, relate to ‘what happens next’—the manner in which the teacher reacts to how the student interprets, accommodates, rejects, and/or reinvents the content and skills, how the student relates and applies the content to other tasks, and how the student reacts in light of success and failure.”

We are good at identifying ways of reacting in the short term, but we are bad at predicting the results of our interventions in the long term, even though we often adopt a convenient illusion—usually based upon average effects—that we do know what has an effect. Unfortunately, averages are not useful when dealing with human beings because, as Rose (2016) observes, almost certainly there is no such thing as an average person. Rose gives the example of aircraft seats designed to fit the average pilot that, because there was literally no average pilot to be found, resulted in many plane crashes. There are reasonable grounds to suppose that education presents a similar case: an intervention intended to suit the average learner, in all likelihood, might suit no learner at all or even be harmful.

Teaching is more like sailing in changeable winds than like driving on well-marked empty roads. Knowing the direction in which we are headed, we must adapt constantly to conditions as they change around us, not keep to an unerring path. There are methods that we can and must learn, techniques that we can and should hone, rules of thumb that we can apply, but each circumstance demands different responses, and there is a high likelihood of encountering novel situations on a regular basis. To do this successfully, we need diverse skills more than mechanical procedures. We must be tinkerers and bricoleurs rather than engineers. Again, this implies that many of the most fruitful research avenues in education lie not in identifying effective teaching methods but in identifying effective teachers and what makes them so. When we find them, we can tell stories, perform design-based research (Anderson & Shattuck, 2012), offer rich case studies, use appreciative inquiry (Cooperrider & Srivastava, 1987; Cooperrider & Whitney, 2011), and so on so that others can gain inspiration, adapt our pedagogies, or imitate our tools in their own orchestrations in their own contexts. In doing so, we can add to and refine the technologies of learning as well as help others to develop their own techniques. This seems to be valuable enough in itself. We do not need to aspire to the kind of predictive certainty found in hard sciences.

The softer the technology, the harder it is to make accurate predictions about it. It would make no sense at all, for instance, to claim that email is good (or bad) for learning, and there is no research method that could be used to prove it unequivocally one way or the other, because there is virtually nothing fixed about the phenomena that it can utilize or the orchestrations of which it can be a part. For very soft technologies, the only research methods that make sense are those that attempt to investigate something about how they are used, not to discover their relative effectiveness in general. However, it is then that the (always unique) orchestration matters rather than the parts of the assembly.

As greater hardening is applied, the number of points of comparison becomes more salient, and it becomes more possible (though seldom particularly wise) to make more general (though always provisional) statements about them. For instance, though LMSs are mostly very soft technologies (for teachers) and can be assembled with other technologies to soften them further, they do impose many pedagogical constraints that lead to greater consistency between instances that, in some cases, can make them more comparable. They create adjacent impossibles as well as possibles. We might be able, for example, to generalize, albeit within a limited domain, about a particular instance of an LMS’s discussion tools (e.g., the effects of limiting the size of text boxes on discussion posts) or quiz modules (e.g., the effects of being able to display only a single question at a time) within a specific pedagogical orchestration, though all bets would be off were we to compare different versions of the same tool across different orchestrations. Although they might cause issues in some orchestrations, almost certainly there would be ways of using the constraints to some pedagogical advantage in others, not to mention ways of assembling them with other tools, so we should resist drawing firm general conclusions.

The fact that every meaningful learning transaction is irreducibly unique does not mean that we should despair of seeking patterns or be wary of reifying soft techniques into harder tools and methods. Quite the opposite. This kind of sense making, pattern recognition, and pattern formation is how we make progress, how we learn to teach better, how we learn better in general, both individually and collectively. That said, those of us who seek to understand and research how this happens are left with what appears to be an intractably complex, complicated, and deeply situated view of learning and teaching, which might make one despair of ever being able to come up with any generalizable rules about how it should be done or, indeed, to make any reliable inferences about how learning occurred in any given instance. It is the essence of the art and nature of learning and teaching, a constant and ever-shifting interplay between knowing and discovery, in which what we do affects the structure of how we do it, and that structure in turn affects what we do. It is a constant and never-ending state of becoming, a creative evolutionary process with no beginning and no end. It is what makes us human.

This lack of generalizable predictability should be a cause for celebration, not for despair. If this is the nature of the beast, then we can identify ways to make the most effective use of it and to avoid pitfalls that await us should we get it wrong. What matters most is therefore awareness of how learners are learning and which effects our teaching is having. This has to be combined with a broader understanding of how the technologies of teaching work—all the technologies of teaching, from pedagogies to Google Search—and how they can work together. Teachers—all of us—are orchestrators of technologies that, if used effectively, at the right time, in the right place, with the right people, can lead to learning.