Volume 4. Studying Virtual Math Teams
* Download PDF free for reading online or printing: svmt.pdf
(This is the best version: up-to-date, complete, full color, hi-res.)
* Download iBook version free for iPad, Mac, etc.: svmt.epub
* Download MOBI version free for Kindle: svmt.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
Introduction to Part I: Introducing Group Cognition in Virtual Math Teams
This introductory chapter frames the entire volume by situating the Virtual Math Teams (VMT) Project within the broader field of computer-supported collaborative learning. The VMT Project, launched in 2002 at the Math Forum, pursued three intertwined goals: providing an online mathematical community service for students, designing a technology environment to support small-group collaboration, and conducting research on how groups achieve cognitive tasks together. The book itself embodies the collaborative nature of its subject, assembling contributions from a wide research team whose members co-constructed the project's methods, findings, and theory over several years. The introduction establishes the project's foundational commitment to the small group as the unit of analysis — neither the isolated individual nor the broad community — and positions "group cognition" as a phenomenon that exceeds what any individual contributes, emerging from the interactive processes among participants. Design-based research is described as the methodological stance that ties service, design, and research together, allowing iterative cycles in which theoretical insights reshape system design and vice versa.
A Chat about Chat
Presented as an extended interview, this chapter diagnoses the state of computer-supported collaborative learning and identifies its central challenge: the field's persistent individualistic bias. Most research and most educational technology still treats learning as something that happens inside individual minds, even when students work together. The chapter argues that this bias is the primary barrier to understanding what is unique and powerful about collaborative learning. The alternative proposed is a social-constructivist view in which knowledge is built through group interaction, particularly through synchronous small-group text chat. Chat is singled out as a medium with distinctive properties that make it especially productive for collaborative knowledge construction: it slows conversation enough to invite reflection, it externalizes discourse in readable form, and it creates engagement that reduces the sense of isolation often accompanying distance learning. The chapter advocates for shifting the object of research from individual cognition to group meaning-making as the site where genuinely new knowledge is produced.
The VMT Vision
This chapter presents a public-facing account of the VMT Project's mission and design. Funded by NSF beginning in 2003, the project aimed to build an online math discourse community for K-12 students worldwide. Small groups of students — typically three to five — would meet in chat rooms for focused math problem-solving sessions, supported by a shared whiteboard and organized around problems posed by the Math Forum. The VMT Vision emphasized the creation of authentic mathematical discourse communities across geographic and institutional boundaries. The chapter describes the design-based research cycles through which the project iterated between user testing, analysis of recorded sessions, and redesign of the environment. It positions the project as both a practical service and a research testbed, explaining how the simultaneous pursuit of service, design, and inquiry is not a contradiction but a productive methodological commitment.
Mathematical Discourse as Group Cognition
This chapter presents a close review of a major theoretical work on mathematics learning and uses it to ground the VMT Project's research orientation. The central argument is that mathematical objects — numbers, geometrical figures, algebraic expressions — are not pre-existing entities discovered by individual minds but reifications of discursive processes: they come into being through communicative acts performed by communities of practice. The chapter draws on the distinction between an acquisition metaphor for learning (in which knowledge is a possession transferred from teacher to student) and a participation metaphor (in which learning is a process of becoming a member of a discourse community). The analysis shows how this theoretical framework makes collaborative mathematical discourse the primary site of mathematical knowledge construction, rather than a secondary medium for expressing pre-formed individual ideas. For the VMT Project, this means that group chat interactions are not mere communication about math but the very place where mathematical thinking occurs.
Interactional Methods and Social Practices in VMT
This chapter identifies three persistent challenges facing online collaborative learning and examines how the VMT Project grapples with each. The first challenge is achieving genuine depth of learning rather than superficial exchange, ensuring that students engage in productive mathematical inquiry rather than performing participation. The second is introducing inquiry-based mathematical practice into contexts dominated by formal schooling conventions that reward answer-finding over problem exploration. The third is the integration of pedagogy, technology, and sociability — ensuring that the social and motivational dynamics of student interaction reinforce rather than undermine mathematical learning. The chapter argues that meeting these challenges requires attention to the emergent social practices and interactional methods that groups develop in the VMT environment. These practices are not prescribed by the design but arise through participants' navigation of the technological affordances and interpersonal dynamics of the chat medium.
From Individual Representations to Group Cognition
This chapter presents one of the book's core methodological demonstrations: a single VMT chat episode analyzed twice using two different frameworks. The first analysis treats each posting as an expression of an individual participant's mental representation — a sender-receiver model of communication. The second treats the same sequence of postings as a jointly co-constructed process in which meaning emerges from the interaction itself rather than from any individual mind. The chapter shows that neither analysis alone is adequate: the individual-cognitive lens misses the collaborative accomplishment, while the group-cognitive lens cannot attend to the specific contributions of particular participants. The key argument is that groupware design must support the group as a genuine unit of analysis. Software that merely aggregates individual inputs or presents information to individuals separately cannot support group cognition. The chat interface must be designed to make the shared, sequential, and referential structure of interaction visible and actionable for participants.
Introduction to Part II: Studying Group Cognition in Virtual Math Teams
This introduction to Part II frames the four doctoral dissertations that form the section's core as case studies of group cognition in VMT. Each dissertation applies ethnomethodological conversation analysis to examine specific aspects of how virtual math teams construct shared understanding through sequential interaction. The introduction reflects on a notable parallel: the VMT research team's own collaborative process — building a shared research agenda across rotating membership over several years — mirrors the very group-cognitive processes it studies. The problem space of the research project itself was co-constructed through the team's joint discourse, demonstrating that group cognition is not only an object of inquiry for the VMT Project but a condition of its possibility.
The Sequential Co-construction of the Joint Problem Space
This chapter develops the concept of the joint problem space (JPS) as a group-level phenomenon constructed through sequential interaction. Originally proposed in cognitive science to describe the representation an individual solver maintains while working on a problem, JPS is extended here to characterize what a small group builds together in the course of collaborative problem solving. The chapter identifies three dimensions of the JPS: the social-relational dimension (who is participating and in what roles), the domain-content dimension (the mathematical ideas and representations in play), and the temporal dimension (how prior contributions are referenced and accumulated over time). The temporal dimension is shown to be critical and often underappreciated: groups do not merely share a spatial or topical common ground but construct a shared history of interaction that each new posting draws upon and extends. Collective remembering and deictic referencing are identified as the primary mechanisms through which the JPS is maintained and extended.
The Organization of Graphical, Narrative and Symbolic Interactions
This chapter offers the most technically detailed analysis in Part II, a fine-grained ethnomethodological study of an eighteen-minute VMT excerpt in which three students work simultaneously in text chat and a shared whiteboard. The chapter documents how participants coordinate three semiotic modalities — graphical inscriptions on the whiteboard, narrative text in chat, and symbolic mathematical notation — to construct a shared indexical field that serves as common ground for group cognition. The analysis traces how references cross between the two spaces: a whiteboard inscription becomes relevant through a chat posting that indexes it, and a chat posting is disambiguated by simultaneous whiteboard activity. Sequential organization is shown to structure not only chat turn-taking but also the timing and placement of whiteboard contributions. The chapter argues that the dual-interaction space is not simply a convenience but a constitutive feature of VMT group cognition.
Question Co-Construction in VMT Chats
This chapter examines how questions function in VMT group interaction. Rather than treating questions as simple information requests issued by one participant and answered by another, the analysis shows that questioning in VMT is a complex interactional achievement. Questions are not posed by individuals with pre-formed informational needs but are co-constructed through sequences of postings in which the object of the question — what is being asked about — emerges from the group's joint engagement with the problem. The chapter also examines how differences in perceived mathematical competence among participants shape questioning sequences: who poses questions, who is addressed, and how responses are evaluated are all entangled with the ongoing negotiation of participant roles. Question-and-response sequences are shown to be among the primary mechanisms through which groups build and maintain the joint problem space.
Resolving Differences of Perspective in a VMT Session
This chapter investigates how small groups manage disagreement and competing approaches in collaborative math problem solving. The analysis draws on traditions emphasizing both cognitive conflict (the productive confrontation of incompatible ideas) and the inter-animation of perspectives (the dialogic process of incorporating another's view into one's own thinking). Through posting-by-posting analysis of a VMT session in which participants pursue divergent solution strategies, the chapter identifies the specific interactional methods groups use to resolve differences: rephrasing a prior posting in terms acceptable to both participants, redirecting attention toward a shared element, and explicitly negotiating criteria for evaluating candidate solutions. The chapter shows that resolution is not simply the victory of one perspective over another but the group's construction of a new, shared orientation that neither participant held individually before the interaction.
Introduction to Part III: Studying Group Discourse in Virtual Math Teams
This introduction situates the chapters of Part III as reflections on the analytical methods the VMT Project developed for studying chat interaction. Design-based research, as practiced in VMT, requires not only designing and deploying a system but also developing appropriate tools for analyzing the interaction the system supports. The introduction explains how VMT researchers drew on ethnomethodological conversation analysis as a starting point but found it necessary to adapt that tradition significantly: ordinary conversation analysis was designed for talk-in-interaction, not for text chat, and it assumes an analyst who stands outside the social setting being analyzed, whereas VMT researchers were simultaneously designers of the environment they studied. Part III's chapters thus reflect on the larger methodological and theoretical implications of VMT data analysis across multiple sessions and multiple teams.
Representational practices in VMT
This chapter presents an external team's analysis of VMT Spring Fest 2006 session data, examining how small groups develop and transform their representational practices across multiple sessions. The central finding is that the representations groups use — their diagrams, notational choices, and vocabulary — are not merely tools selected for immediate problem-solving convenience but practices that carry forward from one session to the next, shaping how participants approach new problems. The chapter introduces the concept of situated cognition contingent on chronologically prior resources: the group's current activity is structured not only by the immediate situation but by the history of prior interactions that established the representational practices now in use. Inscriptions become representations — become meaningful — through the interactive process by which participants interpret and build upon them in subsequent postings.
Student and Team Agency in VMT
This chapter examines agency — the capacity to initiate, direct, and evaluate action — as both an individual and a collective phenomenon in VMT. Drawing on concepts from anthropology, psychology, and sociology, the chapter distinguishes individual epistemic agency (the capacity to generate and evaluate knowledge claims) from team agency (the group's capacity to direct its own collaborative process). The VMT environment is shown to afford and encourage both forms of agency: the absence of an adult authority figure in the chat room, the openness of the mathematical problems, and the technical design of the shared space all support students in taking ownership of their mathematical inquiry. The chapter uses concepts from structuration theory and the notion of habitus to explain how individual and team agency are mutually constituted through the ongoing interaction, with each posting both enacting and reshaping the group's dispositions and social structure.
Group Creativity in VMT
This chapter reconceptualizes creativity as a social and interactional phenomenon rather than a property of individual minds. The analysis identifies three fundamental processes through which groups produce creative work in VMT: indexical referencing (using a posting or inscription to pick out and focus group attention on a particular element), group remembering (the collective maintenance and reactivation of prior contributions), and bridging across discontinuities (connecting ideas across gaps in time, topic, or participation). The chapter examines creativity at two levels: the micro-level of a single session, where novel ideas emerge from specific sequential moves within a chat, and the evolutionary level across multiple sessions and teams, where creative discoveries accumulate and diffuse. The analysis challenges the assumption that creative insight belongs to individuals, showing instead that the most significant mathematical ideas in VMT data emerge from the interactional space between participants.
Inscriptions, Mathematical Ideas and Reasoning in VMT
This chapter shifts analytical focus from text chat to whiteboard inscriptions, examining how graphical marks made on the shared whiteboard participate in the collaborative construction of mathematical ideas. The Rutgers research team analyzes VMT sessions involving combinatorics problems, tracking how students use whiteboard inscriptions as heuristic devices — provisional, exploratory marks that help the group feel its way toward a solution rather than executing a predetermined algorithm. The chapter shows that inscriptions are layered signs: each new mark draws on and transforms prior ones, building a cumulative sign structure through which mathematical understanding develops. Collaborative interaction functions not simply as communication about pre-existing mathematical ideas but as the process through which those ideas come into being, with participants challenging each other's inscriptions and prompting revisions that advance the group's mathematical reasoning.
Reading’s work in VMT
This chapter analyzes the distinctive features of chat as a medium for interaction, focusing on what reading — not writing — accomplishes in chat conversations. In face-to-face conversation, turn-taking is managed through vocal and gestural cues that are visible to all participants as a turn is being produced. In text chat, the composition of a message is invisible until it is sent, making it impossible for co-participants to monitor message production as it unfolds. The chapter identifies the interactional work that reading performs in this context: establishing co-presence, coordinating indexical grounds among participants, and organizing the sequential structure of interaction through the choice of which prior postings to respond to. The analysis shows that the systematics of chat — the rules governing how a well-formed chat interaction is produced — differ fundamentally from the systematics of spoken conversation, requiring different analytic concepts and frameworks.
Introduction to Part IV: Designing the VMT Collaboration Environment
This brief introduction frames the design history of the VMT collaboration environment, centering on the development of ConcertChat by a German research team and its adoption and adaptation by the VMT Project. The dual-interaction space — combining a synchronous text chat with a shared graphical whiteboard — is introduced as the design's central architectural feature, motivated by the conviction that neither text alone nor graphics alone can support the full range of mathematical communication that collaborative groups require. The introduction also describes the VMT Lobby, which allowed groups to move between problem sessions and community spaces, and the wiki, which provided an asynchronous medium for recording and sharing mathematical work across sessions. Three layers of scripting used in VMT's Spring Fest events are identified as a further design response to the challenge of supporting productive collaborative engagement.
The Integration of Dual-interaction Spaces
This chapter documents the design of ConcertChat, focusing specifically on the technical and interactional challenge of integrating the text-chat space and the shared-whiteboard space into a coherent collaboration environment. The German design team identifies three distinct integration problems and the design solutions developed to address them. First, participants needed a way to connect their chat postings to specific objects in the whiteboard space; this was addressed with a graphical referencing tool that draws an arrow from a posting to any referenced whiteboard element. Second, participants needed to track activity in both spaces simultaneously; this was addressed with a history feature that records and replays the sequence of events in both spaces. Third, participants needed to maintain awareness of each other's activity across the two spaces; this was addressed through a social-awareness display. The chapter discusses the different persistence characteristics of the two media and the interactional consequences of each design choice.
Designing a Mix of Synchronous and Asynchronous Media for VMT
This chapter takes a broad human-computer interaction perspective on the design of the VMT environment, arguing that the field of HCI must shift its fundamental model from individual human-computer interaction to group human-computer interaction. The VMT system is analyzed as an integrated ecology of media operating at multiple timescales and social levels: the synchronous chat room supports immediate small-group interaction; the VMT Lobby supports movement between rooms and asynchronous coordination; the wiki supports the accumulation and sharing of mathematical work across sessions and teams. The chapter argues that this multi-level architecture is not incidental but necessary for supporting learning at all three levels — individual, small-group, and community — simultaneously and in their proper relationships. A college HCI design course that used VMT as its subject matter illustrates how the design principles can be taught and critically examined.
Deictic Referencing in VMT
This chapter examines the graphical referencing tool in ConcertChat — the feature that allows a chat posting to be visually linked by an arrow to an element in the shared whiteboard — as a case study in the theory of deixis. Deixis, the use of linguistic or gestural expressions (words like "this," "here," "that one") to pick out objects in a shared situation, is identified as a foundational mechanism of intersubjective cognitive processes. Without the capacity to jointly attend to objects in a shared environment, groups cannot build the common ground needed for collaborative reasoning. The chapter argues that cognitive tools designed to support collaborative communities are fundamentally different in kind from tools designed to support individual cognition: they must support the establishment of shared attention and joint reference among multiple participants, not merely amplify one person's ability to process information. A detailed case study traces how a group uses the graphical referencing tool to orient collectively toward a specific mathematical object.
Scripting Group Processes in VMT
This chapter revisits the concept of scripts — structured plans or schemas that guide interaction sequences — and reconceptualizes them as situated resources rather than implementable plans. The chapter traces the concept of scripting from its origins in cognitive science through Vygotskian notions of structured social activity and the situated-action critique of plan-following, arriving at a view of scripts as flexible, contextually interpreted resources that participants draw upon in structuring their interaction. VMT's Spring Fest events used three layers of scripting: a high-level narrative arc across the event, a set of problems structured to invite progressive inquiry, and a set of facilitator prompts designed to support groups encountering difficulties. The chapter argues that these scripts functioned not as algorithms controlling group behavior but as resources that groups could take up, modify, and deviate from in the course of their own situated activity.
Helping Agents in VMT
This chapter reports on an effort by the Carnegie Mellon team to develop automated computational agents that can support collaborative mathematical problem solving in VMT. The chapter describes two systems: Tag-Helper, which uses text classification to detect features of student interaction, and Basilica, a dialogue agent platform that generates contextually sensitive messages to student groups. The research goal is to support productive collaboration by eliciting the help-seeking, help-giving, and reflective behaviors that research has shown to be associated with learning gains. The methodology follows a full-circle design pattern: naturalistic observation of student groups establishes baseline patterns; laboratory study under controlled conditions identifies variables; technical implementation embeds insights in the agent; and re-deployment in naturalistic settings tests the implementation. A key finding is that groups tend to converge on a single solution approach prematurely; the agents are designed to encourage exploration of alternatives.
Introduction to Part V: Representing Group Interaction in VMT
This introduction frames Part V's collection of analytical methods for representing and analyzing the structure of VMT chat interactions. A central practical problem motivating the section is response-structure confusion: because multiple participants can post simultaneously in a chat, postings often appear in a sequence that does not reflect their functional relationships, making it difficult to identify who is responding to whom. Threading — the explicit or reconstructed linking of postings into response chains — is presented as a necessary first step for any serious analysis of chat data. The introduction also describes a tension in the field between coding approaches (which categorize postings according to predefined schemes and count their frequencies) and conversation-analytic approaches (which analyze the sequential organization of interaction without imposing external categories). Part V's chapters navigate this tension from multiple directions.
Thread-based Analysis of Patterns in VMT
This chapter develops a computational model for analyzing the threading structure of VMT chat logs. The model represents a chat session as a directed graph in which each node contains the full information of one posting — its text, its author, its timestamp — and edges represent response relations between postings. This graph representation makes it possible to mine the chat for recurring structural patterns at multiple scales: dyadic patterns (two-posting sequences such as question-answer or proposal-evaluation pairs), triadic patterns (three-posting sequences involving a third participant or move), and longer sequences. The approach is designed to be systematic and scalable, allowing the analysis of large corpora of chat data without requiring the analyst to read each log in full. The chapter illustrates how the graph-based threading model reveals fine-grained patterns of social, communicative, and mathematical-problem-solving interaction that are invisible to surface-level message counts.
Studying Response-Structure Confusion in VMT
This chapter addresses a problem that is both practical and theoretically significant: the confusion that arises in group chats when multiple participants post simultaneously, creating overlapping threads of discussion whose references become difficult to interpret. The chapter distinguishes explicit confusion (when participants themselves express that they have lost track of the conversational thread) from implicit confusion (when the structure of postings reveals broken reference chains that the analyst must reconstruct). The analysis draws on data from VMT staff research-discussion chats, which involved the largest participant groups in the VMT corpus and therefore the highest rates of parallel posting. The graphical referencing tool in ConcertChat is evaluated for its effectiveness in reducing response-structure confusion. The chapter proposes a typology of reference types — including references to prior postings, to whiteboard elements, and to mathematical concepts — and analyzes how different types of reference contribute to or resolve confusion.
A Multidimensional Coding Scheme for VMT
This chapter reports on an attempt to develop and apply a systematic coding scheme to VMT chat data. The chapter identifies four methodological problems that arise when coding is applied to small-group chat — problems that are endemic to the medium rather than incidental to this particular study. The first is the unit of analysis problem: chat postings vary enormously in length and complexity, making it difficult to define a consistent unit for coding. The second is unit fragmentation: a single conversational move is often distributed across multiple postings, and a single posting may contain elements of multiple moves. The third is the challenge of determining intercoder reliability without inflating it through the structure of the coding scheme itself. The fourth is the validity question: coding schemes developed for face-to-face or dyadic interaction may not be valid for the more complex interaction patterns of multi-party chat. Threading is shown to be essential for managing all four problems.
Combining Coding and Conversation Analysis of VMT Chats
This chapter addresses the methodological tension between statistical-coding approaches and conversation-analytic approaches to VMT data, arguing that the two can be productively combined if the combination is properly sequenced. The chapter reports a research episode in which a hypothesis about collaboration patterns was tested statistically using a coding scheme, yielding a puzzling result that the coding framework could not explain. Conversation analysis of the relevant data then identified the phenomenon responsible for the anomalous result. This experience led to a proposed "mixed methods" approach in which conversation analysis is used first to identify and describe the interactional sequences that will then be coded for quantitative analysis. The chapter introduces a distinction between two broad modes of participation in VMT sessions — expository and exploratory — and shows how transitions between these modes constitute the longer sequences that quantitative analysis should target.
Polyphonic Inter-Animation of Voices in VMT
This chapter introduces a theoretical framework for analyzing collaborative chat interaction drawn from musical theory and the philosophy of language. The central concept is polyphony: in music, multiple independent melodic lines (voices) unfold simultaneously and are woven together into a unified whole through the rules of counterpoint. The chapter argues that collaborative chat exhibits an analogous structure — multiple participants contribute independent lines of thought that are progressively integrated through the give-and-take of posting and response. Drawing on a theory of dialog, the chapter extends the musical metaphor to argue that each participant's voice in a chat carries with it prior utterances, social positions, and cultural resources that are brought into contact with other voices, generating creative tension and the potential for new understanding. Dissonances in the polyphonic structure — moments of conflict or incommensurability between voices — are productive rather than merely disruptive, driving the discourse forward. Software tools for visualizing polyphonic threading are also presented.
A Model for Analyzing Math Knowledge Building in VMT
This chapter proposes a formal model for representing the flow of mathematical knowledge construction in VMT chat sessions. The Collaborative Interaction Model (CIM) represents the discourse as a network of uptake relations: each posting is linked to prior postings from which it takes up — interprets, responds to, builds upon, or challenges — particular informational or conceptual elements. The model distinguishes individual uptake (one participant responding to another's posting) from collaborative uptake (the joint construction of a shared mathematical representation through a sequence of postings). The chapter identifies several analytically significant features that the model makes visible: segmentations (boundaries between distinct phases of problem-solving activity), meaning-making paths (chains of uptake through which a mathematical idea develops), and pivotal moments (postings that significantly redirect the discourse). The model is applied to VMT sessions involving three types of problem design, showing how problem structure influences the shape of knowledge-building discourse.
Introduction to Part VI: Conceptualizing Group Cognition in VMT
This introduction frames the volume's final section as a theoretical and methodological synthesis. It revisits the analytical approaches developed across Parts I through V and reflects on what they collectively reveal about the nature and structure of group cognition. The introduction traces a conceptual arc from the notion of joint problem space, through the indexical ground of shared interaction, to the referential network that constitutes group meaning making. It identifies three phenomena — temporality, objectivity, and intersubjectivity — as the theoretical anchors of both the critical ethnographic tradition and the VMT Project's analytical practice. The final chapter is previewed as a programmatic statement about what a science of group cognition would look like and what the VMT Project contributes toward its foundation.
Meaning Making in VMT
This chapter offers the volume's most detailed analysis of meaning making as a collaborative interactional process. The chapter takes a three-minute excerpt from a VMT session — three students reflecting on mathematical patterns they have identified — and subjects it to fine-grained analysis of the referential structure of the participants' postings. The analysis identifies a complex network of references in which each posting draws on multiple prior resources: earlier chat postings, whiteboard inscriptions, mathematical concepts, and elements of the problem statement. This referential network is shown to be the concrete structure of the group's shared meaning making. The chapter then derives a set of conditions and preconditions for this kind of high-order collaborative cognitive work: both immediate conditions (that the group has an established indexical ground and a joint problem space) and broader preconditions (that the design of the environment supports the persistence and accessibility of relevant resources). The chapter argues that understanding these conditions is essential both for designing effective CSCL settings and for developing a science of group cognition.
Critical Ethnography in the VMT Project
This chapter argues that the VMT Project's research practice can be understood not only as design-based research but as a form of critical ethnography. The chapter traces the history of critical ethnography through two generations of thought rooted in critical social theory, showing how this tradition developed its distinctive focus on temporality, objectification, and intersubjectivity as the three key phenomena of social life. It then shows that VMT analyses — particularly those examining how groups construct shared mathematical objects through interaction, how groups establish a temporal dimension in their joint problem space, and how groups achieve intersubjective understanding — exemplify the analytical concerns of critical ethnography. The chapter also develops the parallel between critical ethnography's commitment to studying culture in order to support human emancipation and design-based research's commitment to studying interaction in order to improve educational practice. The chapter includes an extended illustration of critical ethnographic practice drawn from research on AI-system design.
Toward a Science of Group Cognition
This long concluding chapter steps back from individual analyses to reflect on what the VMT Project as a whole contributes to a potential new scientific discipline: a science of group cognition. The chapter argues that existing sciences of the mind — cognitive psychology, cognitive neuroscience, educational psychology — focus on the individual, while social theories focus on communities and institutions; neither addresses the distinctive processes of small-group interaction in which collaborative knowledge building actually occurs. The VMT Project is positioned as a set of preliminary explorations for this needed science. Three tasks for founding a new science are identified: defining and capturing the domain in the form of a data corpus; selecting, adapting, and refining methods for analyzing that data; and organizing findings in a theoretical framework. The chapter reviews how VMT approached all three tasks — creating the VMT service as a data-collection testbed, developing chat interaction analysis as a rigorous empirical method, and elaborating group cognition as the theoretical framework — and closes with a programmatic statement about the directions in which this work must continue.