OVER THE LAST FORTY YEARS, the growth of digital games has profoundly changed the landscape of entertainment in the networked world. Some of my earliest childhood memories are of computer gaming: batting balls with handheld wheel controllers in Pong on an Apple II+, blasting space aliens in Zaxxon on a Commodore 64, and typing “go north” to venture into the unknown in the Zork text adventures. In my 1980s childhood, video gaming was a niche hobby marketed primarily to young boys, but the gaming industry now rivals film and television in size, scope, and cultural significance. People of all ages, all genders, and all walks of life play billions of hours of games every year.1
Games offer a nice microcosm of learning at scale and a good place to recap the major themes from Part I of this book. In the early 2010s, games and “gamification”—the process of adding game elements to learning technologies—experienced a surge of interest against the backdrop of the widespread growth of gaming. Futurists imagined a more playful future for schools, and technology developers created new educational games and, in some cases, promised dramatic results. From 2012 to 2014, the Horizon Report—a publication from education futurists—predicted that “games and gamifications” were “two to three years” away from generating considerable impact in formal education. Some interesting experiments, which I will discuss in this chapter, proved to be popular and effective in a few niches within the ecology of human development, but overall, the learning game movement remains another unrequited disruption.2
Games are indisputably great engines of learning: ask a passionate Pokémon player about the virtues of Charizard versus Pikachu or the strategies for deploying these monsters in imaginary battles, and you can unleash a torrent of factual knowledge, strategic thinking, and hard-won wisdom from hours of experience. Many modern games are immensely complex, and developers deploy a variety of features for helping players learn that complexity. They gradually and dynamically adjust difficulty, adding levels, new elements, and new challenges in response to player success and development, which keeps players at the sweet spot between what a learner knows how to do without help and what a learner can’t yet do, the liminal space that psychologist Lev Vygotsky called the zone of proximal development. Games provide a narrative world of meaning, consequence, and relevance to motivate and engage players. Hinting systems, online wikis, video tutorials, and discussion boards provide as-needed resources for just-in-time learning as players seek to improve. Through these kinds of strategies, gamers are unquestionably learning and getting better at the game. The core question of learning games is one of transfer: Do people who develop new skills, knowledge, and proficiencies within a game world flexibly deploy those new insights back in the humdrum of everyday life?3
While many educational technologists make claims about product benefits that far exceed the evidence, very few developers manage to venture so far into the realm of falsehood that they attract the attention of the Federal Trade Commission. The developers of Lumosity, then, hold a special place of ignominy in the history of the 2010s edtech hype cycle for earning a $2 million fine from the FTC for false advertising.4
Lumosity develops “brain training games.” They take cognitive tests of mental capacities like working memory and divided attention, and they turn these tests into mini-games. Lumosity advertised to users that practicing these games would lead to more generalized benefits “in every aspect of life,” including improvements in school work, age-related cognitive decline, and brain injuries. If a person played a game that helped them improve their working memory in a puzzle, for example, then Lumosity claimed that their performance would improve on a wide range of real-world tasks that require working memory. In 2016, the FTC found no evidence to support these claims, and psychology researchers conducting experimental evaluations of these programs also found no evidence of these general benefits.5
In psychological terms, the Lumosity advertisers were making a claim about the concept of “transfer,” the idea that what people learn in one situation (such as a game) can be applied to novel situations (such as in day-to-day life). One of the first psychologists to study transfer was Edward Thorndike, the pioneering education scientist whom we met briefly in Chapter 1. In the early twentieth century (and long before), educators and curriculum developers claimed that the rigorous study of Greek and Latin built up “mental muscles” that, once strengthened, could be productively used for tackling other cognitive problems. Transfer emerged as a critique of this line of reasoning. Thorndike observed that when learners developed new knowledge or skills, they were far more likely to be able to apply those skills in novel contexts if the new context had many similarities—Thorndike called them “identical elements”—to the original learning context. Situations that are only slightly novel are known as near transfer; if you learn to drive in a sedan and then hop into a station wagon, you are practicing near transfer. Learning contexts that are substantially different from the original are known as far transfer, like learning to drive a car and then trying to fly a helicopter, or learning Latin and then trying to do math, or playing puzzles on your phone and then being a better thinker in everyday life.6
While Lumosity’s claims were sufficiently specific and incorrect to merit regulatory attention, claims about the general cognitive benefits of games and pastimes are quite common. For example, many people believe that chess experts develop generalizable strategic thinking skills. A recent meta-analysis examined studies of chess training, music training, and working-memory training, and found little compelling evidence that any of these three practices improved people’s general cognitive performance. It turns out that chess expertise primarily depends upon an encyclopedic knowledge of common chess moves and board positions; if you show chess board positions that come from actual, realistic chess situations to both masters and novices, masters are much more likely to be able to recreate those situations from memory. If you show chess masters and novices board positions that are nearly impossible to occur in actual chess games, then masters have little advantage in re-creating the nonsense boards. The knowledge of common board positions is essential to getting better at chess, but this knowledge is mostly useless when trying to play other games or conduct other strategic tasks. Similarly, the research from Luminosity shows that people who play working-memory games indeed get better at other working-memory games (near transfer), but getting better at these working-memory games does not help with other kinds of cognitive tasks (far transfer).7
The implications of this feature of human development are quite significant and quite challenging for educators. As a society, we hope that schools can teach domain-independent, broadly useful skills like critical thinking, collaboration, and communication, but it turns out that most skills are actually quite domain specific—thinking critically about a chess move requires different knowledge and skills than thinking critically about the interpretation of a novel. This also proves to be a substantial challenge for the field of educational games. Part of what makes games fun and engaging is immersing people in an alternate world, but theories of transfer suggest that the more distance between those alternate worlds and our own, the less likely it is that learners will be able to deploy game-world learning in the real world.
So if far transfer doesn’t work, are educational games worth pursuing? Learning games, like adaptive tutors, have been used long enough in schools and other settings that a track record of research exists about their effectiveness. One of the best ways to evaluate a class of learning experiences is to look not just at an individual study but at collections of studies, or meta-analyses. In a meta-analysis, researchers collect a set of published research studies—usually experimental and quasi-experimental designs that draw comparisons between an intervention group and a control group—and draw comparisons across a whole set of findings. Two major meta-analyses of classroom use of games were published in 2013 and 2016, and they both pointed in the same direction. Across the studies, students who participated in game-based learning experiences had modestly better learning outcomes on measures of knowledge and intrapersonal domains like intellectual openness, work ethic, and contentiousness. Playing games over multiple sessions was more effective than one-time games, and when basic versions of a game were compared with versions with more advanced and theoretically informed features, the more advanced games led to better results. Yet even experiments using games with simple mechanics—limited narrative, goals targeting lower-order thinking skills, basic content exercises with badges, stars, and points layered on top—showed modestly better outcomes than control conditions without games. Of course, what happens on average won’t perfectly predict what will happen in any particular classroom or school, but these kinds of studies provide some useful guidelines for reasonable expectations. Using learning games as part of teaching can probably lead to modest improvement in student learning and motivation. Enthusiasts promising a dramatic transforming of schooling and learning through games and gamification should be regarded with skepticism.8
The research on learning games isn’t overwhelmingly negative or positive, and the effects of individual games vary. But by using the concept of transfer and applying the genres of educational technology at scale that we have learned over the first part of this book, we can imagine how individual games might interact with school systems. Think back to the last educational game that you played. Who designed the order of your activities and experiences in the game world? Did you move from one set piece to another, and was the order of set pieces determined by the designers? Did your actions or answers in one part of the game trigger algorithmic decisions that determined what happened next? Did engagements with peers shape your playing experience? Most learning games fit reasonably well into one of the three genres of instructor-guided, algorithm-guided, and peer-guided learning at scale. Placing games in those genres helps throw into relief where any given game might provide targeted benefits to some learners. Other strategies introduced throughout the last three chapters—asking “What’s really new here?,” reviewing published evidence of effectiveness, and finding the alignments or misalignments between learning technologies and existing educational systems—can all prove useful in reviewing learning games and imagining how they might support learning in different parts of the education landscape, even if we can be confident that they won’t profoundly transform schools.
Most learning games are simple to classify, and most are instructor-guided experiences. For many in my generation, Math Blaster was the first learning game they encountered. The on-screen playing field was arranged roughly like Space Invaders with aliens from the top of the screen descending upon a village below while the player shoots laser beams at the aliens. In this case, however, the lasers are inexplicably powered by math problems, and the game stops periodically to have students answer a question, where the variables in the questions are randomized but basically arranged in predesignated sequences. As students complete the sequence, they get to do harder problems. My nine-year-old daughter has a math app from school called XtraMath where the conceit is different—she’s racing against a “teacher” to provide math facts (for problems like “12 minus 9”), but the mechanism for XtraMath and Math Blaster are basically the same: solve simple math problems, get points, solve harder math problems, get more points. These instructor-guided games exist on a walled-garden platform or inside a software package or app, assess student performance and progress through pattern-matching autograders, and draw pedagogical inspiration from pail fillers rather than flame kindlers.
Game researcher Brenda Laurel developed the vivid analogy of “chocolate-covered broccoli” to describe these kinds of games. The core activity in Math Blaster or XtraMath is no different from the core activity on a worksheet: solve arithmetic problems. Since many students experience worksheets as dreadfully boring, game designers add a layer of points, stars, beeps, and other rewards on top of drill-and-practice activities. Underneath this layer of external rewards and incentives are very traditional math activities. The process of pouring behaviorist chocolate over instructionist broccoli is often described as “gamification,” and these practices have a broad foothold in schools. Gamification can be found as elements in learning software, like the points and badges awarded in Khan Academy. Platforms such as Kahoot allow teachers to author their own content within a gaming platform, turning typical classroom routines such as quizzes and review sessions into classroom game shows.9
These approaches to gamification fit relatively easily into traditional school settings by making minimal changes to what David Tyack and Larry Cuban call the “grammar” of schooling, or the unquestioned processes, beliefs, and assumptions deeply embedded in the educational system.10 As we’ve discussed in previous chapters, the grammar of schooling tends toward Thorndike’s pedagogical philosophy, in which students learn through organized, direct instruction, and their learning can be measured. Gamified learning exercises are simple to use and short to play, making them easy to assign in class in lieu of similar kinds of activities. They take worksheet problems and add game elements to them. One purpose of doing worksheet problems is practicing for the kinds of classroom and standardized tests that serve as gatekeepers to advancement in the education system. In that regard, learning games have the advantage of aiming to bridge a problem of near transfer, using games about problems found on math tests to help people do better on math tests.
Relatively few learning games have managed to become breakout hits beyond the classroom, but one of the most successful efforts in recent years is the gamified language-learning app Duolingo. A game-like, algorithm-guided, adaptive tutor for language learning, Duolingo was cofounded by Luis von Ahn, a computer scientist at Carnegie Mellon and the inventor of the CAPTCHA crowdsourcing system. Most Duolingo activities are some form of translation or recognition activity, where students earn points, complete progress bars, and earn badges for translating text between the target and native language in speech and writing. As of 2018, over 300 million people had signed up for an account with Duolingo, making it one of the largest platforms for independent learning in the world. One of the distinctive features of Duolingo is that it includes adaptive features that allow for personalized spaced repetition. These adaptive features offer some interesting targeted benefits for learners, but as with other algorithm-guided large-scale learning technologies, limitations of autograding are an important constraint on the overall utility of these kinds of tools.11
Going back nearly a century, psychologists have recognized that people remember things better when they practice recalling them over a long period of time rather than through cramming. If you have a choice between studying for an hour one day before a test or studying for twenty minutes each of the three days before a test, the spaced practice is almost universally better. These systems can be improved further if the studying experience focuses most on the facts or topics that a learner remembers least well. When studying language facts on flash cards, learners should spend little time on the flash cards they always get right (despite the emotional rewards of doing so) and nearly all their time on flash cards that they always get wrong. The benefits of spaced repetition are some of the oldest and most well-established findings from cognitive psychology with obvious implications for learning, but they are very rarely implemented in actual classrooms.12
Your Spanish teacher probably could not implement personalized space repetition in your class because it is very logistically complicated—an instructor has to identify word definitions or other language facts (like verb conjugations) that each student has not mastered, provide opportunities to practice these facts alongside introducing new content, and then slowly withdraw facts from practice as students demonstrate mastery. Tossing a few vocabulary words from week two on the week-six test is not too difficult for an instructor, but personalizing practice tests for dozens of students on the basis of their individual progress and mastery is nearly impossible for a typical teacher to organize. Computers, however, can implement these complex, personalized schemes of spaced repetition for each student. The results are promising, at least for the introductory parts of learning a language. In 2012, independent researchers found that Duolingo users who spent an average of thirty-four hours on Duolingo learning Spanish would learn material roughly equivalent to the first semester of college Spanish.13
An interesting feature of a language-learning curriculum is that midway through a typical course progression, the cognitive complexity of the learning sharply increases. In introductory Spanish, students are memorizing the Spanish word for cat and how to conjugate the verb to have. In advanced Spanish, students are reading and interpreting Cervantes. Autograders are much more useful for the former kinds of tasks with well-defined correct answers than for the latter kinds of interpretive tasks, so it is unlikely that Duolingo’s usefulness in learning introductory language concepts will extend to more advanced language acquisition skills. The assessments in Duolingo can evaluate whether a person has defined or translated a word or short phrase correctly; they cannot evaluate a student’s arguments for the impact of Don Quixote on Spanish literature and culture. Language-learning games may be a great way for people to start learning a language, but for the foreseeable future, developing real fluency will require engagements with native speakers and culture that are not possible through an autograded app.
If the chocolate-covered broccoli approach is to slather gamification elements on top of traditional schooling activities, then the alternative is to search within content areas to find fun and playful elements that already exist. In this approach to developing learning games, the fun isn’t getting to shoot aliens after doing some math, the fun is doing the math.
My colleagues in the MIT Education Arcade develop what they call “resonant games,” games that try to immerse players, individually and in communities, into activities that are personally engaging and provide rich insights into academic content. If most learning games align best with the instructionist, banking model of education—filling students with content and testing their recall—then the Education Arcade’s resonant games tend to look more like the flame-kindling, apprenticeship model of learning, where the game world immerses players in some kind of cognitive apprenticeship.14
For instance, the game Vanished doesn’t quiz people on science facts, but rather immerses players as members of a community of scientists trying to understand the fate of a lifeless planet. Developed by researchers and designers at the Education Arcade in partnership with the Smithsonian Institution, Vanished was a massive multiplayer puzzle experience played out on the open web—a kind of online escape room for thousands of people at once. In the game narrative, scientists from the future send messages back in time to warn humanity about a forthcoming asteroid-induced apocalypse. They communicate to the players through a series of puzzles and mini-games hidden throughout the web and at physical museums. Some of the puzzles required large-scale collaboration; each user randomly received one of ninety-nine codes that needed to be assembled to solve a puzzle. Players could purchase documents with points that they earned throughout the game, and some documents were so expensive that players could buy them only while pooling points. Over 6,000 player accounts were registered in the game, and over 650 were active toward the end.
While not explicitly influenced by connectivism, the game has much in common with connectivism-inspired pedagogies—participants in the game form networks, communicate with one another, share resources and solutions. Players both learn about a topic (in this case, climate change that renders humanity incapable of responding to the asteroid threat) and develop a shared identity as scientists. The story provided an opportunity for players to learn about a range of scientific content areas, from unit conversion to forensic anthropology, but perhaps more importantly, it gave participants the opportunity to develop their identities as scientists. As one player wrote as part of an evaluation of the game, “I really feel like a future scientist now. Imagine, when we all have famous jobs at research centers across the world, someone will discover how we all as kids worked on a game.” Vanished shares a series of common challenges with other peer-guided learning experiences: attrition limited the number of learners who benefited from the full experience, participant experiences were idiosyncratic and learning outcomes uneven, and a long-term, unfamiliar learning experience was challenging to integrate into traditional classroom practices. For those who invested deeply in the experience, however, the game provided a uniquely powerful learning experience that was not just about science but also about what it is like to be a scientist.15
Math Blaster, Vanished, and Duolingo represent efforts to create games or gamified experiences designed for educational purposes, but another approach to learning games is to find commercial games that have both widespread appeal and the potential to foster powerful learning experiences. Rather than making and marketing new games, what if educators could take existing games and use them for teaching and learning academic content? In 2020, Minecraft and Minecraft: Education Edition represent one of the most ambitious efforts along those lines.
Minecraft is one of the world’s largest learning communities—millions of young and young at heart around the world play, build, and explore in the Minecraft world. Minecraft is an open-world game made up of square blocks of various kinds—dirt, stone, sand, water, lava, iron, gold, diamond—that can be mined with tools and then recombined to make Lego-block-like creations in the game world. Resources that drop from different blocks can be combined to create a wide variety of items, from new tools, weapons, and armor to decorative elements like doors and carpets. It is among the most popular games of all time, with over 180 million copies sold. Minecraft worlds can be set up as multiplayer servers so people can enter the same world and play, build, share, and collaborate.16
The scope of the learning community on Minecraft is extraordinary in its breadth. Like many recent games, Minecraft is highly complex but ships without any kind of manual. Through in-game experimentation and examination of game files, users essentially have cocreated the manual themselves on sites like the Gamepedia Minecraft wiki. That wiki is maintained by over three hundred active contributors and has over four thousand articles on topics that range from the original staff of the Swedish game developer Mojang to the probability distributions of finding diamonds at various depths in the game world. There are countless YouTube accounts devoted to demonstrating features in Minecraft, including genuine global celebrities such as Joseph Garrett, also known as Stampylonghead, whose YouTube channel has 9 million subscribers and whose How to Minecraft introductory series has tens of millions of views. Twitch has hundreds of active streamers playing Minecraft at any given time; Reddit has a subreddit with photos of incredible creations and gifs of funny moments. Any one of these content distribution channels is an endless rabbit hole of narrative and creativity.17
All of these digital learning resources form a dense, complex, and intricate peer learning network, with resources to guide players from their first actions in the Minecraft world to the complex management of resources necessary to visit the far reaches of the game world or to create massive built environments atop the randomly generated game world. There is an easily recognizable peer-guided learning-at-scale network that teaches people about Minecraft, but to what extent can the game itself be used for learning both within and outside of formal education systems?
Given that Minecraft has consumed billions of hours of youth time over the last decade and is increasingly being seen in the classroom, there is surprisingly little research about the experience of playing Minecraft and what benefits it might accrue. Studying these kinds of informal learning environments is profoundly difficult. When students come to school, we have particular goals for them, we give them assessments, and we combine these assessments with observations of classroom processes to understand what learning is taking place. Tracking the learning that happens across tens of millions of households as kids play informally is much, much harder.
Researchers who study play and informal learning can point to a variety of behaviors in Minecraft that could potentially lead to positive outcomes for young people. Games encourage discovery and perseverance, requiring self-regulation in single-player settings and communication and collaboration in multiplayer settings. Game designer and researcher Katie Salen suggests a list of Minecraft related skills: teamwork, strategic communication, asking for help, persistence, recovery from failure, negotiation, planning, time management, decision-making, and spatial awareness. This is the optimist’s view of transfer, that time invested in learning these skills in a game like Minecraft will translate into domain-independent skills. Teamwork skills developed through collaboratively building a castle in Minecraft might prove useful when building teams in a workplace or civic setting.18
A more pessimistic view built upon the research on transfer is that people who spend a lot of time playing Minecraft will primarily learn about playing Minecraft. If people who invest considerable time mastering chess do not appear to be developing domain-independent strategic thinking skills, then it is not clear that people playing Minecraft are necessarily developing any particular domain-independent architectural or design skills, or any of the other skills developed in Minecraft.
My own view is that there probably are some domain-independent problem-solving skills that players develop in Minecraft. For instance, players might develop the intuition that if they get stuck on a task, there is likely to be a community of teachers and learners online who have documented solutions to similar tasks or would be willing to engage in online dialogue about how to solve the problem. As the last chapter on peer-guided learning at scale suggests, that is an enormously useful disposition to adopt in a networked world, because in nearly any domain of human endeavor, there are people online willing to help. That said, there are particular skills and facts required to take advantage of each of those distinct networks of learning. To figure out how to do something in Minecraft, a player must understand the mechanics of the game, the vocabulary of the game, the most common repositories of reliable knowledge, and the norms of information dissemination and online discourse. One has to learn similar things to figure out how to write a particular function and debug a program in JavaScript, but in another domain: the mechanics and syntax of the JavaScript language, the vocabulary of its functions and primitives, the common repositories of reliable knowledge, how to search within Stack Overflow, and how to ask new questions within community norms. The domain-independent skills, a disposition to seek out peer-guided online learning networks, and an understanding of their common structures are a little bit helpful in many different circumstances; extensive content knowledge in a domain, like memorizing common board positions in chess, is essential to mastery in a local context and not very useful elsewhere. These distinctions do not reduce the value of the broader disposition of online help-seeking and participation in online networks, but that disposition alone will not bridge the chasm of far transfer. Domain-independent skills are slightly useful in lots of different domains but not deeply useful in any particular domain.
Recognizing the popularity and potential for learning in Minecraft, some classroom educators have explored making tighter connections between Minecraft play and classroom learning. At Carnegie Mellon University, materials science professor B. Reeja Jayan used Minecraft to teach the basics of material science to engineers by having students build models of atoms and molecules out of digital blocks. The Minecraft universe includes a substance called “redstone” that can be used to create circuits within the game, and educators have developed a variety of tasks for learning the basics of physics, electrical engineering, and computer science. In a kind of learning-at-scale special crossover episode, faculty at UC San Diego have created a MOOC about teaching coding in Minecraft.19
The open-world design of Minecraft is terrific for play and exploration but imperfectly suited for targeted content learning that happens in schools. In an effort to add more classroom-friendly features, in 2011, TeacherGaming released MinecraftEDU, a version of Minecraft developed with specific modifications to aid teachers in using Minecraft in their classrooms. The modifications made certain logistical tasks easier, such as registering students and assigning them to a world. It also allowed educators to create specific worlds with specific tasks, constraints, and game rules so that teachers could give students a more instructor-guided experience within the open-world platform. In the 2010s, both Minecraft and MinecraftEDU were bought by Microsoft, which released a Minecraft: Education Edition that provides new ways for instructors to control the game world and a set of turnkey lesson plans and game worlds that teachers can use for teaching academic content ranging from biodiversity and extinction to Boolean logic in circuits.
In a sense, these modifications run counter to some of the fundamental principles of Minecraft. For most players, Minecraft is fun because they can do whatever they want: building structures, inventing arbitrary challenges (harvesting enough diamonds to build a full set of armor), or just exploring to see if there are any llamas around the next corner. The Minecraft: Education Edition projects constrain some of this creativity; in order to teach specific content, they ask students to do what teachers want them to do. In the worst cases, the imposition on the freedom of Minecraft will spoil the fun, and the high time-cost for instructors setting up these worlds won’t be worth the modest gains in engagement. In the best cases, these additional scaffolds will form a productive hybrid of open-world play and teacher-guided problem-based learning. Little rigorous research exists as yet about Minecraft in education, but I would predict findings similar to what researchers have discovered with other approaches to learning at scale that fit uncomfortably in the constraints of formal schools: a handful of really extraordinary applications in a few institutions but no widespread adoption across many schools.
A worked example is a teaching strategy where an instructor explains, step by step, how he or she would solve a type of problem. In this chapter, I have tried to offer a worked example of a classification exercise that takes a type of large-scale learning technology, in this case educational games, and situates that technology within the learning-at-scale taxonomy to demonstrate what patterns emerge. Throughout the chapter, I provided several examples of games that fit well in the three learning-at-scale genres: Math Blaster as an example of an instructor-guided learning experience; Duolingo as an adaptive, algorithm-guided learning app; and Vanished and Minecraft as peer-guided large-scale learning communities. Placing games in these categories provides a way to quickly frame some of their strengths and weaknesses; if you can see that Duolingo is an adaptive tutor, then you can review what you know about the history of adaptive tutors and make some good guesses about its strength and limitations. That history can also reveal ways in which a new product is distinct; we’ve had decades of adaptive tutors, but few have been as widely adopted by individuals outside of formal learning environments as Duolingo. Something about the design of Duolingo—its gamified elements, the subject matter of language learning, a mobile-first platform—led to a wider adoption than might have been otherwise expected. Designers can tinker with these standout aspects of a particular approach and consider which of them might be applicable to other technologies within that same genre or beyond.
Not all learning-at-scale approaches fit neatly within the three genres that I have proposed, and hybridity is a promising site for sources of innovation. In finding novel combinations of elements within a game or other new technology, we might be able to take inspiration in devising new approaches to learning at scale. One example of a learning game that demonstrates this kind of hybridity is The Logical Journey of the Zoombinis, one of the all-time great mathematics games, developed by Scot Osterweil.
In the game, players have a small band of Zoombinis, which are basically little blue heads with feet, who are on a journey across a series of landscapes to find a new homeland after being enslaved by the evil Bloats. Each Zoombini has four distinct characteristics—eyes, nose, hair, and feet—and each of these characteristics has five options (for instance, wide eyes, one eye, sleepy eyes, glasses, and sunglasses) for a total of 625 unique Zoombinis. The player must bring bands of these Zoombinis through a series of puzzles in which the puzzles are responsive to the characteristics of the Zoombinis. For instance, in the Allergic Cliffs level, there are two bridges, and each bridge has a set of allergies to certain Zoombini characteristics—like wide eyes or green noses; although the player is given no instructions about these particular bridges, and it is left to the player to discover their properties. If the player tries to move a Zoombini with an allergen across the bridge, the bridge will sneeze and blow the Zoombini back. The game is designed to help players explore ideas about logic, pattern recognition, and combinatorics.20
To classify the game, we can start by asking who controls the pace and pathway of the learning. As a single-player game, there is no peer element, but the game has a blend of features that support both algorithm-guided and instructor-guided elements. Within a puzzle, there is no “Next” or “Previous” button; players try a series of steps, moves, and combinations, and the game levels react to player choices. When players complete a puzzle level, they are exposed to a branching structure to choose the next puzzle that splits autonomy between designers and players. The results are somewhere between a fully sequenced system like a MOOC and a fully adaptive system. While the visual skin of Zoombinis and edX are quite different, they share a set of sequenced learning activities evaluated by autograders.
The Logical Journey of the Zoombinis defies easy categorization in the learning-at-scale schema that I have proposed, but the genres help us to situate the gameplay and learning relative to other learning experiences. Pedagogically, the game is an open-ended immersion into logical thinking exercises, with a strong emphasis on inquiry. Levels have no written instructions, and there are no demonstrations or worked examples; rather, players need to discover the rules of each puzzle through visual contextual clues, trial and error, and logic. Despite wildly different appearances, the technological infrastructure of the game has important commonalities with MOOCs, as a walled garden with an automated assessment system. The pacing and gated objectives have more in common with MOOCs or adaptive tutors than with Scratch or the Rainbow Loom community. Despite these similarities to MOOCs and intelligent tutors, the pedagogy behind the game is more about apprenticeship and play than instruction; it has a closer philosophical kinship with Seymour Papert’s Logo programming language than with Salman Khan’s practice problems.
Zoombinis is a learning experience that cuts against the grains of the patterns that I have described in the past chapters, and it can provide inspiration for new avenues of development. What might it look like to have an automated assessment system in the Scratch platform that maintained the values of playfulness and apprenticeship? What might it look like to have the ideas of playfulness and apprenticeship embedded into an intelligent tutor or an xMOOC? Tinkering with hybridity offer potential ways of offsetting the limitation of one learning-at-scale genre with the strengths from another.
In Part I of this book, I highlighted the differences among large-scale learning environments to define three genres, and in the next four chapters I turn to a set of similarities. In some respect, the technology innovations that I’ve described over the past four chapters are astonishing: free online courses in nearly any subject from anatomy to zoology, a repository of free online videos that cover the entire mathematics curriculum, adaptive tutors that personalize practice for individual students, peer communities where learners gather to study computational creativity or online learning, and games for learning languages or building worlds. As these innovations were introduced, particularly in the late 2000s and early 2010s, they were accompanied by dramatic predictions about how these tools might pave the way for fundamental transformations of educational systems. Yet for all the adoption of large-scale learning technologies in informal learning environments and formal institutions, fundamental transformations have been elusive. New learning-at-scale technologies have proven limited but useful supplements to traditional education systems, rather than levers for fundamentally remaking those systems. Why?
Across very different kinds of large-scale learning systems—inspired by Dewey or Thorndike, found behind paywalls or on the open web, guided by instructors, algorithms, or peer communities—certain dilemmas emerge over and over again. The curse of the familiar starts from the observation that technologies that look like typical elements in schools—like the practice problems on Khan Academy—scale much more easily than things that look very different from anything that has come before, like the open-ended programming environment on Scratch. Schools are complex institutions finely tuned to a kind of homeostasis. And if old ideas are easier to adopt than truly novel, and potentially much more powerful, approaches, how then can schools change and evolve to meet the challenges of the future? From MOOCs to adaptive tutors to Scratch, evidence suggests that an edtech Matthew effect is quite common: that new technologies disproportionately benefit learners with the financial, social, and technical capital to take advantage of new innovations. How might we design learning technologies that ameliorate rather than exacerbate opportunity gaps? I have described technologies as unevenly useful across different subject domains, and one of the core sources of unevenness is assessment technologies. These technologies work much better in domains where problem solving is structured and routine; the trap of routine assessment is that the places where automated assessment works best may overlap closely with the domains where automation and robotics are most likely to replace human work. Throughout each of these first four chapters, I have observed that one of the characteristics of the best learning technology systems is that they are subject to constant research, iteration, and refinement. This kind of research might be greatly aided by the vast stores of data collected by online learning environments, and yet these data also include deeply personal information about people’s lives and learning experiences—how they learn, where they succeed and fail, how they rank compared to others, their interests and beliefs. Navigating this tension requires addressing the toxic power of data and experiments and weighing the privacy risks with the research benefits of data collection. Across very different technologies, pedagogies, and designs, all learning-at-scale initiatives are forced to confront these common challenges that take up the second half of this book.