Volume 3. Group Cognition: Computer Support for Building Collaborative Knowledge
* Download PDF free for reading online or printing: gc.pdf
(This is the best version: up-to-date, complete, full color, hi-res.)
* Download iBook version free for iPad, Mac, etc.: gc.epub
* Download MOBI version free for Kindle: gc.mobi
* Order paperback from Lulu at printing cost: Lulu page for Gerry Stahl
* Order paperback from Amazon at printing cost: Amazon page for Gerry Stahl
Essays on Technology, Interaction and Cognition
This opening essay establishes the book's central challenge: how can networked computer technology support the collaborative building of knowledge? The author argues that while the internet and distributed computing offer enormous potential for connecting people across space and time, the software and social practices needed to realize this potential have yet to be fully conceived or adopted. The essay identifies three complementary approaches that organize the book: designing software systems that support collaborative interaction, analyzing empirical instances of collaboration in detail, and theorizing the concepts and phenomena involved. Throughout, the author emphasizes that computer support for collaboration is not merely about automating existing forms of individual work, but about enabling new forms of group cognition — collective achievements of understanding that exceed what individual participants could accomplish alone.
(Introduction)
The introduction argues that computational power, distributed networks, and carefully designed software can help small groups transcend the limits of individual cognition and construct forms of shared understanding — shared meanings, collaborative knowledge, and group interpretations — that are not attributable to any single person but emerge from group interaction. At the same time, the author is candid that early attempts to build such systems have encountered significant technical and social barriers. The transition to computer-supported collaborative knowledge building is compared to the historical passage from oral to literate culture: it requires innovations at multiple levels simultaneously and cannot be accomplished by technology alone. The introduction locates the book's inquiry at the intersection of technology design, interaction analysis, and social theory, arguing that all three perspectives are necessary for understanding how small groups build shared knowledge together through digital technology.
(Introduction to Part I: Studies of Technology Design)
This brief introduction to Part I contextualizes eight design studies written across the 1990s. The author traces their theoretical background in two intellectual currents: situated cognition, which challenged traditional artificial intelligence by showing that human knowledge is fundamentally tacit, contextual, and social; and domain-oriented design environments, which proposed building software systems that support domain specialists by integrating task-specific knowledge, critiquing rules, and shared community resources. The author notes that this theoretical framing was always present in the design work, even if the broader theory of group cognition only became explicit later. He reflects from a 2004 vantage point on what each study contributed to the overall inquiry and acknowledges that these studies are exploratory and fragmentary — they pose questions and allow feedback from practice to be experienced, rather than conducting controlled experiments with rigorous conclusions. The introduction prepares the reader for a series of case studies that range from AI-driven educational tools to large-scale community collaboration platforms, united by a common underlying question: what software functionality does collaborative knowledge building require?
Share Globally, Adapt Locally
This chapter describes the design and prototyping of the Teacher's Curriculum Assistant (TCA), a software environment to help classroom teachers find, adapt, and share constructivist instructional materials over the internet. The study emerged from a practical problem: teachers committed to constructivist pedagogical principles had neither the time nor the resources to develop innovative curriculum materials from scratch, and while the internet was beginning to make educational resources available, there were no adequate tools for locating, tailoring, and sharing them effectively. TCA was designed as a domain-oriented design environment for curriculum development, integrating search mechanisms, annotation tools for adapting retrieved materials to local classroom needs, and facilities for uploading revised curricula to shared internet repositories so that other teachers could benefit.
The chapter reflects candidly on the project's fate. It did not advance beyond the prototype phase, partly because funding reviewers could not see how the quality and accuracy of community-contributed content could be guaranteed, and partly because potential commercial partners could not identify a near-term financial return. The author notes that in the decade following this study, large national initiatives and commercial companies attempted similar goals — educational digital libraries, shared content repositories — with only partial success, suggesting that the institutional and social challenges of community curriculum sharing are real and persistent. The study illustrates a theme that will recur throughout Part I: software innovation for collaboration must reckon with the social, institutional, and political contexts in which that software would be adopted. Technical capability alone does not drive adoption.
Evolving a Learning Environment
This chapter describes State the Essence, a computer-based learning environment that used latent semantic analysis (LSA) to provide automated feedback to students composing summaries of assigned texts. LSA is a statistical method that computes a numeric representation of word meanings from the co-occurrence of words in large text corpora; it can compare a student summary to the original source text and estimate how much of the text's semantic content the summary captures, without requiring the computer to actually understand natural language. The chapter follows the two-year evolution of the software through iterative classroom testing in a middle school, showing how each cycle of use by teachers and students revealed inadequacies and prompted adaptations in the algorithms, the interface design, and the pedagogical framing of the task.
The resulting system successfully encouraged students to revise and improve their summaries across multiple drafts, with the LSA-based feedback functioning as a partial substitute for expert teacher response. The chapter is candid about LSA's significant limitations: its performance depends on extensive fine-tuning and is adequate only in carefully constrained applications. The comparison between a student summary and a source text is only one small part of the interpretive competence that genuine understanding involves; LSA captures something about semantic overlap but nothing about the student's situated knowledge, interpretive perspective, or tacit understanding. The study suggests that LSA and similar technologies are useful only when they are carefully integrated into larger educational contexts that account for these limitations. The chapter anticipates the book's later argument that human collaborative understanding cannot be captured by algorithmic processing of linguistic co-occurrence.
Armchair Missions to Mars
This chapter describes CREW, a software application designed to advise NASA planners on the composition of astronaut crews for long-duration space missions. The practical problem was that future missions would involve crews far more diverse than those of the past — mixed gender, multinational, intergenerational — and confined together for months at a time under conditions of isolation and extreme stress. Little empirical data existed to predict how such crews would perform psychologically. CREW combined case-based reasoning (adapting outcome data from analogous past missions in extreme conditions such as Antarctic winter-overs and submarine deployments) with interrupted time series analysis to model how individual psychological factors might evolve and interact over the course of a mission.
The chapter documents the technical ingenuity required to adapt standard case-based reasoning algorithms to this domain, given the scarcity and inconsistency of available data. Novel modifications were necessary to handle the complexity of comparing past cases to projected future scenarios with many unknown parameters. The study ultimately reveals the fundamental limits of AI approaches when applied to group formation under conditions of data scarcity. Without sufficient high-quality empirical data, case-based systems cannot make reliable predictions; and collecting adequate data about crew psychology in isolated extreme environments is itself practically and politically constrained. The chapter illustrates a general argument about AI: without adequate data, computational power produces nothing reliable. It also introduces the problem of group formation — selecting the right people to work together — as an important and underexplored challenge for computer support of collaboration.
Supporting Situated Interpretation
This chapter presents Hermes, a prototype software system designed to support the interpretive work of engineers designing lunar habitats, developed as part of the author's dissertation research on CSCW support for NASA. The chapter argues that innovative design requires making tacit knowledge explicit through interpretation: designers draw on background understanding of their situation, view the same problem from different professional perspectives, and use shared vocabulary to articulate what they tacitly know. These claims are grounded in the design methodology literature of Alexander, Rittel, and Schön; in a protocol analysis of an actual lunar habitat design session; and in Heidegger's philosophy of interpretation.
Hermes was designed to represent the design situation, to organize domain knowledge according to multiple professional perspectives (structural, thermal, ergonomic, and so on), and to provide a shared terminological framework for the design community. The system allowed designers to view a developing design from their own professional standpoint while making their interpretive perspective visible to colleagues with different concerns. The chapter introduces concepts — interpretation, perspective, tacit knowledge, shared language — that recur in each subsequent part of the book. It also introduces the hermeneutic philosophical tradition as a resource for understanding collaboration: the insight that understanding is always interpretive, always perspectival, and always conditioned by a prior familiarity with the context of one's work. This distinguishes the genuinely human capacities that software must support from the explicit symbol manipulation that computers can perform.
Collaboration Technology for Communities
This chapter traces three stages in the evolution of software for supporting collaborative knowledge building in communities of practice. NetSuite illustrates the domain-oriented design environment approach applied to individual designers: a commercial software product that delivers relevant domain knowledge to designers at the moment it is needed for their current task. WebNet extends this model to small groups, providing a shared information space where community members can contribute and access collectively relevant knowledge. WebGuide goes further still, incorporating a computational perspectives mechanism designed to support the collaborative construction of shared understanding within groups through structured, perspectival access to a common knowledge space.
The chapter argues that as the need shifts from supporting individual specialists to supporting communities of collaborative learners and workers, the locus of intelligence must shift from the clever software to the human group cognition that the software mediates. This transition — from domain-oriented design environment to collaborative information environment — reflects a broader argument running through Part I: designing software for collaboration requires rethinking from the ground up what software is for. Rather than a sophisticated information retrieval or critique system for individual users, a collaborative information environment must support the communicative and knowledge- building processes by which groups construct shared understanding together. The chapter also reviews theoretical arguments from communities-of-practice research to the effect that knowledge in a community is always evolving, tacit in important ways, and distributed across participants — which sets further requirements for any software that would support it.
Perspectives on Collaborative Learning
This chapter reflects on the design, testing, and ultimate failure of WebGuide — the author's most ambitious software development effort — as a study of what is required to support collaborative learning through perspectival access to a shared knowledge space. WebGuide was a Java application that allowed individuals and groups to develop and express distinctive points of view on a shared topic. It was tested in a middle school environmental science course (where students investigated alternative perspectives on the problem of acid mine drainage) and in a graduate doctoral seminar examining theoretical texts. Each round of use generated feedback that drove redesigns of the interface and the underlying algorithmic mechanisms.
Despite repeated refinement over two years, the system proved too complex for users to adopt fluidly. The chapter's significance lies in the author's reflective candor about what this failure revealed. WebGuide's difficulties were not merely technical; they pointed to a deeper theoretical problem. The design had assumed that computer-mediated perspectival access to shared information was itself sufficient to generate collaborative knowledge building. Experience showed that collaborative knowledge building is a far more complex social and communicative process than the software model had assumed. The failure of WebGuide, and the persistent failure of similar systems at other research universities, convinced the author that a careful empirical study of how collaborative learning actually happens in practice was necessary before better software could be designed. This recognition is what motivates the methodological turn of Part II.
Groupware Goes to School
This chapter describes the transformation of BSCW — a widely used CSCW groupware platform used by hundreds of thousands of people for cooperative work — into BSCL (the Basic Support for Collaborative Learning component of Synergeia), a CSCL system adapted for use in educational settings. The chapter provides a sustained analysis of the important differences between CSCW and CSCL contexts. In the workplace, groups of professional experts collaborate on tasks with commercially valued outcomes; in the classroom, a teacher structures the goals and activities of a group of relative novices whose primary objective is learning, not task completion. These contextual differences required fundamentally redesigning the software's model of group interaction.
The adaptations documented in the chapter include creating virtual learning places with distinct organizational structures, supporting collaborative knowledge building and portfolio sharing in addition to document management, implementing perspective-based views of shared workspaces, and laying the groundwork for knowledge negotiation support. Each of these features reflects a theoretical judgment about what collaborative learning requires that mere cooperative work does not. The chapter concludes that taking groupware to school is not a matter of minor interface adjustments but of rethinking what the software is for. The experience also suggests that the distinction between CSCW and CSCL is not arbitrary — the social contexts are genuinely different — but that each domain can learn from the other, particularly as knowledge work in the economy increasingly requires continuous collaborative learning.
Knowledge Negotiation Online
This chapter develops the concept of "knowledge negotiation" as a process qualitatively different from the voting-based negotiation supported by standard CSCW systems, and describes its implementation in BSCL. In CSCW, negotiation is typically modeled as the process by which group members choose among pre-existing options — a quasi-democratic mechanism for reaching group decisions. In collaborative learning, what must be negotiated is something different: the status of knowledge artifacts that the group itself is in the process of constructing. Knowledge negotiation in BSCL involves collectively refining a shared document or proposal from an initial contribution to a mutually acceptable published artifact, with explicit support for proposal, discussion, revision, and publication stages.
The chapter reviews relevant CSCW approaches to negotiation — particularly those developed by German colleagues working with related groupware platforms — and explains why each is inadequate for the collaborative learning context. It then describes the specific implementation of negotiation support in BSCL. The chapter concludes on a cautionary note: when the author subsequently tried to deploy the negotiation system in his own university courses in the United States, students did not use it as intended. This failure underscored the importance of studying empirically how negotiation is actually conducted in practice by online collaborative groups. This recognition rounds out Part I's consistent lesson: software designed for collaboration must be grounded in a genuine understanding of how collaboration works, and such understanding requires careful empirical investigation.
Introduction to Part II: Studies of Interaction Analysis
This introduction looks back on the Part I studies and forward to Part II's methodological turn. The author reflects on what each of the eight design studies ultimately revealed: that the consistent failure to achieve the intended forms of collaborative knowledge building was not primarily a technical failure but a theoretical one. Designers — including the author himself — did not yet understand well enough the fine-grained social and communicative processes involved in actual collaborative interaction. Widely used CSCL research methodologies, borrowed from educational and cognitive psychology, tended to quantify and aggregate collaborative interactions in ways that lost precisely the phenomena most relevant to understanding what happens when collaboration succeeds or fails.
The introduction proposes a methodological alternative: detailed, qualitative analysis of specific moments of collaborative interaction, drawing on conversation analysis and related approaches from ethnomethodology and communication theory. This approach treats the detailed content of utterances, their sequential organization, their relation to artifacts and context, and the moment-by-moment construction of shared meaning as the primary data for studying collaboration. The studies in Part II implement this alternative and, in doing so, provide the first clear empirical evidence in the book for the reality of group cognition as a phenomenon that cannot be reduced to the contributions of individual participants.
A Model of Collaborative Knowledge Building
This chapter presents a diagrammatic model of collaborative knowledge building that identifies a cycle of interrelated phases at both the personal and social levels. Beginning with personal background understanding — the tacit, pre-reflective familiarity with a domain that constitutes the basis of individual knowing — the model traces a path through the articulation of tacit understanding in language, public expression of that articulation in discourse, social processes of negotiation and refinement among group members, and the codification of the resulting collaborative knowledge in shared cultural and computational artifacts. These artifacts, in turn, shape the background understanding of individual participants in subsequent cycles of activity.
The model draws on neo-Piagetian conflict theory, cultural-historical activity theory, social practice theory, and Deweyan transactional inquiry, attempting to synthesize their insights into a unified framework while acknowledging that any model necessarily reifies what is in reality fluid and complex. Crucially, the model addresses the relationship between individual cognitive processes and social-cultural processes, arguing that they are mutually constituting: personal understanding and group knowledge each require the other. The chapter presents the model as a starting point for conceptualizing CSCL software design and for evaluating collaborative interactions — not a definitive theory but a useful schema for identifying which cognitive and social processes a given software environment supports and which it leaves without support.
Rediscovering the Collaboration
Written as a commentary on two representative CSCL studies in the CSCL2 volume, this chapter argues that widely used research methodologies in the field systematically lose sight of what is most important and most interesting about collaborative learning. The two studies under review — one examining university-level collaborative knowledge building with CSILE, the other investigating inquiry-based scientific thinking in Finnish schools — reach conclusions about the teacher's role in supporting deep collaborative learning through quantitative analysis of students' contributions to shared online databases. Both papers follow the same structure: review of abstract pedagogical principles, statistical analysis of note types in CSILE databases, conclusions about individual student learning.
The author's critique is methodological: by coding student contributions and applying statistical analysis to the counts, these studies reduce rich collaborative interactions to a small number of measurable categorical distinctions. The most interesting phenomena — the moment-by-moment construction of shared understanding through interaction — become invisible. The chapter argues that CSCL cannot afford to let the most powerful available statistical methods determine what gets studied; rather, the phenomena of collaborative learning — genuine instances of groups building shared knowledge together — should determine the methodological choices. The chapter recommends detailed conversation analysis as a necessary complement to quantitative approaches, and sets up the methodological alternative that the remaining chapters of Part II implement.
Contributions to a Theory of Collaboration
Combining the author's introduction to the CSCL 2002 conference with his paper from that conference, this chapter proposes four themes as the foundations for a theory of computer- supported collaborative learning. The first is collaborative knowledge building: the framing of learning as a group-level process that produces shared understanding and collaborative artifacts, rather than as the individual accumulation of facts. The second is group and personal perspectives: the recognition that collaborative interaction involves the interplay of individual viewpoints and group-level understanding, and that both must be taken into account in research and design. The third is mediation by artifacts: the insight that collaboration always occurs through and with material, digital, and linguistic artifacts, which both carry and transform the meanings being constructed. The fourth is micro-analysis of interaction: the methodological commitment to studying the detailed sequential structure of collaborative discourse as the primary data for understanding how learning and knowledge building actually occur.
The chapter argues that CSCL has reached a point where it needs a theoretical paradigm appropriate to its distinctive concerns — one that takes the group, not the individual, as the primary unit of analysis, and that grounds theoretical claims in detailed empirical study of the interaction through which groups actually build shared knowledge. The chapter also argues that learning in knowledge work and in schools is converging: both require collaborative inquiry, not the transmission and reception of fixed facts. Computer support must be designed for this new reality.
In a Moment of Collaboration
This chapter presents a detailed conversation analysis of approximately thirty seconds of interaction among five middle school students working with SimRocket, a computer simulation of model rockets. The analysis focuses on how the students' brief, fragmented utterances — most consisting of only one to four words — nonetheless accomplish complex shared meaning through three properties: indexicality (implicit reference to the current situation of the participants and the artifact), ellipsis (reliance on prior context so that utterances make sense only in sequence), and projectivity (each utterance shapes what the next speaker can coherently say). The students encounter confusion about which rockets they are comparing; this confusion is repaired through a rapid sequence of mutually oriented contributions; and a shared understanding emerges about which rockets in the simulation list are comparable.
The analysis demonstrates that this shared understanding — the recognition that two specific rockets share certain properties and differ in others — cannot be attributed to any individual participant. It is constructed in the interaction itself, visible in the sequence of utterances and their relation to the shared artifact, and constitutes a genuine group cognitive achievement. This is the book's central empirical finding: that collaborative groups can build understanding together that is not reducible to the sum of what the participants individually knew before the interaction. Group cognition is not a theoretical postulate but an observable phenomenon in the data of real collaborative interactions.
Collaborating with Relational References
Extending the SimRocket analysis from the previous chapter, this study follows the same group of students as they collectively master a more complex way of understanding the simulation's list of rockets. In the previous chapter, the group had trouble identifying which rocket was which; in this chapter, the analysis reveals that the difficulty ran deeper: the students had to learn to see the list not merely as a set of individual rockets but as a set of paired configurations designed to allow the systematic testing of one variable at a time while holding the others constant. This relational way of seeing — understanding the list as encoding controlled comparisons — is the prerequisite for conducting proper scientific experiments. The chapter traces the step-by-step process by which the group, guided by an adult mentor and by each other's contributions, collectively arrives at this understanding.
No individual student possessed this relational understanding before the collaborative session; it was constructed through group interaction with the artifact and with each other. The analysis shows how the artifact — the SimRocket list with its deliberately designed structure — embodied a meaning that the designer had encoded into it, and that the students had to learn to read out of it through collaborative discourse. This case study illustrates the general argument that collaborative conceptual change is a genuine group-level achievement: a new way of seeing something that the group collectively arrives at and that cannot be adequately described as the sum of individual changes in understanding.
Introduction to Part III: Studies of Collaboration Theory
This introduction reviews the empirical findings of Part II — above all the evidence that group cognition occurs in the SimRocket collaboration — and argues that these findings call for a theoretical framework adequate to understand them. The group's shared understanding of the rocket list was constructed in the interaction itself; it was not in the mind of any individual before the collaboration, and it cannot be fully explained by reference to what any individual brought to the interaction. This is a striking finding that challenges the individualistic assumptions of mainstream cognitive science. The introduction provides the intellectual background for Part III by tracing the author's philosophical formation through the three main currents of twentieth-century European philosophy: critical social theory (Marx, Adorno), common language analysis (Wittgenstein, Austin), and existential phenomenology (Heidegger, Merleau-Ponty). Each of these traditions undertook a fundamental critique of the positivist, mentalist, individualist assumptions that have dominated cognitive science, and each took a "linguistic turn" — recognizing that thought and meaning making are fundamentally linguistic phenomena and therefore potentially social ones. These philosophical resources provide the conceptual vocabulary for Part III's theory of group cognition.
Communicating with Technology
This chapter surveys the major families of communication theory — cybernetics, semiotics, conversation analysis, message production and reception, symbolic interaction, socio-cultural theory, phenomenological hermeneutics, and critical theory — and assesses their relevance to the design of groupware as a medium for online collaborative learning. The author argues that most received communication theories are structured around the model of a message transferred from a sender to a receiver, with communication success defined as the accurate transmission of pre-formed meaning. This model is adequate for many purposes but deeply inadequate for understanding collaborative knowledge building, in which the meaning being communicated is not pre-formed but is itself constructed through the communicative process.
The chapter draws consequences for the design of CSCW and CSCL systems: groupware must be understood not as a channel for transmitting pre-existing meanings but as a medium that enables new forms of shared meaning to emerge through group interaction. Computer- mediated collaboration lacks the non-verbal cues of face-to-face interaction — gesture, gaze, tone of voice, physical proximity — and must therefore develop its own explicit forms of communicative coordination. Drawing on ethnomethodology, the chapter argues that designing effective groupware requires understanding how meaning is actually constructed in group discourse — how turns are organized, how references are established and repaired, how group understanding is displayed and acknowledged — and that this understanding must come from detailed empirical study of real collaborative interaction.
Building Collaborative Knowing
This chapter develops the book's most systematic theoretical account of group cognition under the concept of "building collaborative knowing." The deliberately awkward phrase is chosen to direct attention away from the individual accumulation of mental representations toward the group process of constructing shared understanding. The chapter synthesizes insights from situated cognition, cultural-historical activity theory, distributed cognition, social practice theory, and ethnomethodology to argue that collaborative knowing — a group collectively arriving at a new degree of understanding about a topic it is investigating — is a core phenomenon for both CSCW and CSCL.
The chapter introduces key theoretical concepts and shows how they cohere in a social theory of collaboration. Intersubjectivity names the shared world of meanings that participants inhabit and that constitutes the context of their interaction. The zone of proximal development, borrowed from Vygotsky, is extended from the individual-teacher dyad to the collaborative group. Artifact mediation recognizes that the objects and tools of collaborative activity — physical, digital, and linguistic — are not neutral instruments but carriers of meaning that actively shape the interaction. The group perspective denotes the emergent viewpoint of the group as a whole, not reducible to any individual's perspective. Building on these concepts, the chapter argues that the mark of successful collaboration is precisely that something new comes out of the group's interaction that cannot be attributed to any individual participant's prior contribution — an emergent achievement of group cognition.
Group Meaning / Individual Interpretation
This chapter develops a philosophical distinction between meaning and interpretation that is central to the theory of group cognition. Meaning, the author argues, is essentially shared: it exists in the intersubjective world of discourse, language, and cultural artifacts, not in individual minds. When a word, phrase, or artifact means something, it means it for a community of speakers and users who share the relevant practices and contexts. Interpretation, by contrast, is necessarily perspectival: individuals understand shared meanings from their own particular positions, experiences, and concerns, which may differ from each other's even when they share a language and a situation.
This distinction — traced through the philosophical tradition from Plato through Hegel, Husserl, Heidegger, and Wittgenstein — has direct implications for CSCL. If meanings are shared rather than private, then the analysis of collaborative learning should focus on the public discourse and the shared artifacts in which group meanings are displayed and negotiated, rather than making inferences about the private mental states of individual participants. The chapter responds to the field's consensus programmatic definition — that CSCL is centrally concerned with meaning and the practices of meaning making — by arguing that this phrase should be read as pointing to irreducibly social phenomena, not to psychological processes occurring inside individual heads. Collaborative learning is not just individuals processing information in proximity; it is the joint construction of shared meaning in and through group interaction.
Shared Meaning, Common Ground, Group Cognition
This chapter critically examines the concept of "common ground" as it is standardly used in cognitive science and CSCL — the idea that interlocutors share meaning by virtue of the overlap between their individual mental representations of relevant facts and assumptions — and argues that this account falls short of what is needed for understanding group cognition. Common ground, on the standard view, is established through the mutual acknowledgment that specific information has been successfully transmitted between individuals; it is essentially an accumulation of individually held beliefs that happen to be shared. Group cognition, however, involves something more and different: the collaborative construction of understanding that is not reducible to any prior set of individual beliefs, but emerges from the interaction itself.
The chapter proposes instead to treat shared meaning as an emergent property of group discourse: something constructed in and through collaborative interaction, observable in the visible and recorded discourse, and not reducible to the sum of individual mental states. An analysis of an online chat excerpt from the Virtual Math Teams project provides a complementary empirical case alongside the SimRocket data from Part II: in the VMT example, students collaborating on a mathematical problem in a text chat environment construct a shared mathematical insight through their sequenced contributions. The chapter argues that studying group cognition requires analyzing the discourse at the level of the group — taking the group's interaction as the unit of analysis — not aggregating the mental states of individual participants.
Making Group Cognition Visible
This chapter addresses the core methodological challenge for CSCL: how can researchers observe and analyze processes of group cognition? The author proposes a framework of five interpretive perspectives that must be distinguished in any rigorous analysis of collaborative interaction. Individual group members interpret each other's contributions during the live event of collaboration. The group as a whole constructs shared meanings and knowledge artifacts through the interaction of those individual contributions. The broader community of practice provides the socio-cultural context within which the group operates. Researchers analyze the group's behavior by studying recorded data — video clips, transcripts, chat logs — from a perspective outside and after the event. Designers of technology, pedagogy, or social practices interpret the findings of research in order to develop new forms of support for future collaborative groups.
Carefully distinguishing these five perspectives, the chapter argues, is what allows researchers to study group meaning rather than projecting individual-level interpretations onto group phenomena. Drawing on Garfinkel's ethnomethodology, the chapter proposes that the data needed to study group cognition is everywhere in the visible, recorded discourse of collaborative groups: group meaning is displayed in the sequential organization and content of the utterances themselves, not hidden in unobservable mental states. This methodological principle is illustrated with both the SimRocket data and initial data from the Virtual Math Teams project. The chapter argues that making group cognition visible is not a matter of developing special observational instruments but of adopting the right unit of analysis and the appropriate interpretive framework.
Can Collaborative Groups Think?
By analogy with Turing's classic question "Can machines think?" this chapter asks whether collaborative small groups can be said to think. The analogy is not merely rhetorical. Three arguments were developed in the decades-long AI debate — that a system can be credited with thinking if it can produce behavior indistinguishable from intelligent human performance (Turing), if it understands the symbols it processes (Searle), and if it can act intelligently in a real-world context (Dreyfus) — and the chapter applies each to the case of collaborative groups. On all three criteria, the author argues, small groups of collaborating people are stronger candidates for cognition than computers are.
Groups produce outputs — collaborative knowledge, shared decisions, creative solutions — that could not have been predicted from or attributed to individual members. Groups understand the meanings of the symbols they exchange: the meaning of an utterance is constituted in and through the interaction, not merely manipulated as a formal symbol. And groups act intelligently in the world: they adapt to context, repair misunderstandings, and accomplish tasks that no member could accomplish alone. At the same time, accepting group cognition does not require positing a mysterious "group mind" as a separate entity; it requires only adopting the group as the unit of analysis for understanding processes of shared meaning making that cannot be adequately explained at the level of individual participants. The chapter argues that CSCL should take the concept of group cognition as seriously as AI once took the concept of machine intelligence, and that doing so could define a research agenda as productive as Turing's question proved to be for computing.
Opening New Worlds for Collaboration
This chapter draws on Heidegger's philosophy of language and world-disclosure, alongside Marx's critique of ideological thinking, to argue that CSCL needs to develop its own theoretical tradition rather than uncritically importing conceptual frameworks from engineering or individual cognitive psychology. The chapter contrasts what the author calls the "American engineering mentality" — represented by Shannon's mathematical theory of communication as information transfer from sender to receiver — with the tradition of German philosophy, which sees language not as a neutral channel but as the medium through which worlds of shared meaning are opened up. Applied to CSCL, the engineering model treats software as a neutral delivery channel and learning as the efficient reception of transmitted information. The philosophical alternative sees the design of collaborative environments as the opening of new spaces in which groups can build shared understanding together in ways that would not otherwise be possible.
The chapter draws on Heidegger's concept of world-disclosure: great artworks, founding political events, and scientific discoveries each open new worlds by establishing new frameworks of shared meaning within which things can appear and matter in new ways. Computer-supported collaborative learning environments, designed well, might do something analogous: create new conditions under which groups can engage in forms of collaborative knowledge building not otherwise available. Marx's ideology critique contributes the warning that taken-for-granted theoretical assumptions — including individualism, the sender-receiver model of communication, and the engineering metaphor of computation — may prevent researchers from seeing what is genuinely new and important about collaborative cognition.
Thinking at the SmallGroup Unit of Analysis
The book's concluding chapter synthesizes its methodological, analytic, and theoretical contributions through an introduction to the Virtual Math Teams (VMT) project, a continuing research program studying how small groups of students collaborate on mathematical problem solving in computer-mediated chat environments. The chapter presents initial examples of VMT data — excerpts of online mathematical discussion in a text chat system — and analyzes how group-level mathematical reasoning and argumentation emerge through the interaction. In particular, the chapter shows how students collectively develop mathematical conjectures, test them through collaborative discourse, and arrive at group-level conclusions that no individual participant had formulated before the interaction began.
The chapter addresses three senses in which "thinking at the small-group unit of analysis" matters. Methodologically, it argues that CSCL researchers should focus on the group discourse as their primary data, since this is where the phenomena of collaborative knowledge building are visible. Analytically, it argues that what takes place in that discourse — the sequenced construction of shared mathematical meaning — meets any reasonable criteria for genuine thinking. Theoretically, it distinguishes the concept of group cognition as developed in the book from related concepts such as distributed cognition, arguing that group cognition specifically concerns the emergent meaning-making of small groups engaged in face-to-face or computer-mediated interaction. The chapter closes by sketching the research agenda that the VMT project represents: systematic empirical investigation of how groups think together, guided by the theoretical framework that the book has developed.
In addition to the full Webinar on Group Cognition created online with relatively poor video quality in Dec 2012, ISLS created shorter, high quality versions in June 2013:
Webinar video of Gerry Stahl on "Group Cognition, the Foundation of the Learning Sciences: The Philosophy of Group Cognition" from the ISLS NAPLES video project (90 minutes).
Abstract: Cognition is no longer confined to the solitary musings of an armchair philosopher, but takes place, for instance, in problem-solving efforts of teams of people distributed around the world and involving various artifacts. The study of such cognition can unfold at multiple units of analysis. Here, three cases of problem solving by virtual math teams demonstrate the mix of individual, group and social levels of cognition. They show how a resource like a mathematical topic can bridge the different levels. Focusing on the under-researched phenomena of group cognition, the presentation highlights three pre-conditions for the constitution of group cognition: longer sequences of responses, persistent co-attention and shared understanding. Together, these structure a virtual analog of physical embodiment: being-there-together, where what is “there” is taken by the participants as co-experienced.