WHEN THE rhetoric of “personalized learning” seized the education world starting in 2010, I was struck by the diverse constituencies advocating for technology-mediated personalization. The enthusiasm for personalized learning cut across many of the typical partisan divides in the politics of education reform. From pedagogical progressives to free-market reformers, people who agreed about nothing else—charter schools, unions, school boards, direct instruction, national standards—agreed that personalization was (1) going to be enabled by technology and (2) going to improve student learning.
The consensus about the great potential of personalized learning depended on a stark disagreement about what the term actually meant.1 For the advocates of adaptive tutors and blended learning whom we met in Chapter 2, personalization meant that each individual child would be able to spend part or all of her day proceeding through technology-mediated learning experiences at her own pace. For other educators, it wasn’t the pace that should be personalized; it was the content and learning experience. For these educators—usually aligned with John Dewey’s vision of apprenticeship models of education—personalization meant that students would be able to leverage online networks to explore their own interests. Students would identify passions, join online learning communities, study topics of their choosing, and create performances and artifacts of their learning that could be shared online. Learning in schools would look more like the experience of the Rainbow Loomers whom we met in the Introduction.
These twin visions of personalization—personalization as algorithmically optimizing a student’s pathway through established, traditional curriculum and personalization as students choosing topics for study and communities for participation—are not only very different, they are in some sense irreconcilable. It is only possible for adaptive tutors to algorithmically optimize student pathways through content if educators define all of that content in advance and limit assessment to those domains in which computational assessment is tractable. If learners are to be empowered to choose their own topics of study and demonstrate their understanding through different kinds of assessment, then online networks that can support diverse investigations become more essential than adaptive tutors that can accelerate learners through pre-defined content.
Behind these two perspectives on personalization, there is another important distinction in how these camps view the notion of “scale.” For most systems of instructor-guided and algorithm-guided learning at scale, the tutorial is considered the ideal mode of learning, and the goal is to bring the best possible tutorial experience to as many learners as possible. And since human tutors are too expensive, the model uses technology to create something as close to the tutorial ideal as possible. The massive scale of human-learning needs is a problem in this model; scale is a hurdle to be overcome through technology.
An alternative vision sees scale not as a hurdle, but as a crucial resource for creating powerful learning experiences. Scale means knitting together a community of learners from across the networked world, leveraging their interests, talents, and inclinations to teach and share. In this chapter, we will focus on examples from the peer-guided genre of learning at scale, where learning designers and instructional leaders are intentional about weaving networked learning environments into formal educational institutions. In the peer-guided genre of learning at scale, a learner’s progress through an experience is decided not by an instructor or an algorithm but by the learners themselves, who navigate a network of learning experiences generated by a community of peers and curated by a set of designers and instructional leaders.
Whereas much that happens in MOOCs and adaptive tutors feels very familiar to anyone who has spent time in traditional schools and colleges, many designs in the peer-guided genre of learning at scale can appear novel or foreign to both learners and educators. One place to begin exploring the opportunities and challenges presented by this new approach is with the original learning experiences that called themselves MOOCs but differed in form and philosophy from the instructor-guided MOOCs discussed in Chapter 1. These connectivist MOOCs, or cMOOCs, formed primarily in Canada several years before Thrun and Norvig’s Introduction to Artificial Intelligence course started the MOOC phenomenon in elite higher education.
The term massive open online course was coined in 2008 by David Cormier, an instructional technologist on Prince Edward Island in Canada, to describe a new kind of online course that a handful of educational technologists were experimenting with. These early MOOCs had a few thousand participants whose primary learning activity was engaging in conversations over social media, and they differed dramatically from what Coursera and edX would create four years later. One of the first MOOCs was called Connectivism and Connected Knowledge, taught in 2008, and known by its social media hashtag #CCK08. It was offered for credit to twenty-four students at Manitoba University, but through its open design, over 2,200 students participated in the course in some way.2
The form of the CCK08 learning experience was influenced heavily by its subject matter, the epistemology of connectivism. Two Canadian instructional technologists, George Siemens and Stephen Downes, were the principal architects of the theory, which argues that knowledge exists in networks. At the biological level, this means that knowledge exists in the networked structure of the brain; at the sociological level, knowledge exists within communities of people and practitioners. This epistemological position on the nature of knowledge—it primarily exists in networks—led naturally to a pedagogical position: the way to increase knowledge is to generate richer, denser networks. In this model, the best learning happens when learners connect with other people and resources that support ongoing inquiry.3
In the original connectivist MOOCs, the home base for a course was a publicly accessible site on the open web—no logins, no paywalls. This home base offered shared content, guidelines, and instructions for students. Instructors encouraged students to create their own online web presence, typically by creating individual blogs and social media accounts. Learners came to the home base to find shared texts (reading assignments) and prompts for discussion and interaction, and then they responded to those prompts on their individual blogs and social media accounts that were networked with other students on social media. To organize this cacophony of activity, instructors developed techniques that came to be called syndication.4
One of the simplest syndication techniques was using a course hashtag. Students could write a blogpost on their own blog and then tweet the link using the #CCK08 hashtag so that other people following the course could find it. More sophisticated syndication techniques used RSS, or real simple syndication, which was one of the gems of the open web that has been marginalized by the growth of walled-garden platforms like Facebook. Stephen Downes developed an RSS software toolkit called gRSShopper, which allowed students to register their individual blogs and other content sources, and then gRSShopper would make a copy of each submission and aggregate it elsewhere in a variety of forms. gRSShopper automatically published a daily digest of all submissions, and it also allowed the instructors to easily curate a few highlights from each week in a course.5
While instructors played an important role in shaping the direction, membership, and cadence of activity within these communities, the learning experience of each individual student in the course was dramatically shaped by the student’s peer network. Siemens and Downes argued that it was the discussions and connections in the network, rather than the instructor-selected content in the home base, that defined the learning experience. Stephen Downes wrote that the “content is a MacGuffin,” the narrative trick in a movie that brings people together. For an educator, this is a provocative stance: that the content of a course is a kind of trick designed to bring people together into conversation, and it is through this conversation—rather than through direct instruction—that the learning happens.6
The primary learning activities in the Connectivism and Connected Knowledge course were reading and commenting on other people’s thoughts via blog posts, Twitter threads, and other forms of social media, and then responding with a student’s own posts and perspectives. Among the most important learning tasks were connecting with other people—adding new people to follow on Twitter, bookmarking blogs, and adding others’ RSS feeds. Successful learners used technology to create a learning community.
To those steeped in pedagogical theory, the approach of Siemens and Downes had much in common with what Jean Lave and Etienne Wenger called “situated learning.” Lave and Wegner studied vocational communities and how apprentices in those communities developed their expertise. They argued that a central part of apprenticeship was a mode of interaction called “legitimate peripheral participation.” Legitimate peripheral participation is when a novice hangs out on the edge of a community of experts, looking for opportunities to move from the edge toward the middle—a kid hangs around the auto repair shop, watching the mechanics at work, until one day, a mechanic asks him to hold a bolt in place for a minute, and the next week he’s asked to actually tighten the bolt, then he’s hired a few hours a week, and from there, the journey commences.
Stephen Downes argued that what makes someone a physicist is only in part her knowledge of the facts and formulas of physics. Even more important to becoming a physicist is having colleagues who are physicists, knowing the current debates in physics, and becoming inculcated in a physics community. Situated learning and connectivism are pedagogical approaches that are attentive to the social and cultural dimensions of learning. In particular, they encourage designs that let people move from the periphery to the core of a learning experience or learning community. In the twenty-first century, those communities are often defined by their online networked connections. The technological scaffolding of connectivism—blog posts, Twitter hashtags, and other open web technology practices—were novel and attuned to a moment when social networking was transforming society, but as with so many things in education and education technology, it built upon ideas and practices that had come before.7
In the connectivist vision for peer-guided learning at scale, learners need to develop a variety of online learning skills as a precursor to learning about particular topics or subjects. Learners need to be able to set up blogs and social media accounts, use social networking features such as following accounts or feeds, and navigate a decentralized web of resources and people. Engaging in these processes just to access the learning experience is much more complex than figuring out how to click “Next” in a MOOC or how to submit an answer in an adaptive tutor and wait for the next problem to appear. Just getting started in a cMOOC required that learners develop a whole set of new skills for participating in online learning. While most cMOOCs addressed these needs through online tutorials or peer mentoring sessions, a few places experimented with building institutional infrastructure to help students develop these skills.
The most ambitious efforts to have students develop the technical fluency needed to participate in connectivist-inspired learning communities were centered at the University of Mary Washington, where a team of innovative instructional technologists tried to reimagine digital learning infrastructure in higher education. One of the leaders of this effort was Jim Groom, who in 2008 defined the term edupunk to describe a way of relating to education technology that rejected corporate solutions, especially learning management systems, and promoted student ownership of the means of technological production. In 2010, Groom helped develop the online course Digital Storytelling, or DS106, a computer science course with a goal of helping students develop skills related to media production, web development, and storytelling online. Like the Canadian cMOOCs, DS106 developed an open online component that let other universities participate in the course (parallel sessions have been offered at the University of Michigan, Kansas State, and several other colleges) and individual learners from the web join in.8
As in other connectivist, peer-guided learning experiences, the home page of DS106 serves as a guide, a syllabus, and an aggregator, syndicating the feeds of blogs, Flickr, YouTube, and other accounts from learners around the world. There is a DS106 online radio station and a livestream video station for media projects. Perhaps the most distinctive feature of the course is the Daily Create—a challenge to make media in twenty minutes or less every day. The Daily Creates are inspired by materials out of the Assignment Bank, a repository of media creation prompts (“Make a video where you tell the stories of the keys on your keychain”). These assignments have been submitted over nearly a decade by instructors, enrolled students, and passersby. Through these kinds of assignments, students developed the skills in media production, web hosting, and social networking to be able to participate in peer-guided large-scale learning communities. If CCK08 was a cMOOC about the ideas animating cMOOCs, then DS106 was a cMOOC about the technical skills required to participate in cMOOCs.9
Having every student who was enrolled in DS106 create his or her own blog through a commercial provider was a logistical barrier to student participation, so Groom and colleagues developed their own blog-hosting solution for the school. This eventually turned into a Domain of One’s Own, a project to give every freshman at University of Mary Washington his or her own server space and online presence. Much in the same way that learning management systems provided institutional infrastructure for teacher-directed instructionist learning, the Domain of One’s Own project attempted to create an institutional infrastructure for connectivist learning. Other universities took an interest in the Domain of One’s Own approach, and Groom left the University of Mary Washington to start Reclaim Hosting. To reclaim universities’ web presence from learning management systems and to reclaim the web from centralized commercial interests more broadly, Reclaim Hosting offered a turnkey solution for universities to create their own Domain of One’s Own projects. A number of universities—University of Oklahoma, Drew University, Brigham Young University, and others—tested using Reclaim Hosting to make a student-controlled online space a central part of their information technology infrastructure.10
Groom and colleagues realized that implementing their student-centered, peer-guided vision would require not just making a new site or app, but also developing an entirely new technology infrastructure for supporting higher education. Few universities to date have taken this path, but Reclaim Hosting maintains an alternate, edupunk, indieweb approach that, like a global seed vault, stores possibilities for alternative futures.11
For those who had the technical prowess to generate and navigate content on the open web and the time to invest in navigating these communities, the connectivist MOOCs were powerful learning experiences. Participants explored new ideas, developed new technical skills, and perhaps most importantly, developed a set of relationships and connections that in some cases long outlived their original course communities. One could head onto Twitter a decade after CCK08 and still find participants occasionally posting on the #CCK08 hashtag. But despite the passion that cMOOCs evoked among enthusiasts, they never expanded much further than a few hothouses of fertile experimentation.
For a time, it looked like both connectivist MOOCs and instructionist MOOCs might coexist side by side in an online learning ecosystem, and commentators came up with the terms cMOOCs and xMOOCs to distinguish the Canadian open web experiences from the increasingly paywalled, linear learning experiences offered primarily by elite universities. Researchers conducted comparative studies of both approaches, and the xMOOC-mania of 2012 gave rise to a small surge of renewed interest in and attention to cMOOCs. In 2012, as a response to public attention on instructionist MOOCs, a pair of women’s studies professors, Anne Balsamo and Alexandra Juhasz, came together to create a distributed open collaborative course called FemTechNet around feminist dialogues in technology. The course was designed much like CCK08 or DS106, with a series of online resources to support small local “nodal” classes facilitated by local instructors with ideas and feedback permeating back to the core. If Downes and Siemens framed their project as pedagogical, FemTechNet was more explicitly political, contesting not just the instructionist pedagogy of xMOOCs but also their hegemonic model, where elite universities sent their digital emissaries to the far corners of the world to instruct rather than to listen and share. In these efforts, cMOOCs were not just an alternative to xMOOCs but a critique of them.12
Connectivist MOOCs blossomed at a peculiar historical moment in the history of the web, in the pivot point between a dramatic increase in the number of people creating content online and the capture of all that activity by a handful of proprietary platforms. The peer-guided cMOOCs were made possible by a series of new technologies called Web 2.0—WYSIWYG (“what you see is what you get”) web editors that let people create web content without HTML or CSS, and hosting solutions for blogs and websites that let people upload images and other files without using file transfer protocol (FTP) services. In 2008, when CCK08 started, people on the web created their own blogs through Blogger and Wordpress, hosted and shared their pictures on Flikr, and read through news and blog feeds through Google Reader. In the years that followed, the largest technology companies were successful at integrating all of those different features into their “walled-garden” platforms. People could share thoughts, host images, and scroll through a news feed all in one place on Facebook, LinkedIn, or Snapchat. Advocates of the indieweb have lamented the concentration of power within a few platforms, and the loss of the richness that emerged from the multiple, distributed voices on the open web. But for most users, the integrated experience of posting ideas, connecting with people, and reading content in a single, tidy, walled garden was simpler and more compelling than the additional efforts required to maintain and participate in the open web.
The simpler, centralized, linear approach won out among MOOCs as well. In the same way that Facebook came to define what it meant to “go online,” the edX and Coursera xMOOC cemented what it meant to be a MOOC in the public consciousness. Courses and platforms adopting the xMOOC model accumulated the overwhelming majority of registrations in large-scale learning experiences. Learners consistently found cMOOCs confusing and difficult to navigate, like wandering through a corn maze rather than the tended linear paths of xMOOCs There are still a few cMOOCs offered every year, mostly to other educators, but the connectivst experiment in higher education increasingly appears to be the road not taken, or perhaps a road to be reclaimed.13
While peer-guided approaches to learning at scale have generally foundered in schools, there is one striking exception: the Scratch programming language and online community that has been translated into over sixty languages and widely adopted by schools and systems around the world.
Scratch is a block-based programming language, which means that rather than learning to write software code through syntax (print:“Hello World”), people learn to write code by snapping together digital blocks, each of which represents a function, variable, or other programming element. Scratch was developed by Mitch Resnick, Natalie Rusk, and their team at the Lifelong Kindergarten lab at MIT. Scratch has integrated graphics editing, and when combined with the programming language, it is a powerful platform for making animations, games, and other visually appealing programs.14
From the beginning, Scratch was imagined as a creative learning community of users, called Scratchers. Each Scratch program contributed by a user is automatically made available for inspection and remixing—starting a new project with a copy of another project—by anyone else on the site. In his recent book, Lifelong Kindergarten, Mitch Resnick describes the Scratch platform as the intersection of four alliterative learning dimensions: projects, passion, peers, and play. There are a few tutorials on the site and some exemplar projects created by staff, but the instructional approach of Scratch leans heavily on learners sharing examples of their projects with the community. The home page hosts a collection of curated examples, sometimes from project staff and sometimes from community members, along with a few algorithmically curated sets of projects based on what’s currently being “loved” by the community or remixed. Every project page includes space for the author to post instructions, notes, and credits, along with a comment thread where other Scratchers can offer feedback, ask questions, and interact with the project author. Scratchers also communicate about their projects in the very active forums, where they ask for suggestions, share tips and tricks, and discuss projects with the community as a whole.15
For Resnick and the Scratch team, the whole point of the learning environment is to empower young people to explore their passions through creativity and design. There is no right way to program in Scratch or right pathway to learning how to program, so the site generally stays away from the kinds of linear instruction provided by MOOCs or adaptive tutoring systems. Resnick and his collaborators were influenced by the ideas of Seymour Papert and Cynthia Solomon, who codeveloped the Logo programming language that many of us from Generation X used when we were in elementary school. Papert argued that programming environments for young people should offer “low floors and high ceilings”; it should be easy to get started programming in the environment, but still possible to create sophisticated programs. To this, Resnick adds the idea of wide walls; community members should be able to create a wide variety of projects, based on their interests and passions, with different themes and purposes.16
Learning in the Scratch community looks much like learning in the connectivist MOOCs. Scratchers use the features of the platform to develop their own identity, make connections with people and resources, and develop their skills along the lines of their interests. The learning environment supports a wide variety of activities and levels of participation. While tens of millions of registered users have made over 40 million projects on the Scratch platform (as of 2019), many of these learners pass through the system relatively quickly, lurking on a few projects or starting one or a few simple ones. A small percentage of Scratchers get seriously into creating, remixing, and commenting on projects, and a very small number of users become leaders in the community, creating tutorials, moderating forums, editing the wiki, curating collections, and so forth. People choose how they want to participate, and they do so to varying degrees.17
Just as connectivism provides a useful framework for explaining the learning designs in cMOOCs and higher education, the theory of connected learning, developed by cultural anthropologist Mizuko Ito and colleagues, provides a lens for understanding networked learning among younger learners. Connected learning is interest-driven and peer-supported, but crucially, it also provides opportunities for academic connections; connected learning is realized “when a young person is able to pursue a personal interest or passion with the support of friends and caring adults, and is in turn able to link this learning and interest to academic achievement, career success or civic engagement.”18 Realizing connected learning in the Scratch community means letting kids develop new skills to create projects that reflect their interests, but also supporting young people in seeing how creating animations and games or learning how to program can connect to other academic pursuits in school.
In a 2014 TEDx talk, Resnick tells the story of a boy who was using Scratch to create a video game. Seeing that the boy would need to include mathematical variables in his program if he were to realize his vision, Resnick explained how to encode variables in Scratch programming blocks. Resnick described the moment when the concept of a variable clicked for the boy: “He reached his hand out to me, and he said, ‘Thank you, thank you, thank you.’ And what went through my mind was, ‘How often is it that teachers are thanked by their students for teaching them variables?’ ” By placing the concept of variables in the context of a meaningful project, Resnick and Scratch helped the boy connect his interests in gaming with at least one crucial algebraic concept.19
Scratch was developed out of Resnick and Rusk’s work in Computer Clubhouses, a network of afterschool programs around the world for exploring computational creativity, and in the early years, Scratch was primarily used by individual kids and informal learning programs. More recently, Scratch adoption has taken off in K–12 schools as a way of introducing computing and computer programming, but not always in the ways that the Scratch team had intended. As Resnick noted, “Over the past decade, we’ve found that it’s much easier to spread the technology of Scratch than the educational ideas underlying it.” In Resnick’s TEDx story, a boy starts a project connected to his own interest in video games, and when he encounters a particular challenge, Resnick steps in with some just-in-time learning to help the boy develop the skills and knowledge needed to advance the project. By contrast, Resnick and his colleagues in the Lifelong Kindergarten lab have countless stories of schools where teachers introduce students to the Scratch programming language through teacher-structured activities rather than through open-ended exploration. It is very common for teachers introducing Scratch to create their own model program and then ask students to create a replication of that model, sometimes even by requiring that students reproduce step by step a project that a teacher projects on a screen. Many of the core expectations of schools—that students produce their work independently, that all students complete a project in a similar amount of time, that all students study topics regardless of their interest level—conspire against a pedagogy that seeks to empower students as leaders of their own creativity and learning.20
Having examined all three of the learning-at-scale genres of instructor-guided, algorithm-guided, and peer-guided large-scale learning environments, we are now better equipped to compare them.
People with very different pedagogical proclivities often have surprisingly similar views about the end goals of learning. When sociologist Jal Mehta of Harvard University and educator Sarah Fine of High Tech High Graduate School of Education studied dozens of highly lauded high schools in the United States that approached instruction quite differently, they found that many of them pursued a similar vision of “deeper learning,” a set of interrelated competencies that include traditional disciplinary knowledge as well as the skills of communication, collaboration, problem solving, and self-regulation. Mehta and Fine characterize deeper learning as the intersection between three important learning outcomes: mastery, identity, and creativity. When students experience deeper learning, they develop mastery of deep content knowledge in a domain. They also experience a shift in identity where the learning activity is a part of who they are rather than something that they do—the shift from “learning to swim” to “being a swimmer.” They also have opportunities to create novel, authentic, interesting new projects and performances with their new skills and knowledge.21
Sal Khan of Khan Academy and Mitch Resnick and Natalie Rusk of Scratch would all agree that students should learn math in order to create wonderful things in the world. I suspect they would agree that great mathematicians and great computer programmers have a deep understanding of content knowledge in the domain, develop an identity around their practice, and show true mastery not by replicating what has been done before but by creating things that are new. But if they do not differ in ends, they differ dramatically in how they believe learners should make progress toward these ends.
Traditional instructionist educators believe that content mastery is a necessary precursor to shifts in identity and opportunities for creativity. They rightly observe that people who do the most novel and important creative work in a field tend to have extensive mastery of the domain knowledge in that field, so they start their teaching by focusing on knowledge mastery and hope that content mastery will provoke shifts in identity and that mastery can then lead to creative output. By contrast, social constructivists observe that most motivation for learning comes from opportunities to be creative. As people play with Scratch, they become Scratchers, and the opportunities for creativity unlock their passion for learning about programming and mathematics to make ever more intricate programs and creations.
In Khan Academy, the proper first step toward deeper learning is learning mathematical procedures and facts that might eventually lead to doing interesting collaborative projects. In Scratch, the first step in that journey is getting people to play with tools for computational creativity that will inspire learners to understand how variables and other mathematical concepts can enrich their creations. I think there is room for both models in our education systems. But given that our formal educational systems are overwhelmingly organized around the mastery-first models, I’m enthusiastic about approaches that create more opportunity for creativity-first and identity-first learning in schools and colleges.
The divergent pedagogical beliefs of traditionalists and progressives about idealized pathways to learning lead to differences in how each camp conducts research about their large-scale learning environments.
The research concerning the efficacy of adaptive tutors that we explored in Chapter 2 had a set of shared assumptions, and those shared assumptions were part of what made meta-analyses and comparisons across multiple studies possible. Adaptive-tutor research assumed that instructors and designers identified a body of content knowledge that students should learn, and that teachers and the system as a whole should be judged on the basis of how much progress all learners made toward content mastery. Usually, progress toward that goal was measured by the change in the average proficiency of learners before an intervention (such as the adoption of adaptive tutors in classrooms) and the average proficiency afterward. This research assumed that the goal of education is for every student to make progress toward mastery and that the bell-shaped distribution of student competence, measured in effect sizes, should shift to the right over time.
These assumptions about essential features of a learning environment are not shared by the researchers in Resnick’s Lifelong Kindergarten group or by the designers of connectivist MOOCs. If educators take seriously the idea that learning ought to be driven by the interests of students, then when students decide that Scratch is not really interesting to them but that something else is, then it’s no loss to the Scratch designers to see those students move on to other things. For supporters of peer-supported, interest-driven learning, the concerns are less about whether an entire class of learners is developing new capacities in one subject, but rather whether the subset of learners who are really interested in and devoted to a learning experience can steadily improve their skills and that the environment can successfully invite in new community members who share their interests.22
Part of the challenge of measuring peer-driven learning is that in many of these learning environments, goals are determined by individual students and the networked community, not by teachers or evaluators. If, as Downes says, the content is a MacGuffin, then what would be reasonable measures of learning in that environment? If a participant in #CCK08 knows little or nothing about connectivism but has built a network of new peers and colleagues, is that a successful course? These problems plague other learning-at-scale environments as well—for instance, some xMOOC students are more interested in learning and practicing English than the particulars of course content—but the challenges posed by the multiple aims and goals of learners in peer-guided large-scale learning environments make summarizing their effectiveness particularly challenging.
Because the goals of a peer-guided educational environment are different from those of an instructionist one, so too are the research methods used to evaluate them. Researchers in the Lifelong Kindergarten lab have largely studied the Scratch community through intensive qualitative research—thick descriptions of the lives and practices of individual Scratchers. Much of this research tends to focus on individuals who have powerful learning experiences online and share that learning with others. In a sense, the purpose of Scratch is to create the conditions for these deeply invested learners to thrive, while also allowing other learners at the periphery to participate at less intense levels. Having a learner leave the Scratch community or pass through with only a light touch isn’t necessarily a loss or a concern.23
The goals and research methods for the different types of learning environments then interact with the pedagogies and instructional designs of developers in mutually reinforcing ways. The outlook shapes the questions that designers and researchers ask, the methods to conduct research, and the answers to research questions—and the answers to those questions then feed back into the iterative design of these large-scale learning systems.
A group of advocates for the merits of traditional instruction wrote one of my favorite papers on education, provocatively titled “Why Minimal Guidance during Instruction Does Not Work: An Analysis of the Failure of Constructivist, Discovery, Problem-based, Experiential, and Inquiry Teaching.” They argue that over and over again, experimental studies that contrast traditional methods of direct instruction with minimally guided, open-ended learning demonstrate that for a wide variety of learning outcomes, direct instruction works better. They draw on a set of ideas from cognitive science called cognitive load theory to explain why this is the case. Put simply, people have a limited working memory, and when learners allocate that working memory to solving a problem, they often are not permanently encoding learning about the patterns and practices that let them solve that type of problem. It is more efficient and effective to have an instructor demonstrate through worked examples how to solve an individual problem and less efficient and effective to have students try to discover solutions and patterns from problems without much instruction.24
It is crucial to understand, however, what these authors mean by “successful learning.” For these critics of minimally guided, peer-led instruction, a learning environment that works is one that helps shift a bell-curve-shaped distribution of learners toward higher levels of mastery; indeed, the statistical models they use to test their interventions require as an assumption that there is a measurable skill that is normally distributed across the population of learners. That is one useful definition of a learning intervention that “works,” but it’s not the only one. The Scratch learning community is a powerful refutation of the argument that minimally guided instruction does not work. Scratch does everything wrong according to the advocates for traditional instruction—there is almost no direct instruction, there are very few formal assessment mechanisms, there is no assigned sequence of learning activities, there is no attention given to managing learner cognitive load, there are no experimental tests of interventions, and nearly every change to the Scratch platform is evaluated on the basis of qualitative case studies and user observations rather than randomized control trials. And yet Resnick and his team have built one of the most widely adopted engines of creative learning in the world.
When something “works” for Resnick and the Lifelong Kindergarten team, it allows individuals to explore their passions, publish authentic performances of understanding to the world, and develop deep mastery. They have tuned the Scratch learning environment to allow for widespread participation, but they have also ensured that accommodating widespread participation doesn’t place undue restrictions on the individual pathways of the most devoted learners. As a result, Scratch “works” brilliantly, in the sense that millions of students are introduced to block-based programming through the system, and a subset of those young people develop remarkably deep understandings of block-based programming, creative digital expression, and computational thinking through the system.
I have a vigorous commitment to methodological pluralism; I think both of these approaches to learning are necessary and can lead to great outcomes for learners. Our society needs instructional systems that address both kinds of aims. We need our entire population to have fundamental skills in reading, writing, numeracy, civics, science literacy, and communication; in these domains, we need to take the entire distribution of learners and help them move toward mastery. We also need learning environments that let young people discover their interests and explore them deeply, much more deeply than might be allowed if the environment were equally concerned with bringing along the unenthusiastic with the enthusiastic. We need learning environments that shift whole distributions to the right, and we need learning environments that enable deep learning for a self-selected few.
That said, one of the challenges in understanding peer-guided learning environments is that the research defies easy summarization. For peer-guided learning environments, we know that some learners become deeply immersed in these learning environments and can develop very high levels of proficiency, but we have less understanding of what learning looks like across the whole distribution of people who engage.
Pedagogies come bundled with philosophies and moralities. For instance, advocates of instructionist approaches tend to emphasize that learning is difficult and results from struggle; advocates of progressive pedagogies tend to emphasize that learning is natural and easy. Seymour Papert argued that just as people learned the French language naturally from living in France, so too the Logo programming language could become a “Mathland” where people naturally and easily learn math. In Jean-Jacques Rousseau’s novel Emile, or On Education, the protagonist’s education emphasizes exploration and observation in the natural world over formal study, and Rousseau associates this naturalistic approach with preparation for a more democratic society that would transcend the feudal and monarchical structures of eighteenth-century Europe. An assumption that cuts through these older ideas and contemporary approaches to peer-guided learning is that if young people are given the opportunity to explore their interests and passions, then they will generally choose interests and passions that are enriching and interesting.
Throughout this chapter and this book, I’ve celebrated peer-guided learning at scale, not as the best learning methodology for all learning, but as an approach to empowering learners that provides an important counterbalance to the instructionism that dominates schools, colleges, and formal learning institutions. I have also celebrated the peer-led, informal learning that happens among Rainbow Loomers and enthusiasts of all kinds. Part of my enthusiasm for peer-guided learning at scale is rooted in a general optimism about learning and humanity that Papert and Rousseau shared: given the freedom and resources to learn, people will generally choose to learn about worthwhile things.
Unfortunately, peer-guided learning environments can be used to recruit people into dark and hateful ideologies in much the same way that people can learn Rainbow Loom or digital storytelling. In 2018, there was a terrible incident where a man in his mid-twenties rented a van and proceeded to drive it down a busy Toronto sidewalk, killing ten and injuring fifteen. As police and others explored the online history of the murderer, they found that he had participated in an online community of men that call themselves incels, or involuntary celibates. These are men who gather on subreddits and the troll-filled message board of 4chan to lament their inability to persuade women to have sexual relationships with them. Incels promulgate a worldview of male radicalization according to which all men are owed sex from women, but women only provide these opportunities to men from certain social strata. The ideology is a mixture of resentment and madness, and periodically it explodes into violence. The Toronto van murder was inspired by a similar incel mass murder in California some years earlier.25
Even the introduction here of the term incel in explaining this particular murder is in itself a potentially dangerous act. The word incel functions as a potential aggregator for male radicalization in the same way that “starburst bracelet” might operate for a Rainbow Loom enthusiast or “#CCK08” works for a cMOOC enthusiast: it is a search term that can bring people into a broader online community interested in educating new members. In an October 2018 talk by danah boyd, founder of Data and Society, boyd excoriated members of the media for broadcasting the term incel in the wake of the Montreal murder:26
I understand that the term “incel” was provocative and would excite your readers to learn more, but were those of you who propagated this term intending to open a portal to hell? What made amplifying this term newsworthy? You could’ve conveyed the same information without giving people a search term that served as a recruiting vehicle for those propagating toxic masculinity. Choosing not to amplify hateful recruiting terms is not censorship. You wouldn’t give your readers a phone number to join the KKK, so why give them a digital calling card?
When boyd accuses journalists of helping their readers “learn more” about incels, it is useful to realize that incels and other advocates of male radicalism have built a sophisticated online learning environment on the open web—hidden in plain sight—that has much in common with connectivist-inspired learning environments: it is a distributed network of people and resources that seek to invite new members to join them in a community in which they will learn new ideas, knowledge, and skills. Those seeking to inculcate new—primarily young—men into male radicalism post a range of materials online, from mainstream critiques of political correctness to targeted social campaigns like Gamergate to extremist forums, and these communities strategically guide people along these paths toward extremism.27
There is mounting evidence that some of the architectural features of online networks can, unfortunately, be more powerful in amplifying extremist messages than more moderate messages. The video-hosting platform YouTube illustrates this phenomenon best, but the features of the YouTube recommendation engine can be seen in other algorithmic recommendation engines as well. In 2018, sociologist Zeynep Tufekci observed that in a variety of situations, after a viewer watches a YouTube video, YouTube will recommend another video with content more intense, extreme, and disturbing than the last. After watching videos in support of Donald Trump, she observed recommendations for videos about white supremacy or Holocaust denial. After watching Bernie Sanders videos, she got recommendations for videos claiming that 9 / 11 was an inside government job. Watch videos about vegetarianism, and you’ll get recommendations for veganism. Watch videos about jogging, and you’ll get recommendations for videos about ultramarathons. As Tufekci argues, “It seems you are never ‘hard core’ enough for YouTube’s recommendation engine.”28
Of course, in learning, “getting more hard core” is often quite wonderful, as when a Rainbow Loomer graduates from a simple design to a more complex one. When a casual watcher of Rainbow Loom videos starts posting comments and then making her own videos, we can celebrate the process of legitimate peripheral participation, moving from the periphery of a community to a core. Online educators and policymakers, however, must come to understand that the bad guys have learned these educational techniques as well. Anti-vaccine conspiracy theorists understand that videos with open-ended questions about vaccine safety can draw in new “anti-vaccine learners” and that those videos, comment threads, and recommendations can be used to move people toward more hard-core anti-government conspiracies.
The proliferation of online communities organized around hate groups or the kinds of conspiracies that Richard Hofstadter called the “paranoid style in American politics” reveals one of the virtues of the centralized learning experiences provided by publishers of adaptive tutoring systems or university providers of MOOCs. These traditional institutions provide an editorial filter. This filter is imperfect, and elite educational consensus can countenance truly terrible ideas (such as the dark history of segregated schooling). For all these flaws, though, it would be virtually impossible for Coursera to host a MOOC that indoctrinated learners in explicitly white nationalist ideology or for Carnegie Learning to produce an intelligent tutor on the physics of a flat earth. And if they did, there are various watchdogs and other methods to police such transgressions. When Walter Lewin, a physics professor whose extraordinarily popular lectures were available online, was found to have sexually harassed women in online course contexts, his lectures and MITx courses were removed.29
In the introduction to this book, I argued that new technologies make this the greatest time in history to be a learner. In this chapter, I have enthusiastically endorsed approaches to peer-guided learning that give learners agency and help them learn to navigate online networks of peers and resources to develop new skills and knowledge. But participating in vast network of online learning resources doesn’t guarantee that people will inevitably learn ideas and skills that will bring about better individuals and a better society. Our technologies and learning resources are shaped by the broader culture, and political battles about whether we have a culture of dignity, respect, and inclusion or a culture of divisiveness and tribalism will determine whether or not our extraordinary infrastructures for learning will, in fact, lead to a better, more just world.
Participation in open-web, peer-guided learning environments is ubiquitous. People all over the world have hopped online to learn how to use a certain block in Minecraft, how to debug a software problem, how to cook an apple pie, or how to crochet a dragon out of rubber bands. Quantifying the exact scope of this learning is difficult, but when millions of people have created Scratch accounts and YouTube videos like “How to Make a Rainbow Loom Starburst Bracelet” have tens of millions of views and hundreds of comments, it seems clear that the global community of online learners is massive.
One of the complexities of peer-guided learning environments is that participants can find them both completely intuitive and utterly baffling. With instructor-guided and algorithm-guided learning environments, students find it easy to use the system but may not always be motivated to do so. People get bored going through xMOOCs and cognitive tutors, and they quit, but it’s less common that they are so confused that they don’t know how to participate. By contrast, many cMOOC participants find these networked learning environments overwhelming, and students “getting stuck”—not knowing what to do next to advance their learning—is a common challenge in classrooms adopting Scratch. One of the signature design challenges of peer-guided learning environments is to figure out how to make them more accessible to novices without turning them into instructor-led learning environments.30
Peering into these kinds of mysteries reminds us that learning and teaching remain, after millennia of practice and study, unfathomably complex. Somehow, millions of people around the world find ways to teach and learn with one another online without any formal training and sometimes without any organization, and yet when designers try to create these kinds of environments with intention, they encounter substantial challenges with motivation and comprehension. It’s frustrating how far we are from understanding how best to create online learning communities where people support one another’s learning in powerful ways, but inspiring to see how despite our limited understanding, people make progress anyway.
And all of the challenges of designing powerful, accessible, and equitable large-scale learning environments are magnified by the additional challenge of integrating these environments in formal education systems. Schools and colleges are tasked with doing more than just helping individual people learn whatever they want. Formal education systems mandate that all students learn certain fundamentals, whether or not they have an intrinsic inclination to do so. Teachers evaluate learners in part to provide feedback, but also so that learners can be ranked, sorted, and tracked into different parts of the educational system. While there may be opportunities for collaboration and peer learning, students in schools are expected to tackle many of their most consequential assessments alone so that their individual competency can be measured. These expectations for teaching, assessing, ranking, and sorting individuals create an inhospitable institutional climate for peer-guided learning environments to take root.
Forward-looking schools, therefore, face a challenging dilemma. Peer-guided networked learning environments will be central to how people, young and old, learn across their lifetimes. In some professions, participation in these kinds of networks will be essential to professional advancement. For instance, computer programming languages advance so quickly that it is almost impossible to be a successful computer programmer without participating in networked learning communities, such as Stack Overflow, where people ask and answer questions about specific programming languages or coding approaches.31 Formal education systems need to teach students how to engage with and learn in this type of open, large-scale, peer-guided network. But the learning practices in these environments grate against some of the key commitments of formal educational systems; they mix like oil and water. A few schools will respond to these challenges by dramatically changing their practices to more closely match the patterns of learning that happen outside of schools, and some schools will simply ignore the changes happening beyond their classroom walls. The most adaptive approach in the near-term is probably for creative educators to find more spaces where peer-guided large-scale learning can be woven into the periphery of schools—in electives, extracurriculars, and untested subjects—so that learners can have some practice in navigating these new networks with a community of local peers and mentors to support them.