(PDF) Sandboxes for Model-Based Inquiry

J Sci Educ Technol (2015) 24:265–286 DOI 10.1007/s10956-014-9506-8 Sandboxes for Model-Based Inquiry Corey Brady • Nathan Holbert • Firat Soylu • Michael Novak • Uri Wilensky Published online: 13 July 2014 Springer Science+Business Media New York 2014 Abstract In this article, we introduce a class of con- Introduction structionist learning environments that we call Emergent Systems Sandboxes (ESSs), which have served as a cen- In this article, we describe and illustrate a novel con- terpiece of our recent work in developing curriculum to structionist (Kafai 2006; Papert and Harel 1991) approach support scalable model-based learning in classroom set- to model-based inquiry (Buckley et al. 2004; Lehrer and tings. ESSs are a carefully specified form of virtual con- Schauble 2006; Windschitl et al. 2008). Specifically, we struction environment that support students in creating, introduce a class of learning environments that we call exploring, and sharing computational models of dynamic Emergent Systems Sandboxes (ESSs), which have served as systems that exhibit emergent phenomena. They provide a centerpiece of our recent work in developing curriculum learners with ‘‘entity’’-level construction primitives that to support scalable model-based learning in classroom reflect an underlying scientific model. These primitives can settings. ESS environments are a carefully specified form be directly ‘‘painted’’ into a sandbox space, where they can of construction environments that support students in cre- then be combined, arranged, and manipulated to construct ating, exploring, and sharing virtual models of dynamic complex systems and explore the emergent properties of systems that exhibit emergent phenomena. They differ those systems. We argue that ESSs offer a means of from computational tools used in other model-based design addressing some of the key barriers to adopting rich, con- work, both in terms of their affordances for construction structionist model-based inquiry approaches in science and the ways we use them to structure inquiry-driven classrooms at scale. Situating the ESS in a large-scale explorations for students. We present our ESS construct as science modeling curriculum we are implementing across a response to the Special Issue’s question, ‘‘How can the USA, we describe how the unique ‘‘entity-level’’ technology transform teaching and learning as students primitive design of an ESS facilitates knowledge system develop and use models?’’ and we argue that ESSs offer a refinement at both an individual and social level, we means of addressing some of the key barriers to introduc- describe how it supports flexible modeling practices by ing and pursuing rich model-based inquiry approaches in providing both continuous and discrete modes of executa- science classrooms at scale. bility, and we illustrate how it offers students a variety of We begin by indicating the research context for our opportunities for validating their qualitative understandings design work. We then provide a preliminary definition of of emergent systems as they develop. an ESS focused on the construction primitives it offers to the learner and the relation of these primitives to core Keywords Constructionism Design Agent-based disciplinary knowledge structures; the means it offers users modeling Scalability to run their constructions in various ways; and its use of saving state to support individual and social exploration of the behaviors of systems constructed by learners from its C. Brady (&) N. Holbert F. Soylu M. Novak U. Wilensky primitives. Next, we describe our theoretical framework, Northwestern University, 2120 Campus Drive, Evanston, IL 60208, USA which attends to the individual and social dimensions of e-mail:

our design of the ESS. We then proceed to offer a more 123 266 J Sci Educ Technol (2015) 24:265–286 elaborated description of the ESS construct and of the An extensive history of design research on construc- specific ESS we developed for our Particulate Nature of tionist curriculum in the domain of Chemistry and the Ideal Matter (PNoM) unit. Finally, rooted in this description, we Gas Laws provides a foundation for our PNoM work. Early discuss three features of ESS environments and of the work with an agent-based environment to explore the classroom interactions they foster, which have emerged in PNoM, called GasLab (Wilensky 1999b, 2003) showed the the course of our iterative design and implementation work learning potential of engaging students in developing their with this unit. own computational simulations of phenomena related to the behavior of ideal gases. Here, the process of con- structing a computer model became a process of debugging Research Context not only one’s code but also one’s conceptions of particle behaviors and the implications of these behaviors on Our development of the ESS design construct has emerged aggregate phenomena such as temperature, pressure, and in the context of our work on the ModelSim project (NSF# volume (Wilensky 2003). DRL-1020101). A core objective of this project is to Later work associated with the Connected Chemistry investigate the scalability of three innovative approaches to curriculum (Wilensky et al. 2004; Levy et al. 2006; Stieff science learning supported by the NetLogo environment and Wilensky 2003) sought to scale the GasLab research in (Wilensky 1999a). These approaches are as follows: (1) our first, numeric sense. It provided a more structured model-based inquiry with agent-based modeling (ABM), in sequence of investigations that were experienced by a which scientific phenomena are viewed as emergent, sys- much larger group of students across a variety of school tem-level behaviors and studied through computational and classroom settings, as a part of the Modeling Across simulations; (2) participatory simulations (PartSims), in the Curriculum project (Gobert et al. 2003; Levy and Wi- which classroom groups engage in systems simulations by lensky 2009a, b; Levy et al. 2006; Stieff and Wilensky taking on the role of agents in the system to experience 2003). Although Connected Chemistry was quite success- emergent phenomena firsthand; and (3) bifocal modeling, ful, the roles of construction and programming in that in which students connect virtual agent-based models with curriculum were substantially reduced from the original the physical world by taking in data streams from sensors GasLab studies. Nevertheless, an extension activity within and by generating physical behaviors through motors and Connected Chemistry did provide an open, exploratory other outputs. Each of these modalities has been developed environment, called a ‘‘Particle Sandbox,’’ which in fact over a long history of design research. The ModelSim served as the initial inspiration for our PNoM work. Our project investigates their systematic combination and use in ModelSim design objectives with respect to this first extended curricular units (2 weeks, 10–12 of hours of class dimension of scaling are thus to build on GasLab and time) treating core science topics at the high school level: Connected Chemistry, continuing the pursuit of ways to Population Dynamics; Evolution; Electricity; and the engage students in open-ended construction tasks and to PNoM. The project aims to refine and study both the assess the viability of these tasks in learning environments technological and pedagogical supports for these three beyond the controlled settings of researcher-supported modalities from perspectives of scalability. laboratory studies. In this article, we focus on work connected with the A second dimension of scalability has also proved PNoM unit of the project and especially with the first week important to our learning design work in the ModelSim of that 2-week unit. We do this in order to describe the ESS project: scaling from the individual learner to the class- construct in a particular curricular context that illustrates room group. An innovation is scalable in this sense if it key aspects of the ModelSim project’s conception of sca- makes effective use of the social resources and structures lability. Specifically, we identify two dimensions of sca- of classrooms, providing a basis for the construction of lability that we have considered in our design. shared collective understanding. In particular, such class- The first dimension of scalability deals with the size of room-level scalability can be achieved when students implementations, as a matter of numbers. In this sense, one of experience each others’ work as comprehensible and rele- the challenges of scaling innovative curriculum involves rep- vant to their own, and as building toward shared or com- licating essential features of learning environments developed plementary goals. When this occurs, students act as an in laboratory studies in more complex classroom settings. An engaged and authentic, critical audience for each other’s objective of ModelSim along this dimension is to support work. Such settings also encourage an active facilitation widespread, effective use of construction tasks that are critical role for the teacher, to support students in making the most for students to develop deep, mechanistic understandings of of the findings of their classmates and of the group as a the emergent systems explored in each of our units. whole. 123 J Sci Educ Technol (2015) 24:265–286 267 The ESS contribution we describe in this paper is a type 1. It offers its construction primitives at the ‘‘entity’’ of learning environment that supports this dual sense of level, where we define entities as agents-with-fixed- scale for model-based inquiry. Because the success of the behaviors, and where these fixed behaviors are ModelSim project depends upon our addressing scale at governed by a core scientific model that underlies the both of these levels, the affordances of ESSs will be construction environment. measured by their ability to facilitate this scaling. For 2. It allows flexible execution of constructions, enabling instance, in the current academic year we are implementing users to run them both continuously as they build and our curricular units in approximately 100 classrooms, and discretely as coherent runs that produce outcomes. as will be clear in the discussion below, effective small- 3. It supports saving and sharing of states of construc- group and whole-class interactions are essential to the tions, facilitating iterative experimentation, and shar- successful functioning of the units. Thus, two scalability- ing of in-process artifacts with peers. related design requirements for the ESS are: In later sections of the article, we will unpack this def- 1. It must permit students to engage deeply in individual, inition further, showing how we applied the construct to personally meaningful model construction and model- create the particular ESS we used in the introductory week debugging processes, and of our PNoM unit and then illustrating the affordances of 2. It must permit students to share their constructions that ESSs through the work of students and classroom with classmates and to benefit from interacting with groups who engaged with it to build and explore models of and discussing these artifacts, as an integral component diffusion. of their learning during the construction process. These two requirements stem from the numeric and Theoretical Framework classroom-level dimensions of scaling, respectively. And though they can be considered independently, we will show A wide range of research in mathematics and science how the ESS serves to connect these two dimensions. education has indicated the power of modeling activities, Indeed, the ESS works to bridge individual and social both in illuminating student thinking and in producing aspects of modeling, using each as a means of addressing conceptual change. Specifically, researchers have shown challenges associated with the other. that model-based learning approaches can support content mastery (Stewart et al. 2005), competence, and fluency with disciplinary practices such as argumentation (Pass- A Preliminary Definition of an ESS more and Svoboda 2012) and assimilation of meta- knowledge required to engage in these practices in appro- With the context and design challenges of ModelSim in priate ways as a part of authentic inquiry (Schwarz et al. mind, we can offer a preliminary definition of an ESS. An 2009). Moreover, the Next Generation Science Standards ESS is first and foremost an agent-based computational (NGSS) situate developing and using models as one of the modeling environment for creating and exploring emergent eight core scientific practices. In particular, these standards complex systems. In an agent-based approach to emer- suggest that models should be developed ‘‘to predict and gence, rather than describing and measuring phenomena at show relationships among variables between systems and the aggregate level (using differential equations or a sys- their components in the natural and designed worlds’’ tems dynamics model with ‘‘stocks’’ and ‘‘flows’’), one (NGSS Lead States 2013). instead conceives of these aggregate-level phenomena as emerging from the interactions of many individual, autonomous ‘‘agents.’’ ABM, then, is about attempting to Supporting Individual Construction identify the individual agents of a system along with the behaviors and rules that these agents follow that will lead Model-based inquiry takes as a fundamental premise that in to the emergence of target phenomena at the aggregate most fields, including the sciences, experts are distin- level. An agent-based perspective has been shown to be a guished from novices not only by what they know, but also powerful approach for explaining and understanding phe- by the ways in which they perceive situations (Glaser and nomena across a wide range of domains, including the Chi 1988; Newell and Simon 1972; Simon and Chase natural sciences (Abrahamson and Wilensky 2004; Blik- 1973; Lesh and Doerr 2003a, b; Lesh et al. 2008). Thus, stein and Wilensky 2004; Levy et al. 2006; Sengupta and developing expertise in a domain resides at least as much in Wilensky 2005; Wilkerson-Jerde and Wilensky 2010). building powerful interpretation systems (also known as With this foundation in ABM, an ESS has three key ‘‘models’’), as it does in accumulating mastery over col- properties: lections of facts and skills. This perspective helps to 123 268 J Sci Educ Technol (2015) 24:265–286 emphasize a view of learning that involves the construc- situational cues may lead to less-relevant resources taking tion, appropriation, and synthesis of new models, or the priority. In such instances, learners may offer explanations revision and expansion of the scope of locally applicable that would be categorized as ‘‘misconceptions,’’ regardless models to more general settings. of whether or not this explanation was productive (Smith This research perspective is particularly tuned to pro- et al. 1994). The goal of a knowledge-in-pieces approach, ducing descriptions of learners’ ways of thinking and then, is to help to refine the way learners interpret situa- carefully documenting changes in their ideas during tional cues so that they are more likely to activate pro- learning. Nevertheless, even among theoretical frameworks ductive knowledge resources. In this approach, intuition is that take student thinking as a central focus, there is a range valued and harnessed, rather than seen as a roadblock to be of potentially conflicting perspectives. For instance, his- removed. torically, many technological designs for STEM education To facilitate the refinement of learners’ knowledge have focused on identifying and correcting student mis- systems and to enhance their perception of important sit- conceptions. In these designs, researchers work to catalog uational cues, we adopt a constructionist approach in the common mistakes and misunderstandings about a given design of model-based inquiry environments (Papert and topic, to bring students into confrontation with their mis- Harel 1991; Papert 1980). Constructionist designs conceptions, and then to offer an expert explanation or empower learners to take charge of their own learning definition (Carey 1988; Clement 1982; Driver et al. 1994; through the construction of public artifacts that are per- McCloskey 1984; Posner et al. 1982; Tasar 2010; Trow- sonally meaningful. The act of construction, along with the bridge and McDermott 1980, 1981). While this strategy has process of sharing and critiquing these constructions, cat- proved enormously popular in educational design, it has for alyzes the refinement and reorganization of internal the most part proved ineffective, as studies have shown knowledge structures (Caperton 2010; Kafai 1995; Noss student misconceptions to be particularly ‘‘sticky’’ and and Hoyles 1996; Papert and Harel 1991; Sherin et al. resistant to change (see, e.g., McDermott 1983). 1993; White 1993; Wilensky and Reisman 2006; Wilensky One reason for this stickiness is that ‘‘misconceptions’’ 1996). Over the past 15 years, much work has been done of this kind are in fact often useful in everyday situations, exploring the potential of constructionist designs for providing effective guidance for ordinary action. For model-based inquiry (Abrahamson and Wilensky 2004; example, while it is true that in a frictionless world an Blikstein and Wilensky 2004, 2009; Levy et al. 2006; object in motion will remain in motion, real-world expe- Sengupta and Wilensky 2005, 2009; Stieff and Wilensky rience suggests objects in motion always slow down and 2003). eventually stop. For this reason, it is not surprising that In settings where computational media are available, students resist letting go of their ‘‘misconceptions’’ when there is a natural resonance between such constructionist ‘‘expert’’ explanations contradict thousands of observations model-based inquiry and various forms of computer pro- and experiences! Furthermore, while domain experts often gramming. Student-programmed computational models can hold rich and detailed theories about a phenomenon, nov- act effectively as ‘‘thought-revealing artifacts’’ (Lesh et al. ices tend to apply ideas or explanations in the moment, 2000), offering ‘‘both a model of student thinking about the depending on the ‘‘framing’’ of the specific situation (di- situation and a model that represents how the students have Sessa 1993, 1996; Hammer et al. 2005; Hammer 1996; integrated both interdisciplinary knowledge and the con- Sherin 2006). So while an explanation given by a novice straints and affordances of the problem context.’’ (Martin about a phenomenon may fit an identified ‘‘misconcep- et al. 2006, p. 389). In addition, the ability to easily ‘‘run’’ a tion,’’ the same novice may offer an expert explanation of computational model allows students themselves to quickly the same phenomenon in a different context (diSessa test and revise their constructions, further facilitating 1993). intrinsic motivation and learner self-assessment (Commit- In contrast to a misconceptions-centered approach, we tee for the Workshops on Computational Thinking 2010, have adopted a knowledge-in-pieces perspective for the 2011; Lesh and Doerr 2000, 2003a, b, 2012; Martin et al. design and analysis of our PNoM unit. This manifold 2006). model argues that cognition is emergent from many dis- Figure 1 offers a simplified, schematic account of the parate, low-level knowledge elements that are highly sen- modeling process in such a learning environment. The sitive to the learner’s interactions with their environment learner’s ideas develop through iterative cycles of con- (diSessa 1988, 1993; Minsky 1986). Which knowledge struction and interpretation, throughout the process of resources are activated depends greatly on the learner’s creating the computational artifact or program. Within perception of a given environment (diSessa and Sherin those cycles, running the program (the circular arrow) 1998). In some cases, relevant resources may be activated illuminates hidden consequences of the representation; leading to an expert-like explanation or response. In others, reflecting on it (the thought bubble) stimulates changes to 123 J Sci Educ Technol (2015) 24:265–286 269 consisting of one or more representational artifacts the learner has produced, along with the collection of her intentions, explanations, and perspectives toward those artifacts. At the same time that we are concerned directly with the emergence and refinement of this idiosyncratic sense- making process, we also must consider the models that have been created by the larger scientific community. The process of constructing a computational model involves an effort to capture the essence of a real-world system. For instance, in our PNoM unit, that system might be the phenomenon of diffusion or the behavior of a gas in a closed container. However, there are models from the Fig. 1 Modeling in a computational medium disciplinary community of Chemistry that are also designed to describe these systems. The Kinetic Molecular Theory (KMT) would be a prime example, along with its the learner’s internal conceptions; and construction activity quantitative entailments in the ideal gas laws. Thus, the (the double-arrow) is motivated by disjunctions between modeling process of our learner takes place in relation to the computational artifact and the learner’s emerging intent the modeling processes of a scientific community. or conception (which also includes the intended referent, We apply the word ‘‘model’’ to both of these categories not directly shown in the figure). of conceptual systems—the idiosyncratic and the official. Where, then, are the models in this picture? For con- But it is often important to distinguish between these two creteness, we borrow a definition of ‘‘model’’ from Lesh types as well, if only to aid investigation into the relations and Doerr (2003a, b): that students are able to construct between them. In this article, we designate the former class of models, devel- Models are conceptual systems (consisting of ele- oped and articulated by learners themselves, with the ments, relations, operations, and rules governing phrase ‘‘little-m’’ model. For instance, in our PNoM unit, interactions) that are expressed using external nota- an example little-m model might be one student’s tion systems, and that are used to construct, describe, expressions of contextual and case-specific understandings or explain the behaviors of other system(s)—perhaps of particle interactions in matter, and her speculations so that the other system can be manipulated or pre- about how observable phenomena might emerge from dicted intelligently (p. 10). these interactions. When referring to the latter class of In our case, the ‘‘external notation system’’ is the models, such as the KMT, we use the term ‘‘Big-M’’ programmable computational medium. However, this def- Model. In contrast with little-m models, these conceptual inition does not suggest that the construction process is structures are fundamental elements of the ‘‘paradigms’’ merely an externalization of a fixed internal conception. On (Kuhn 1970) that characterize the ways of interpreting and the contrary, the work of externalization is itself a conceptualizing the world which define entire scientific generative process—giving rise to changes in the learner’s disciplines (Fig. 2). conceptions. As such, the nature of the external represen- Little-m models can be thought of as personal hypoth- tation system used reflects back and has an effect on the eses or theories about how a system functions. While such nature of the conceptual systems of its author (diSessa models do not have the status of a Big-M model, to the 2000; Goody 1977; Wilensky and Papert 2006, 2010). We learner these models are salient and useful for interpreting also see modeling as a multi-faceted, social process, their immediate world. We view the interaction between encompassing the communication of emerging ideas, the models of these two types (Big-M and little-m) as a rich negotiation of overlaps and conflicts among these ideas, and important aspect of research into learning in model- and the integration of personal experiences with disciplin- based inquiry. In fact, a key element of the ESS hinges on ary forms of knowledge. Finally, at any given point in time, the ability to place little-m and Big-M models in closer the learner’s conceptions may not be complete, or the contact than indicated in Fig. 2. The ESS explicitly constructed external artifact may not adequate to articulate encodes a Big-M Model in the rules that govern the her conceptions. For all these reasons, we regard the behavior of virtual objects added to its world. Thus, the learner’s model as neither fully ‘‘in’’ the computational Big-M Model actively governs the running of computa- artifact nor fully ‘‘in’’ her head; rather, we view it as tional artifacts created in the ESS (Fig. 3). 123 270 J Sci Educ Technol (2015) 24:265–286 Fig. 2 An individual’s (little- m) model in relation to a discipline’s official (Big-M) model At the same time, a programming-centered approach to model-based inquiry may also raise issues along the classroom-level dimension of scalability. Extensive, indi- vidualized construction projects may be difficult to manage in classroom, and it may not be straightforward to ‘‘unpack’’ student programs for their instructional value in STEM learning. Fellow students (or teachers, or research- ers) who wish to understand the inner mechanisms of a text-based program must engage with it at the level of its code. This requires both a significant investment of time and a willingness to see the interpretive task of reading Fig. 3 Big-M and little-m models in an individual student’s construction with an ESS another’s code as relevant to learning in the content area. Moreover, in a programming-centered learning environ- ment, it may be difficult to engage students productively in Designing for Social Interactions in Classroom sharing and analyzing the interim versions of each others’ Learning computational artifacts. The activity of debugging code can be an extremely rich learning activity, since unexpected As we have seen, computer programming not only fits the behavior of a computational simulation often reflects demands of model-based inquiry, but it also offers a ‘‘bugs’’ at the level of thinking as well as at the level of powerful medium for supporting conceptual model devel- coding (Wilensky 2003); however, to engage in this opment and refinement as suggested in the constructionist activity, socially or collaboratively is a challenging prop- design paradigm. However, to the extent that this rich osition for classroom management. model of individual learning is pursued on its own, Our theoretical framework attends to the social dimen- potential issues can emerge along both of the dimensions of sion of constructionist learning environments by drawing ‘‘scale’’ that we have described above. For instance, the on the literature of generative design (Stroup et al. 2005, (numeric) scalability of learning designs that require 2007; Davis 2010), which takes inspiration from the teachers and students to become familiar with computer learning approaches of Wittrock et al. (Wittrock 1989, programming languages and their use may be questioned. 1992; Osborne and Wittrock 1985). In these approaches, This barrier can be particularly challenging to overcome in learning activities involve students in generating artifacts the context of classrooms where the primary subject area that reflect their emerging understanding. Stroup and col- being studied is not computer science (Committee for the leagues build upon this work, identifying a gap that is Workshops on Computational Thinking 2011; Guzdial closely related to our notion of classroom-level scaling. 1994; Sherin et al. 1993). Although there is important Specifically, they note that prior generative learning work ongoing research to address this very topic (Jona et al. has ‘‘underutilized the emergent space of behaviors and 2014; Trouille et al. 2013; Sengupta et al. 2013; Wilensky artifacts for classroom-based (group) learning and teach- 2014; Wilensky et al. 2014), this difficulty has also led ing’’ (Stroup et al. 2005, p. 191). In other words, students’ constructionist learning scientists to search for ways to independent work, considered as a collective whole, is seen lower the threshold to authentic computational modeling, as ‘‘space creating’’ in that it indicates the conceptual space for example, by using visual rather than text-based pro- of all potential responses. If this space can be visualized gramming tools (Sengupta et al. 2012; Wilkerson-Jerde and and discussed, it offers rich opportunities for collective Wilensky 2010; Wilkerson-Jerde 2012). discussion and reflection that prior designs did not exploit. 123 J Sci Educ Technol (2015) 24:265–286 271 Fig. 4 Pathways-and-endpoints diagram for a nominally generative task Fig. 6 Construction activity with an ESS that endpoint (the correct or acceptable method(s)). Stroup and colleagues describe such tasks as ‘‘nominally generative.’’ In contrast, design tasks, including the tasks common to constructionist learning environments, may admit a wide range of both pathways and endpoints (Fig. 5). Here, the challenge is to make the diversity of different students’ contributions and conceptions meaningful to one another so that they build to a collective understanding in which Fig. 5 Pathways-and-endpoints diagram for design tasks or con- ‘‘the whole is greater than the sum of its parts.’’ structionist activities With this pathways-and-endpoints framework in mind, we can articulate our strategies for increasing classroom- level scalability with the ESS. At the endpoint level, our In working to fill this gap, Stroup and colleagues noted a activity design provides students with a shared experience— resonance between their learning designs that engaged with in the case of PNoM, the diffusion of an odor through the classroom-level patterns in student work and the affor- classroom—which students seek to illuminate or explain dances of new classroom network technologies (Stroup with their ESS constructions. Because all students work et al. 2005; Stroup and Wilensky 2014). Such networks toward endpoints that are related through this shared repre- support a range of activity structures (Brady et al. 2013)— sentational goal, collections of student responses are neces- from PartSims (Wilensky and Stroup 1999b) (in which the sarily relevant to one another as different perspectives on entire class can enter a virtual world and interact as agents related facets of the phenomenon. At the pathway level, the in a shared simulation) to the sharing of rich artifacts ESS is designed to facilitate sharing of in-process artifacts. It designed by individual students offline in parallel con- does so both by enabling students to quickly make sense of struction work. In the ModelSim project, we have provided each other’s constructions (i.e., without grappling with them support for this range of activity structures, by using the at the level of computer code) and by making it easy to share HubNet architecture (Wilensky and Stroup 1999a) for artifacts-in-process to foster peer commenting and feedback. PartSims and by creating custom network technologies to In terms of the pathways-and-endpoints diagram, the ESS support flexible sharing and review of computational enables connectivity between pathways and the emergence artifacts. of a collective endpoint (Fig. 6). To document and conceptualize the range of generative Finally, we can show the effect of attending to the social activity structures, Stroup et al. (2007) developed a dimension in terms of our modeling diagram (Fig. 7). The ‘‘pathways-and-endpoints’’ task analysis framework that individual modeling process of Fig. 3 is here augmented by emphasizes the role assigned by an activity to the diversity a lateral, social process, driven by sharing and exploring of thinking in the classroom group. For instance, a tradi- the constructions of classmates. The act of sharing a con- tional, single-right-answer task such as the algebra problem struction itself stimulates further reflection, as Student 1 shown in Fig. 4 consists of a single endpoint (the correct begins to move from m1 to m10 . Student 2, who is con- answer) and a single pathway (or very few pathways) to currently at work on her own construction (reflecting on 123 272 J Sci Educ Technol (2015) 24:265–286 Fig. 7 Social dimension of modeling: students exchange in- process artifacts during construction m2), downloads and interacts with Student 1’s construc- based programming language. Thus, when users of Net- tion, forming a conception of it represented by m100 . Based Logo ‘‘talk to’’ the agents in their models, they do so by on her (public) comments, Student 1 and 2 together make issuing commands using this programming language. As possible a new conception (m10 ? m100 ). Moreover, as a such, the NetLogo building primitives exist at the pro- result of the experience of reviewing her classmate’s work, gramming language level, in the commands of the NetLogo Student 2 begins to think differently about her own con- language itself. By using primitives at this ‘‘low’’ level, struction, giving rise to m200 , which she will incorporate NetLogo offers its users an extremely versatile toolkit for into her future construction activity. modeling systems across many domains, and it enables its users to define agent-level behaviors with arbitrary com- plexity. (Wilensky 2003; Tisue and Wilensky 2004). An ESS for PNoM Moreover, learning environments created in NetLogo have the ‘‘glass box’’ feature that their mechanisms are always Earlier in the article, we defined an ESS as an agent-based available in the ‘‘Code tab’’ for interested learners. This computational modeling environment where construction provides an always-available route to ‘‘high ceiling’’ primitives have been designed at the ‘‘entity’’ level to inquiry to pursue questions about such simulation envi- reflect an underlying Big-M Model; where learners can ronments and even to modify or extend such environments. flexibly run their artifacts as they build them; and where Another class of modeling environments, including learners can also easily save and share successive states of Interactive Physics (Roth 1995) or the PhET Interactive their constructions. In this section, we will go into more Simulations (Wieman et al. 2008), typically offers detail about the specifics of these components of the ESS ‘‘higher’’ level primitives that focus construction activities definition and describe the particular ESS we produced for at the level of human-scale objects. Here, the construction the first week of our PNoM unit, the Diffusion Sandbox. primitives might include virtual tools such as springs or Every computational modeling environment provides its motors for kinematics-based environments, or switches, users with a particular set of primitives from which larger ammeters, and jumper wires for electricity-based environ- models and systems are constructed. The primitives used in ments. Such environments are not particularly suited to a modeling environment thus serve as a representational modeling emergence and complexity, but because the infrastructure for thinking through the components and human-scale object primitives are highly configurable, they mechanics of the models that are built in that environment. can offer a powerful means for students to explore how For this reason, in designing a modeling environment, the these concrete real-world scientific apparatus work and to form of its primitives and the particular level at which these easily construct a wide variety of scenarios involving the primitives exist are critical factors, which greatly deter- possible configurations of these objects. Even when the mine the kinds of thinking and building that users can programmatic mechanisms of these objects are black boxes engage in when using the environment. To illuminate the to learners, they are similar enough to laboratory apparatus design choice of defining the ESS primitives at the entity to invite further inquiry through exploring the functioning level, we give brief accounts of two other classes of of these physical devices. modeling environment that use different levels of primi- In contrast to environments that utilize programming tives, and we indicate the effects of those choices. language-level primitives and those that offer human-scale For example, in the NetLogo ABM environment, the objects, an ESS provides users with primitives at an properties and behaviors of agents are defined using a text- ‘‘entity’’ level. We define an ‘‘entity’’ as an agent-level 123 J Sci Educ Technol (2015) 24:265–286 273 object with fixed behaviors. Rather than allow users to environment and builds intuitions about the interactions of program the rules that define agents’ behaviors and inter- entities. In this mode, representations in the model are actions, in an ESS, these rules are predefined to encode the executed continuously: the simulation logic can be run logic defined by the scientific (Big-M) model. Users add, while the user adds and manipulates entities in the ESS. arrange, and combine these entities to create working Because in an ESS learners are building at the ‘‘entity’’ systems. level, rather than at the ‘‘language’’ level or at a higher, A consequence of designing primitives at this level is aggregate level, it is particularly important that they attend that the representational expressivity of an ESS is limited closely to the objects that compose a particular structure or to situations that can be adequately understood through phenomenon. A major task of modeling in an ESS is application of that Big-M Model. For instance, in our orchestrating situations for agent-level entities to interact, PNoM unit, the Big-M Model behind our Diffusion Mod- and so these interactions and their consequences become eling Sandbox ESS is the KMT. A student could attempt to salient for learners and are often the focus of iterative utilize this ESS to model a phenomenon for which the construction work. In an ESS environment, users do not KMT is not a useful model—say, snow falling in a forest. discover the rules that govern a complex system by creat- But such a student will find that his ‘‘snowflakes’’ travel ing and coding them; they do so by putting entities that forward at a uniform speed until they collide elastically encode these rules into contact with one another, observing with each other or with his ‘‘trees.’’ While this student may the results and making predictions about how the results be frustrated by the inability of this environment to produce might be different under alternative conditions. a satisfactory model of snowfall, he may actually learn In contrast to the immersive, continuous mode of something powerful about the KMT as a result of this interaction and execution, a second mode is external, ‘‘failure.’’ And herein lies one of the advantages of an ESS. comparative, and discrete. Here, the learner wishes to Because every entity added to its virtual world relentlessly observe the emergent behavior of a system that they have follows the rules of the Big-M Model, the consequences of constructed, to ensure alignment with their target phe- these rules tend to become salient to users who create nomenon. In this mode, it is important to pause execution, artifacts within the ESS. Moreover, because the rules are establish the arrangement and conditions of the dynamic not open to editing, a construction in an ESS will always entities, and then cause the environment to run for a period run, and it will always behave faithfully to the Big-M of time to test and revise conjectures. Moreover, because Model. It may not produce the aggregate behaviors that its the two activities of intuition-building (continuous execu- author intends, but it will always produce outcomes that are tion) and conjecture testing (discrete execution) are fluid, the logical consequence of the Big-M Model, gradually both modes must be present in an ESS and learners should nudging the learner’s intuitions into alignment. be able to quickly and easily transition between the two. The nature of an ESS’s primitives also distinguishes it Finally, the ESS’s facilities for saving and sharing the from ‘‘macro-level’’ modeling environments. The ESS’s states of constructions also follow from the nature of the entities exist at the agent level of the created system, and so primitives and the two interaction modes described above. the ESS is well suited to produce complex, emergent To support the discrete mode of interaction, it is vital that phenomena. Moreover, because these entities are not learners are able to establish ‘‘initial’’ conditions in the directly linked to human-scale objects, different students virtual world, ‘‘run’’ this construction, and then return to can make different choices about what human-scale objects the prior state to make modifications for further explora- they will build in their ESS constructions and how they will tions. Flexible saving of states supports discrete executa- build these objects out of the entity primitives. In spite of bility in enabling this iterative process. the fact that there are a limited number of entity types, all But state saving (and publishing) also enables an of them having fixed, predefined behaviors, these features entirely new dimension of exploration, at the social level. make it common for students to be surprised by each Because there are only a few entity types in an ESS, stu- others’ representational choices and intrigued by their dents find each others’ constructions comprehensible on ingenuity. Here again, the nature of the ESS primitives visual inspection. While surprise occurs as a result of dif- correlates with the learning affordances of the ferent choices of how to use a given entity to construct a environment. given aggregate object or phenomenon, by the very defi- The design and level of the primitives of an ESS also nition of the ESS, the entities themselves and their support two distinct and powerful modes of interaction behaviors are part of a shared lexicon. On the other hand, with the environment and enable the learner to move freely because these constructions exhibit complex behaviors, between these modes. One mode is immersive, ‘‘embod- students can gain significant insight into the nature of the ied’’ modeling (Wilensky and Reisman 2006), in which the entities and of the underlying Big-M Model by interacting learner explores what it is like to exist within the ESS with their peers’ constructions. Moreover, because students 123 274 J Sci Educ Technol (2015) 24:265–286 Fig. 8 Data visualizations from three moments in the whole-class odor-diffusion experience know their peers are building models of the same shared ‘‘entities’’ expanded. As you can see, the environment phenomenon as they are, they have an incentive to study provides a fairly open canvas for student construction, and respond to each others’ perspectives on the phenom- offering only a collection of green particles for a start (in enon, by downloading and interacting with their con- fact, even this initial feature can be removed or altered by structions. In this peer review process, both the continuous the learner, by adjusting the initial-#-particles slider and and discrete modes of execution gain another level of pressing the setup button). The set of construction tools utility, in supporting student’s understanding and assimi- simply suggests that student work could involve creating lation of ideas from each others’ little-m models in process. ‘‘walls’’ of various kinds (fixed and removable, of different In the remainder of this section, we advance the colors), particles of different colors, and sensors that detect description of the ESS construct by introducing the par- each of the different particle types. There are also tools for ticular ESS we developed for the PNoM unit of the Mod- changing particles’ color, for speeding up or slowing down elSim project. This ESS, called the Diffusion Modeling specified particles, and for removing entities that have been Sandbox, was used extensively in the first week of the unit. created in the sandbox. To add or modify entities in the On the first day of the unit, students engage as a whole- environment, learners simply select in the chooser the class group with an experience of odor-diffusion in their appropriate action and ‘‘paint’’ the objects or changes physical classroom. Two containers of perfume are intro- directly into the sandbox space. duced into the classroom space—one heated and the other Execution of constructions in the Diffusion Sandbox is at room temperature. The students are told that they will act controlled by the pause-particles? switch. When particles as smell sensors, recording the changing intensity of odor are unpaused, they immediately and continuously execute that they detect overtime. Knowing that the experiment can their rules of motion. Regardless of the state of the pause- only be run once (it is hard, if not impossible, to ‘‘reset’’ particles? switch, the learner can add, modify, or remove one’s nose or clear the room completely), the group dis- entities from the construction. This enables both continu- cusses how best to arrange themselves in the room, and ous and discrete interactions with the environment, as how to record their sense data. After running the experi- described above. Saving, sharing, and loading of states are ment, the class explores a visualization of the intensity- controlled by other buttons in the interface, which work in level data that they have recorded and discusses what they conjunction with a web-based Gallery that we constructed see (Fig. 8). These discussions typically involve identify- for the project, where learners can view, download, and ing patterns in the spatio-temporal spread of the scent, comment on each other’s posted states. observing fluctuations or exceptions in those patterns, In spite of the unstructured and open interface of the thinking about causal factors such as temperature, and Diffusion Modeling Sandbox, the system elements that the discussing various ideas about mechanisms that could learner can construct all behave according to rules that explain the spread of the odor. reflect the underlying Big-M Model of KMT. In particular, The ESS is introduced immediately after this inconclu- in the sandbox, particles collide elastically with each other sive but idea-generating whole-class discussion. Student and with walls of all kinds, and walls provide surfaces pairs are asked to explore the ESS environment and use it against which collisions are also elastic, and where to develop an expressive, executable model that explains reflections obey a rule equating angles of incidence and one or more aspects of the shared experience of the dif- reflection. Because these micro-, agent-level behaviors and fusion phenomenon. Figure 9 shows an image of the initial features are analogous to rules that learners observe in state of this ESS, with its chooser list of available primitive macro-level interactions (e.g., billiard balls and their 123 J Sci Educ Technol (2015) 24:265–286 275 Fig. 9 The initial state of the Diffusion Sandbox environment, and the construction primitives available to the builder behavior in bouncing off walls and each other), learners are the Diffusion Modeling Sandbox, showing how the design able to draw upon intuitions from their everyday lives in components of the ESS gave rise to key features of student working with these objects in the sandbox. At the same activity and interaction conducive to effective, scalable, time, the essence of the KMT is that a system of many model-based inquiry. agents obeying these simple and intuitive rules can give rise to emergent phenomena (Wilensky 2001)—including not only diffusion but also other aggregate-level properties Methods such as pressure and temperature. Students’ constructions within the sandbox environment require them to engage The data presented below are drawn from implementations deeply with the various aspects of this underlying Big-M conducted by two teachers that taught in two different high Model to achieve their personal little-m goals. Model schools, each serving a diverse population in the metro area development in this ESS environment is thus an entry-level of a large Midwestern city. Before implementing the PNoM process of discovering the expressive potential of the unit, these teachers were trained on the technology and agent-level components and their ability to account for and activity structures during a 2-day summer workshop. produce these aggregate phenomena. This construction Researchers were present during the implementation to offer process is interactive, as the Sandbox can provide contin- both technical and instructional support. The implementa- uous feedback to the learner by executing the agent-level tions described here occurred in two honors and one regular behaviors of the components that the learner constructs and chemistry class with students in their freshmen to junior year. arranges. It is iterative, as the learner can save states of the A large corpus of data was collected, including both artifact as it develops, to test the effect of changes in a whole-classroom and ‘‘roving’’ video data; answers to systematic way. And it is social, in that these states and the online questions before, after, and during the unit imple- modeling process as a whole occur in a classroom com- mentation; as well as computational artifacts and models munity of learners who are producing, sharing, and created by students during the unit. While the non-model- reviewing each others’ computational artifacts, which offer building activities of the unit certainly impacted the larger different perspectives on a phenomenon that the group has experience, in this paper we focus specifically on the ways experienced together. learners engaged the tools available in the ESSs, the arti- In the remainder of the article, we will describe and facts created within the ESS environments, and their con- analyze data from implementations of the PNoM unit with versations around these artifacts. 123 276 J Sci Educ Technol (2015) 24:265–286 Features of the ESS in Action that students often produced constructions that were like drawings, rather than immediately attempting to model the In this section, we describe three features of the experience target phenomenon of diffusion. While these drawings of constructing within an ESS, illustrating our discussion varied dramatically, most converged on images and sys- with examples from students’ use of the PNoM unit’s tems that included particulate phenomena such as ‘‘snow’’ Diffusion Modeling Sandbox. In our implementations, we (Fig. 10a). Though initially it might have seemed that have found that the ESS provides students with a refer- learners were neglecting ‘‘more important’’ work in creat- entially underdetermined space for exploration; that it ing such drawings, we came to recognize the importance of supports flexible modeling through its dual modes of this ‘‘messing about’’ (Hawkins 1974) phase. Indeed, dur- continuous and discrete executability; and that it affords ing this time, learners began to develop their intuitions students a variety of opportunities for validating their about how the objects in the sandbox world could be qualitative understandings of emergent systems as these manipulated. Their ideas were reflected in the particular understandings develop. phenomena they chose and the ways they used ‘‘anima- tion’’ in their drawings. As intuitions about virtual objects Opening a Referentially Underdetermined Space and their behaviors grew, this in turn increased students’ for Exploration readiness to employ the virtual objects as referents. By the end of the first day, while students may still have been The construction primitives of an ESS are intentionally ‘‘drawing,’’ these drawings began to take on the features designed to be referentially underdetermined, in the sense and characteristics of objects relevant to the phenomena that they do not map singly or obviously to apparatus in the under study, such as including the layout and features of observable world. For one thing, they exist at a lower, more the classroom (Fig. 10b). fundamental level than things such as desks, flasks, or In an ESS, this ‘‘seeing-as’’ step—seeing a rectangular perfume. Thus, macro-level, human-scale objects must be arrangement of wall patches as the perfume container in a built out of aggregations of these entities. But there is also shared diffusion experiment, for example—is initially left nothing in the ESS that predetermines the mappings up to the learner and is a critical part of the construction between the entity types and macro-level objects. The process. As learners begin to see the ESS as a medium in names, such as ‘‘walls’’ and ‘‘particles,’’ do provide sug- which to simulate scientific phenomena, they begin to tune gestive guidance, but learners are free to explore possible their constructions toward specific questions (or factors) mappings. about how these phenomena unfold. Often students first Designing an ESS to be referentially underdetermined define one object or construction as a central figure (such as engages the learner in two important and linked layers of drawing a box out of wall entities and deciding that it interpretation that drive the modeling process. In the first represents a container of perfume) and then gradually apply layer, learners work to apply meaning to the individual a set of references to objects in relation to that first map- objects and entities; and in the second, learners try to make ping. While some students may exhibit a literalism in their sense of the behaviors of these objects, in terms of their constructions, such as reproducing the classroom along referents. So while learners work to identify what the with the rows of lab-tables (Fig. 10b), others abandon strict ‘‘green particles’’ could represent, they must also make literalism in an attempt to create conditions where a par- sense of how and why these particles move in the way they ticular mechanism or outcome will show itself most do. Why do they move in straight lines? Why do they clearly. Some of these ‘‘schematic’’ designs suggest bounce and reflect off of walls? As learners begin to apply attempts to isolate variables and test-specific experimental a referent to objects in the model, they bring with this conditions, such as exploring the differences between dif- referent a certain set of possible behaviors. Likewise, as fusion with and without air present (Fig. 11). learners interpret behaviors, a certain set of objects become While learners individually define the mappings possible referents and others are excluded. These two between ESS entities and what these entities come to sense-making activities occur simultaneously and are represent, over the course of an implementation whole complementary, in that each act supports and constrains the classes begin to develop a shared understanding and other; and they are central in helping the learner to connect expectations of what ESS entities are and can be. For their personal little-m model with the Big-M Model tar- example, classroom groups may coalesce around particular geted by the designer. strategies for representing macro-level objects such as To encourage students to explore the space of possible ‘‘containers’’ and ‘‘desks,’’ or they may begin to develop references in the Diffusion Sandbox Model, we explicitly classroom norms such as ‘‘we usually use green particles to incorporated free exploration time into our PNoM imple- represent air.’’ While each of these particular examples has mentations. During the first day of explorations, we found been observed in our PNoM implementations, due to the 123 J Sci Educ Technol (2015) 24:265–286 277 Fig. 10 Early explorations included drawings of interesting pictures or representation of particulate phenomena such as ‘‘snow’’ (a). Later constructions more literally modeled the room where the diffusion phenomena took place (b) Fig. 11 Examples of a schematic and experimental approach—exploring how particles would diffuse with (b) and without (a) air ambiguity of the objects and entities that can be added in Because students are encouraged to post in-progress model the Diffusion Sandbox Model, we have found a great designs frequently to the gallery and are explicitly directed diversity between different classes. to explore the models created by their classmates, students This emergence of a common language and expectations must necessarily read and interpret the models of others. around referents shared generally by each class as a whole Making sense of others’ models facilitates a shared is an effect of the social dimension of the ESS, which understanding of entity referents. As students begin to promotes communication about ideas that the group feels comment on or discuss others’ models, they freely apply are important: in this case the normative mappings of the their own mapping to entities and objects in models created environment’s ‘‘underdetermined’’ entity primitives. To by others. For example, after observing and running facilitate multiple rounds of exploration and design, as well another student’s construction, one student posted a com- as to encourage students to share and collaborate during ment on the Gallery to the model author stating, ‘‘I think model development, we developed the means for students you should have made the heated molecules and cold to ‘‘go public’’ with their discoveries by publishing the molecules separate colors. Also, I like how you incorpo- state of their ESS worlds to a shared ‘‘Gallery’’ at any time rated air molecules.’’ While the model authors did not they chose. Once a world-state is posted, anyone in the explicitly state what each object or entity in the model was class can view a snapshot of the world, comment on it, or meant to represent, the commenting student immediately load the full executable state into their own sandbox mapped ‘‘heated’’ and ‘‘cold molecules’’ to particular environment for testing, exploration, or refinement. groupings of circles present in the model and also assumed 123 278 J Sci Educ Technol (2015) 24:265–286 the circles distributed throughout the model must be ‘‘air constructed using entity-level primitives, their meaning and molecules.’’ For students that had not considered how the function exist at a higher ‘‘object level.’’ presence of air particles might impact the modeled phe- While almost all computer programming and modeling nomena, this mapping suggested by another student may environments include executability of some sort as a core seem reasonable and lead them to incorporate this feature feature, the nature and the degree of this executability are in their own models (doing so on the assumption that this significant points in the design of such environments. interpretation of green-circles-as-air-particles is correct), Moreover, different types and levels of executability may further facilitating the spread of this particular mapping. be appropriate for different modes of exploration or dif- By designing primitives to be representationally under- ferent phases in model construction. In a similar way, determined, we have also created an opportunity for stu- recent trends in computing that offer a more fluid interac- dents to be surprised by how others might use an entity in tion between the programmer and the programmed envi- an innovative way. While comments similar to the one ronment, sometimes known under the heading of ‘‘live cited above certainly could occur in other modeling envi- coding’’ (Burg et al. 2013), suggest a new, more dynamic ronments, the entity level at which ESS building primitives relationship between programmer and program that may exist greatly determines the form and purpose of these augment and complement the more traditional ‘‘express- exchanges. For example, when engaged in modeling at the test-revise’’ (Martin et al. 2006) debug cycle common to language level, one may expect conversations and com- most existing programming environments. In describing ments around others’ constructions to focus on algorithmic the executable representations available within an ESS, we or behavioral rule-level choices, such as the decision to use highlight the advantages of both a continuous and a dis- fixed directions for a particle’s motion rather than incor- crete mode of execution. porating an element of randomness. Alternatively, when The continuous mode of executability is akin to the building in an environment such as Interactive Physics, the ‘‘live coding’’ style described above. And though live macro-nature of the building primitives may instead give coding is a fairly novel affordance in computer program- rise to discussions that center around the relative correct- ming environments, continuously executable representa- ness or the effects of connecting various macro-level tions have been studied in other areas of the learning objects with particular configuration settings, similar to sciences for over two decades. For instance, in the research how one might discuss the appropriateness of a particular literature on mathematics learning, the value of such rep- equipment assembly in a laboratory setting. In contrast, the resentations was explored in the 1990s as access to com- use of referentially underdetermined entities allows stu- putational power sufficient to support real-time dents to attend to each others’ reference mappings and the executability ushered in what Balacheff and Kaput effects of these choices: both to recognize mappings and to described as ‘‘a new experiential mathematical realism’’ be surprised by them. This shifts the focus of discussions (1997, p. 470). Here, the term ‘‘realism’’ refers to the from being about either the nuances of programming or the reification of mathematical constructs as virtual objects appropriate use of scientific apparatus to center on both the that become manipulable (e.g., through interaction with novelty of agent-to-aggregate constructions and the value ‘‘hotspots’’) while obeying the structural constraints and of particular entity mappings. relations that define them. The 1990s witnessed the growth of dynamic mathematics software tools such as Kaput’s Supporting Flexible Model Construction own SimCalc MathWorlds (Kaput and Roschelle 1996) with Continuous and Discrete Executability environment for the study of the math of change and var- iation; the Geometric Supposer (Schwartz and Yerushalmy To create experiences that both facilitate intuition-building 1987), Cabri Geometry (Laborde 1990), and the Geome- and encourage learners to move toward hypothesis gener- ter’s Sketchpad (Jackiw 1991) for dynamic geometry; and ation, ESSs allow learners to build constructions that are Fathom (Finzer et al. 2002) and later TinkerPlots (Konold both continuously and discretely executable. Executable and Miller 2005) for the dynamic study of statistics, representations in computational media are designs that probability, and data modeling. exhibit dynamic behavior in response to logic encoded in An ESS enables this type of ‘‘experiential realism’’ in their construction. They thus ‘‘run’’ semi-independently of modeling by supporting a construction mode in which their human authors, which allows those authors to reflect entities move and interact while the learner is in the process on their behavior and/or to interact with them as they run. of building. As such, an ESS has some key features in For instance, an executable representation in the Diffusion common with dynamic mathematics software and espe- Sandbox Model might be a closed container that only cially with modern dynamic geometry environments. In releases particles when triggered to open, or a vent that particular, both support the construction of systems of heats and pushes particles as they approach. While each is virtual objects and relations whose dynamic interactions 123 J Sci Educ Technol (2015) 24:265–286 279 obey the constraints and specifications of an underlying observe results. To support this shift, our ESS makes it Big-M Model. The fluid exchanges between learner and possible not only to pause the dynamic running of the environment enabled by such a design permit constructive environment, but also to save the state of the representa- modeling processes characterized by what Moreno-Arm- tional world and to reload this state at a later time. This ella et al. call cognitive partnership (Moreno-Armella and enables learners to design and develop experiments that Sriraman 2005) and co-action (Moreno-Armella and carry their intuitive understandings to the next level and Hegedus 2009). According to this perspective, the direct produce more robust and shareable results. While ‘‘present- engagement with objects that embody the rules and struc- tense’’ thinking explores the nature of agent-level interac- tures of a discipline offers learners a continuous, dialogic tions in an ESS, this new ‘‘future-tense’’ style can involve relation with those knowledge structures. In this way, live observing and reasoning about the aggregated effects of coding in computer science, hotspot dragging in dynamic these interactions overtime. Moreover, this discrete mode geometry software, and systems construction in an ESS all of execution is particularly powerful along the social offer exciting possibilities for co-action in computationally dimension of modeling. As learners seek to share their mediated environments. These possibilities hinge on the findings with others (or even to align their interpretations creation of a virtual space that affords the user continuous of virtual phenomena within a single working group), they interactions with dynamic objects in a world where disci- are led to formalize their observations and move toward a plinary structures define the phenomenology of those future-tense style, supported in this effort by the ability to interactions. save, load, and share world states publicly in the Gallery. In our work with ESS environments, we have observed Both continuous and discrete modes of executability are the power of such spaces for facilitating student sense- critical to the investigations and constructions facilitated by making, perhaps most notably in the early stages of our PNoM ESS. Continuous executability supports the exploratory use. In these early interactions, learners rapid growth of students’ intuitions both about the Big-M attempt to gain an intuitive understanding of the way the Model of the KMT and about the little-m models reflected virtual world works. Papert (1980) described such activ- in their classmates’ constructions. And as understanding ities as ‘‘getting to know’’ a virtual world and argued for grows, discrete executability enables students to articulate the potential of even text-input-driven microworlds for and share predictions about the aggregate-level behavior of building this type of qualitative understanding of the their virtual constructions, substantiating them through the world’s structures (e.g., p. 137). While we believe it is design of informal but replicable experiments. certainly possible to engage in this intuition-building outside of a continuous execution environment, we argue Offering the Means for Validating Qualitative that the fluid interactions facilitated by such settings are Understandings particularly resonant with this intuition-building objective and that therefore the ESS environment’s support of this A third key feature of ESS environments is that they offer mode is a strong affordance. Finally, while we have learners a variety of opportunities to seek and receive highlighted the value of continuous executability here in validation for the little-m models expressed in their con- the context of individual construction, it has a significant structions, as they develop. An important concern with value for the social dimension as well. The ability to open constructionist learning designs is the question of inhabit and explore a fellow student’s construction offers whether, in the language of this article, students’ little-m a compelling incentive to engage in peer review, and the models eventually converge on the Big-M Model that vividness of the experience encourages students to represents normative disciplinary understandings of key appropriate key elements of the constructions that they phenomena. There are many approaches to ensuring such explore in this way. convergence, but the mechanism we emphasize in our ESS An ESS also supports another key mode of executabil- design is that of external validation. That is, providing ity, the discrete mode, which is conducive to investigations occasions for subjecting the qualitative impressions that of another kind. In general, we observe students’ interac- learners have about the emergent behaviors of their com- tions with ESS environments to move from what we call putational artifacts to increasingly rigorous standards of ‘‘present-tense’’ focused work (e.g., ‘‘when I do X, Y evidence. In this section, we describe two means by which happens’’), which is supported well with continuous ex- students can receive such validation: (1) adding mecha- ecutability, to ‘‘subjunctive’’ or ‘‘future-tense’’ focused nisms to their constructions that produce quantitative work (e.g., ‘‘under condition X, Y will happen’’), which measurements of phenomena, and (2) negotiating meaning requires a more punctuated, discrete form of execution. and resolving conflicting ideas in a social space, among Investigating the behavior of a system at this level requires constructions identified as describing the same or over- the ability to control conditions, repeat experiments, and lapping phenomena. 123 280 J Sci Educ Technol (2015) 24:265–286 The first means of validation, adding quantitative mea- work support an increased sense of need to articulate their surements, is conducive to the kind of ‘‘future-tense’’ findings clearly and precisely, as their fellow students thinking described at the end of the previous section. An constitute an engaged and authentic audience for these ESS can provide students the ability to instrument their findings. virtual environments in various ways; in the PNoM ESS, The social dimension of the PNoM ESS modeling this includes adding ‘‘particle detectors’’ of different kinds environment in fact has a prominent role from the very or measuring changes in selected virtual objects or regions. beginning of the unit, when students act as ‘‘smell sensors’’ By introducing and arranging these sources of quantitative to produce a collective representation of the diffusion data, students can substantiate qualitative impressions that phenomenon that the classroom can continually refer to they have about emergent properties of their constructions. and reflect on throughout the course of the unit (Fig. 8). In Moreover, the ESS provides the ability to plot some of the visualization produced by this activity, each ‘‘dot’’ these data overtime. In combination with the ability to save represents the data collected by a single student. However, and reload states, these quantification features support the significance of these data is only recognized in the learners in quantifying observations, substantiating asser- emergent patterns that appear among the dots, both in tions, and validating predictions about their virtual exper- space and overtime. Thus, the ‘‘agent-level’’ testimony of iments. For example, after adding virtual sensors to her the individual student must be placed in a social context for ESS, one student commented: it to have meaning. Moreover, students’ engagement with the ESS environment is explicitly framed as an effort to Increase in temperature helps speed up molecule investigate, reproduce, or explain some features of this diffusion. The peppermint at the front of the room shared physical experience in their virtual sandboxes. In (which was heated) was detected faster and a lot more this way, students’ independent investigations are implic- strongly than the peppermint at the back of the room itly linked in a communal effort to understand diffusion in (lower temperature), which was not detected at all. general and the particular phenomena that the class has The particles that were sped up were detected more experienced as a group. by the sensors (purple in our case), and the slower To build upon this social foundation of inquiry, students particles not as much. Particles that were sped up are encouraged to post works-in-progress and completed bounced around more and passed over the sensors models in the ESS Gallery (as described in the section on more often. Particles that had a higher temp bounced the referentially underdetermined feature of the ESS) as around more and with more force thus giving them a well as to run and comment on the models posted by better chance of being detected by the sensors. others. We argue that the existence of a ready-to-hand Here, the student’s comments blend claims about real- means for ‘‘going public’’ provides an authentic audience world diffusion phenomena with evidence and argumen- for the students’ work and applies an appropriate pressure tation about their virtual construction. The conflation of to articulate their intuitive understanding of the model in statements about physical and virtual worlds suggests that more formal ways—specifically, as observations and she has begun to see the mechanisms of her virtual con- communicable findings. Furthermore, due to the entity- struction as offering a (little-m) model of the physical level primitives used in ESS models, students can easily world experience. The sensor readings over multiple runs read and interpret classmates’ models without needing to of her construction have provided her with a warrant for make sense of complex computer code. By removing these her claims about the relation between temperature and roadblocks to understanding, students are free to attend to diffusion rates. the common or distinctive elements among their class- A second means of external validation in the ESS arises mates’ constructions. In our implementations, we found in the social space. At a basic level, students are aware that evidence of learners comparing and contrasting various their construction efforts are related to one another, since designs of virtual experiments in written comments posted they are all attempting to create models of an aspect of the to the gallery and during whole-class discussions, and we shared diffusion experience. Moreover, because students observed spontaneous identification of shared lines of publicly post their works-in-progress, they are able to inquiry emerging among multiple student groups working identify relations and overlaps between their construction in parallel. Such evidence suggests that this social dimen- project and those of their classmates. In terms of the sion was an important aspect of learners continued refine- pathways-and-endpoints diagram, the shared experience ment of their little-m models. promotes connectivity between the endpoints of the stu- When commenting on others’ models in the gallery, we dents’ construction efforts, while the ease of sharing sup- found that students often offered endorsements or critiques ports students in identifying affiliated work that their peers of explorations or experimental designs by directly com- are doing. Both forms of connectivity between students’ paring the posted model to mechanisms and designs found 123 J Sci Educ Technol (2015) 24:265–286 281 Fig. 12 Posting to the diffusion gallery Fig. 13 Model produced by the author of the first comment on Fig. 12, and comments on his model from two other classmates in the author’s own work. For example, the first of the Thus, this group’s comment in Fig. 12 acknowledges comments seen in Fig. 12 reads, ‘‘I am kind of confused as the value of an alternative perspective on the question of to why your diffusion starting points are in the corners and air molecules, but maintains a critical stance on the ques- why the sensors are not together. However, good job tion of sensor positions. Because this commentary occurs implementing the air molecules.’’ Behind this comment in a public space, however, the conversation continues. stands a reflective and discursive history. The author of Subsequent commenters on the model in Fig. 12 pick up this comment was from a group that did not include air these themes, expressing different perspectives on sensor molecules in their representation and arranged their sen- placement, while also reinforcing the value of including air sors in a grid at the center of the virtual room. Comments molecules in the representation: on their posting, from still other groups, had critiqued their Your experiment was very different than ours. I liked orderly, centralized sensor arrangement and had also how you placed the sensors all in different places to noted: ‘‘…you forgot to account for the molecules in the show where the particles reached most…. air’’ (Fig. 13). 123 282 J Sci Educ Technol (2015) 24:265–286 Fig. 14 Introducing structures within the perfume containers to slow the escape of particles Fig. 15 Constructing barriers to slow the spread of particles I like how the areas containing the purple and orange The existence of an ongoing channel for communication particles are the same size. I also like how your between the groups and the ability to survey classmates’ model contains what looks to be air molecules. strategies-in-process seems also to have encouraged the spontaneous formation of groups with shared research Later, in the teacher-facilitated, whole-class discussion of interests and lines of inquiry. For instance, a number of the models, issues associated with including or omitting air students in this class became interested in or concerned molecules were raised as well, with different student about the rapidity of diffusion in the virtual world and groups explaining their decisions and the impact they felt wanted to introduce mechanisms that would produce the these decisions had. Interestingly, in this discussion, the slow and gradual diffusion that they had observed in the question of the validity of a model without air was not physical world during the classroom experiment. To do settled simply through the assertion that ‘‘the world is not this, some of these groups introduced structural features in like that,’’ though verisimilitude or fidelity to the real world their virtual containers to slow the exit of particles, as in was certainly important to some students. Rather, student the work shown in Fig. 14. Other students, pursing the groups that had designed a vacuum argued that while same area of interest, created barriers in the open area to realism was certainly sacrificed in this sense, working with affect the rate at which particles moved through the space, a vacuum allowed them to more closely investigate other as in Fig. 15. aspects of particle behavior. That is, these constructions This spontaneously formed interest-based subgroup of maintained their status as models of particle diffusion, the class provides a microcosm that illustrates how groups’ while departing from being models of the particular independent work supported the emergence of collective phenomenon and conditions of perfume diffusion as understandings. Because different groups explored differ- experienced in the opening classroom experiment. ent facets of the diffusion phenomenon, or explored the 123 J Sci Educ Technol (2015) 24:265–286 283 same facet from different perspectives, the class as a whole modeling processes that can be studied in classroom set- covered more ground than any individual group. This tings using an ESS. enabled whole-class discussions to be an opportunity for We also discussed three key elements of the classroom external validation as the class worked to fit individual experience of the ESS that we designed for our PNoM unit. groups’ findings together, identifying both complementar- First, we showed how constructions of representationally ity and conflict, and creating the basis for a shared model of underdetermined entities enabled creative exploration of a the diffusion phenomenon. While it is still in principle range of possible phenomena that could be produced with possible for learners at the end of the unit to have an particulate dynamics governed by the KMT. At the indi- interpretation of the dynamics of the ESS that is at odds vidual level, we showed how the construction process with the KMT, this is much less likely when their findings dialectically balanced explorations of entity dynamics with are integrated with those of their classmates in constructing explorations of possible referents or mappings of the a shared descriptive model. entities in the ESS to aspects of phenomena of interest (in Furthermore, teachers can make strategic use of whole- this case, phenomena associated with the diffusion of class discussions to capitalize on group resources that odors). And at the social level, we described the emergence develop over these iterative cycles of model construction of shared interpretations and shared conventions among the and peer commenting. In the ModelSim project, we have students in particular classroom groups. Second, we recognized the fundamental role of the teacher in facilitat- showed how the discrete and continuous forms of execu- ing consensus-building discussions to draw upon the dis- tability of the ESS facilitated both an immersive experience tributed insights that the classroom group has encountered. of the KMT ‘‘world’’ and the ability to run increasingly In this way, teachers can support the class in creating shared controlled and repeatable experiments on virtual con- expressions of disciplinary core ideas, as facets of the structions. We indicated how these features supported underlying ‘‘Big-M’’ model of the KMT that are directly understanding and argumentation at both the individual and connected to the group’s collective construction work. Our social levels. And third, we discussed the role of external professional development has focused specifically on validation in the modeling process. This occurred through alerting teachers to these opportunities and on exploring increasing use of instrumentation and measurement in high-leverage teacher moves to facilitate such discussions. students’ constructions; through discursive reference to interpretations of the shared diffusion experience; and through recognition and resolution of perceived conflicts or Conclusion incompatibilities between the constructions of different student groups. Finally, we suggested how the teacher In this article, we have introduced the design construct of plays an essential role in capitalizing on the learning an ESS. We have described its role in ongoing research opportunities and insights that arise as the classroom group into the scalability of innovative constructionist approa- engages with the ESS. ches to ABM and model-based inquiry. Here, we defined Each of the features of the ESS that we have discussed scalability along two dimensions: first, as the ability for a in this article was carried forward beyond the diffusion model of implementation to succeed in its learning goals in experiment and into the second week of the PNoM unit. In classroom settings where the multifarious support of the course of that instructional sequence, the initial ESS researchers and designers, characteristic of laboratory was augmented with new components (such as movable studies, is not present. Such scalability allows an innova- platforms that responded to particle impacts and release tion to be applied in a greater number of classrooms and valves to remove particles from constructed systems); under more diverse and variable conditions. Our second increased tools for instrumentation and measurement dimension of scalability referred to the ability of a class- (including tools to measure particle concentrations, the room implementation to make use of the social resources of variable or constant volume of enclosed regions, and so the classroom group in materially enhancing the learning forth); additional means of programmatic control (includ- experience. Such scalability allows an approach to move ing the ability to heat or cool specified regions and to inject beyond an individual experience and register its effects on particles into specific regions at specified points in time); the shared understandings of the classroom group—on their and further opportunities to share constructions and results collective ways of talking about and practicing science. We with the classroom group. With these supports, students have argued that these two dimensions can in fact be linked extended their particulate models of diffusion to build through learning designs that support students in going understandings of the behavior of gases under varying public in meaningful ways with their in-process findings to temperature, pressure, and volume. Finally, in a culminat- an authentic peer audience. Building on these ideas of ing engineering task, they used the KMT principles and scalability, we described our vision of the models and the mechanisms studied over the course of the unit to construct 123 284 J Sci Educ Technol (2015) 24:265–286 hybrid virtual-physical machines designed to achieve par- science using agent-based modeling. Int J Comput Math Learn ticular tasks or outcomes (e.g., to drive a saw, hammer a 14:81–119 Brady C, White T, Davis S, Hegedus S (2013) SimCalc and the nail, propel a hydraulic catapult, or power a circular paddle networked classroom. In: Hegedus S, Roschelle J (eds) The wheel). In this construction, they used the augmented ESS SimCalc vision and contributions: democratizing access to to construct virtual pistons; this virtual motion was trans- important mathematics. Springer, New York, NY, pp 99–121 lated in real time to a coupled physical device that then Buckley BC, Gobert JD, Kindfield A, Horwitz P, Tinker R, Gerlits B, Willett J (2004) Model-based teaching and learning with powered Lego-based machines that the students built. BioLogica: what do they learn? how do they learn? how do Future work in the direction of this article will involve we know? J Sci Educ Technol 13:23–41 more extensive analyses of data we are collecting of stu- Burg B, Kuhn A, Parnin C (2013) 1st international workshop dents’ work with ESS environments, across a wider range on live programming (LIVE 2013). In: Proceedings of the 2013 international conference on software engineering, of classroom settings. Such analyses will enable us to pp 1529–1530 Retrieved from http://dl.acm.org/citation. identify connections between the learning opportunities cfm?id=2487068 and resources generated through construction work with Caperton IH (2010) Toward a theory of game-media literacy: playing ESSs and particular patterns in student inquiry behaviors, and building as reading and writing. Int J Gaming Comput Mediat Simul 2(1):1–16 teacher moves, and other implementation features. They Carey S (1988) Reorganization of knowledge in the course of will also enable us to link work within the ESS with acquisition. In: Ontogeny, phylogeny, and historical develop- quantitative measures of student learning. ment. Ablex, Norwood, NJ, pp 1–27 We argue that an approach to learning design centered Clement J (1982) Students’ preconceptions in introductory mechan- ics. Am J Phys 50:66–70 on ESS environments can place the practice of modeling at Committee for the Workshops on Computational Thinking (2010) the center of a generative social process of inquiry, while Report of a workshop on the scope and nature of computational also engaging other key scientific practices identified in the thinking. National Research Council, Washington, DC NGSS Framework. Indeed, in the course of the imple- Committee for the Workshops on Computational Thinking (2011) Report of a workshop on the pedagogical aspects of computa- mentations of the PNoM unit, we have observed student tional thinking. National Research Council, Washington, DC activity with and within ESS environments that engage Davis SM (2010) Generative activities: making sense of 1098 with each of the eight core practices of science and engi- functions. In: Lesh R, Galbraith PL, Haines CR, Hurford A neering identified by the Framework. It is our hope that our (eds) Modeling students’ mathematical modeling competencies. Springer, New York, NY, pp 189–198 experiences in designing and implementing ESS environ- diSessa AA (1988) Knowledge in pieces. In: Constructivism in the ments as expressed in this article might offer insights into computer age. Lawrence Erlbaum Associates, Inc., Hillsdale, NJ, ways in which technology can be used to enhance the pp. 49–70 processes of learning through models and modeling in diSessa AA (1993) Toward an epistemology of physics. Cogn Instr 10:105–225 experiences that support scalability in both of the senses diSessa AA (1996) What do ‘‘just plain folk’’ know about physics? In: we have defined. The handbook of education and human development: new models of learning, teaching, and schooling. Blackwell, Oxford, Acknowledgments The research reported here is based upon work pp 709–730 supported by the National Science Foundation under Grant #DRL- diSessa AA (2000) Changing minds: computers, learning and literacy. 1020101. Any opinions, findings, and conclusions or recommenda- The MIT Press, Cambridge, MA tions expressed in this material are those of the author(s) and do not diSessa AA, Sherin B (1998) What changes in conceptual change? Int necessarily reflect the views of the NSF. J Sci Educ 20:1155–1191 Driver R, Squires A, Rushworth P, Wood-Robinson V (1994) Making sense of secondary science: research into children’s ideas. Routledge, New York, NY References Finzer W, Erickson T, Binker J (2002) Fathom dynamic statistics software. Key Curriculum Press Technologies, Emeryville, CA Abrahamson D, Wilensky U (2004) ProbLab: a computer-supported Glaser R, Chi MTH (1988) Overview. In: The nature of expertise. unit in probability and statistics. In: Proceedings of the 28th Lawrence Erlbaum Associates, Inc., Hillsdale, NJ, pp xv–xxvii annual meeting of the international group for the psychology of Gobert J, Horwitz P, Tinker B, Buckley B, Wilensky U, Levy ST, mathematics education. Bergen, Norway Dede C (2003) Modeling across the curriculum: scaling up Balacheff N, Kaput J (1997) computer-based learning environments modeling using technology. In: Paper presented at the twenty- in mathematics. In: Bishop AJ, Clements K, Keitel C, Kilpatrick fifth annual meeting of the cognitive science society, Boston, J, Laborde C (eds) International handbook of mathematics MA education. Springer, Netherlands, pp 511–564. Retrieved Goody J (1977) The domestication of the savage mind. Cambridge from http://link.springer.com/chapter/10.1007/978-94-009-1465- University Press, Cambridge, MA 0_15 Guzdial M (1994) Software-realized scaffolding to facilitate pro- Blikstein P, Wilensky U (2004) MaterialSim: an agent-based gramming for science learning. Interact Learn Environ simulation toolkit for learning materials science. In: International 4(1):001–044 conference on engineering education. Gainsville, Florida Hammer D (1996) Misconceptions or p-prims: how may alternative Blikstein P, Wilensky U (2009) An atom is known by the company it perspectives of cognitive structure influence instructional per- keeps: a constructionist learning environment for materials ceptions and intentions? J Learn Sci 5:97–127 123 J Sci Educ Technol (2015) 24:265–286 285 Hammer D, Elby A, Scherr RE, Redish EF (2005) Resources, Martin F, Hjalmarson M, Wankat P (2006) When the model is a framing, and transfer. In: Transfer of learning from a modern program. In: Lesh R, Hamilton E, Kaput J (eds) Foundations for multidisciplinary perspective. Information Age Publishing, the future in mathematics education. Lawrence Erlbaum Asso- Greenwich, CT, pp 89–120 ciates, Hillsdale, NJ Hawkins D (1974) The informed vision: essays on learning and McCloskey M (1984) Naive theories of motion. In: Mental models. human nature. Algora Publishing, New York, NY Lawrence Erlbaum, Hillsdale, NJ Jackiw N (1991) The geometer’s sketchpad [Software]. Key Curric- McDermott LC (1983) Critical review of research in the domain of ulum Press Technologies, Emeryville, CA mechanics. In: First international workshop research on physics Jona K, Wilensky U, Trouille L, Horn MS, Orton K, Weintrop D, education. Paris, pp 139–182 Beheshti E (2014) Embedding computational thinking in Minsky M (1986) The society of mind. Simon & Schuster, New York science, technology, engineering, and math (CT-STEM). In: Moreno-Armella L, Hegedus SJ (2009) Co-action with digital Paper presented at the future directions in computer science technologies. ZDM Math Educ 41(4):505–519. doi:10.1007/ education summit meeting, Orlando, FL s11858-009-0200-x Kafai YB (1995) Minds in play: computer game design as a Moreno-Armella L, Sriraman B (2005) The articulation of symbol context for children’s learning. Lawrence Erlbaum Associates, and mediation in mathematics education. ZDM Math Educ Hillsdale, NJ 37(6):476–486 Kafai YB (2006) Constructionism. In: The Cambridge handbook of Newell A, Simon HA (1972) Human problem solving, vol 14. the learning sciences. Cambridge University Press, New York Prentice-Hall, Englewood Cliffs, NJ Kaput J, Roschelle J (1996) SimCalc MathWorlds. University of NGSS Lead States (2013) Next generation science standards: for Massachusetts, Dartmouth, MA states, by states. The National Academies Press, Washington, Konold C, Miller CD (2005) TinkerPlots: dynamic data exploration. DC Key Curriculum Press, Emeryville, CA Noss R, Hoyles C (1996) Windows on mathematical meanings: Kuhn TS (1970) The structure of scientific revolutions. University of learning cultures and computers. Kluwer, Dordrecht Chicago Press, Chicago Osborne R, Wittrock M (1985) The generative learning model and its Laborde JM (1990) CABRI Geometry [Software]. Brooks-Cole implications for science education. Stud Sci Educ 12:59–87 Publishing Co, New York, NY Papert S (1980) Mindstorms. Basic Books, New York Lehrer R, Schauble L (2006) Cultivating model-based reasoning in Papert S, Harel I (1991) Situating constructionism. In: Construction- science education. In: Cambridge handbook learning sciences, ism. Ablex Publishing, New York pp 371–388 Passmore CM, Svoboda J (2012) Exploring opportunities for Lesh R, Doerr H (2000) Symbolizing, communicating, and mathe- argumentation in modelling classrooms. Int J Sci Educ matizing: key components of models and modeling. In: 34(10):1535–1554 Symbolizing and communicating in mathematics classrooms. Posner GJ, Strike KA, Hewson PW, Gertzog WA (1982) Accommo- Lawrence Erlbaum Associates, Hillsdale, NJ, pp 361–384 dation of a scientific conception: toward a theory of conceptual Lesh R, Doerr H (2003) Foundations of a models and modeling change. Sci Educ 66(2):211–227. doi:10.1002/sce.3730660207 perspective on mathematics teaching, learning, and problem Roth WM (1995) Affordances of computers in teacher–student solving. In: Beyond constructivism: a models & modeling interactions: the case of interactive physicsTM. J Res Sci Teach perspective on mathematics teaching, learning, and problem 32(4):329–347 solving. Lawrence Erlbaum Associates, Hillsdale, NJ Schwartz JL, Yerushalmy M (1987) The geometric supposer: an Lesh R, Doerr H (eds) (2003b) Beyond constructivism: a models & intellectual prosthesis for making conjectures. Coll Math J modeling perspective on mathematics teaching, learning, and 18(1):58–65 problems solving. Lawrence Erlbaum Associates, Hillsdale, NJ Schwarz CV, Reiser BJ, Davis EA, Kenyon L, Achér A, Fortus D, Lesh R, Doerr H (2012) Alternatives to trajectories and pathways to Krajcik J (2009) Developing a learning progression for scientific describe development in modeling and problem solving. In: modeling: making scientific modeling accessible and meaningful Blum W, Ferri RB, Maaß K (eds) Mathematikunterricht im for learners. J Res Sci Teach 46(6):632–654 Kontext von Realität, Kultur und Lehrerprofessionalität. Sengupta P, Wilensky U (2005) NIELS: an emergent multi-agent Springer Fachmedien Wiesbaden, Germany, pp 138–147 based modeling environment for learning physics. In: Proceed- Lesh R, Hoover M, Hole B, Kelly A, Post T (2000) Principles for ings of the 4th international joint conference on autonomous developing thought-revealing activities for students and teachers. agents and multi-agent systems (AAMAS). Utrecht, Netherlands In: Kelly A, Lesh R (eds) Handbook of research design in Sengupta P, Wilensky U (2009) Learning electricity with NIELS: mathematics and science education. Lawrence Erlbaum Associ- thinking with electrons and thinking in levels. Int J Comput Math ates, Mahwah, NJ, pp 591–646 Learn 14:21–50 Lesh R, Hamilton E, Kaput J (eds) (2008) Models & modeling as Sengupta P, Farris AV, Wright M (2012) From agents to continuous foundations for the future in mathematics education. Lawrence change via aesthetics: learning mechanics with visual agent-based Erlbaum Associates, Hillsdale, NJ computational modeling. Technol Knowl Learn 17(1–2):23–42 Levy ST, Wilensky U (2009a) Crossing levels and representations: Sengupta P, Kinnebrew JS, Basu S, Biswas G, Clark D (2013) the Connected Chemistry (CC1) curriculum. J Sci Educ Technol Integrating computational thinking with K-12 science education 18:224–242 using agent-based computation: a theoretical framework. Educ Levy ST, Wilensky U (2009b) Students’ learning with the Connected Inf Technol 18:351–380 Chemistry (CC1) curriculum: navigating the complexities of the Sherin B (2006) Common sense clarified: the role of intuitive particulate world. J Sci Educ Technol 18:243–254 knowledge in physics problem solving. J Res Sci Teach Levy ST, Novak M, Wilensky U (2006) Students’ foraging through 43:535–555 the complexities of the particulate world in the Connected Sherin B, diSessa AA, Hammer D (1993) Dynaturtle revisited: Chemistry (MAC) curriculum. In: Annual meeting of the learning physics through the collaborative design of a computer American educational research association, San Francisco. model. Interact Learn Environ 3:91–118 Retrieved from http://citeseerx.ist.psu.edu/viewdoc/download? Simon HA, Chase WG (1973) Skill in chess: experiments with chess- doi=10.1.1.170.3899&rep=rep1&type=pdf playing tasks and computer simulation of skilled performance 123 286 J Sci Educ Technol (2015) 24:265–286 throw light on some human perceptual and memory processes. Wilensky U (1999b) GasLab: an extensible modeling toolkit for Am Sci 61(4):394–403 exploring micro-and-macro- views of gases. In: Roberts N, Smith JP, diSessa AA, Roschelle J (1994) Misconceptions recon- Feurzeig W, Hunter B (eds) Computer modeling and simulation ceived: a constructivist analysis of knowledge in transition. in science education. Springer, Berlin, pp 151–178 J Learn Sci 3(2):115–163. doi:10.1207/s15327809jls0302_1 Wilensky U (2001) Modeling nature’s emergent patterns with multi- Stewart J, Cartier JL, Passmore CM (2005) Developing understanding agent languages. In: Proceedings of EuroLogo 2001. Linz, through model-based inquiry. In: How students learn: science in Austria the classroom. The National Academies Press, Washington, DC, Wilensky U (2003) Statistical mechanics for secondary school: the pp 515–565 GasLab modeling toolkit [special issue]. Int J Comput Math Stieff M, Wilensky U (2003) Connected chemistry—incorporating Learn 8(1): 1–4 interactive simulations into the chemistry classroom. J Sci Educ Wilensky U (2014) Computational thinking through modeling and Technol 12:285–302 simulation. In: Whitepaper presented at the summit on future Stroup W, Wilensky U (2014) On the embedded complementarity of directions in computer education. Orlando, FL. http://www. agent-based and aggregate reasoning in students’ developing stanford.edu/*coopers/2013Summit/WilenskyUriNorthwestern understanding of dynamic systems. Technol Knowl Learn REV.pdf 19(1–2):1–34 Wilensky U, Papert S (2006) Restructurations: reformulations of Stroup W, Ares N, Hurford A (2005) A dialectic analysis of knowledge disciplines through new representational forms. In: generativity: issues of network-supported design in mathematics Annual meeting of the American educational research associa- and science. Math Think Learn 7(3):181–206 tion, San Francisco Stroup W, Ares N, Hurford A, Lesh RA (2007) Diversity by design: Wilensky U, Papert S (2010) Restructurations: reformulations of the what, why and how of generativity in next-generation knowledge disciplines through new representational forms. In: classroom networks. In: Lesh RA, Kaput JJ (eds) Foundations of Proceedings of constructionism 2010 Paris, France the future: twenty-first century models and modeling. Lawrence Wilensky U, Reisman K (2006) Thinking like a wolf, a sheep, or a Erlbaum, New York, NY firefly: learning biology through constructing and testing com- Tasar MH (2010) What part of the concept of acceleration is difficult putational theories—an embodied modeling approach. Cogn to understand: the mathematics, the physics, or both? ZDM Math Instr 24:171–209 Educ 42:469–482 Wilensky U, Stroup W (1999a) HubNet. Center for connected Tisue S, Wilensky U (2004) NetLogo: design and implementation of a learning and computer-based modeling, Northwestern Univer- multi-agent modeling environment. In: Proceedings of the agent sity, Evanston, IL. http://ccl.northwestern.edu/netlogo 2004 conference on social dynamics: interaction, reflexivity and Wilensky U, Stroup W (1999) Learning through participatory emergence, Chicago, IL simulations: Network-based design for systems learning in Trouille L, Beheshti E, Horn M, Jona K, Kalogera V, Weintrop D, classrooms. In: Proceedings of the 1999 conference on computer Wilensky U (2013) Bringing computational thinking into the support for collaborative learning, CSCL ‘99 Palo Alto, CA high school science and math classroom. In: American astro- Wilensky U, Levy S, Novak M (2004) Connected chemistry nomical society meeting abstracts, vol 221 curriculum. Retrieved from http://ccl.northwestern.edu/curricu Trowbridge DE, McDermott LC (1980) Investigation of student lum/ConnectedChemistry/ understanding of the concept of velocity in one dimension. Am J Wilensky U, Brady C, Horn M (2014) Fostering computational Phys 48:1020–1028 literacy in science classrooms. Commun ACM 57(8):17–21 Trowbridge DE, McDermott LC (1981) Investigation of student Wilkerson-Jerde MH (2012) The DeltaTick project: learning quan- understanding of the concept of acceleration in one dimension. titative change in complex systems with expressive technologies. Am J Phys 49:242–253 Retrieved from http://dl.acm.org/citation.cfm?id=2522404 White BY (1993) ThinkerTools: causal models, conceptual change, and Wilkerson-Jerde MH, Wilensky U (2010) Restructuring change, science education. Cogn Instr 10(1):1–100. doi:10.2307/3233779 interpreting changes: the deltatick modeling and analysis toolkit. Wieman CE, Adams WK, Perkins KK (2008) PhET: simulations that In: Proceedings of constructionism 2010. Paris, France enhance learning. Science 322(5902):682–683 Windschitl M, Thompson J, Braaten M (2008) Beyond the scientific Wilensky U (1996) Making sense of probability through paradox and method: model-based inquiry as a new paradigm of preference programming: a case study in a connected mathematics frame- for school science investigations. Sci Educ 92(5):941–967 work. In: Constructionism in practice: designing, thinking, and Wittrock M (1989) Generative processes of comprehension. Educ learning in a digital world. Lawrence Erlbaum, Mahwah, NJ Psychol 24(4):345–376 Wilensky U (1999a) NetLogo [computer software] version. Center for Wittrock M (1992) Generative learning processes of the brain. Educ connected learning and computer-based modeling, Northwestern Psychol 27(4):531–541 University, Evanston, IL. http://ccl.northwestern.edu/netlogo 123 View publication stats