• No results found

Chapter 8 • General discussion

8.1 Summary of the contribution

The contributions of this thesis to the field of CSCW focus on the influence of mutual awareness on collaboration and on the process of mutual modeling.

8.1.1

MLA and coordination

Our main hypothesis was that MLA  as a coordination device  may enrich the shared understanding of the group and consequently, facilitate the modeling of each others’ intents, a process called ‘mutual modeling’. We also hypothesized that enhancement of the mutual modeling by a MLA tool would also improve the collaborative task performance. This question was investigated through three controlled experiments and the design of visualizations to make these influences more explicit.

Qualitative analyses of how MLA has been perceived and interpreted by players showed how it could contribute to the common ground of the groups. Figure 68 describes the five main uses players made of a MLA coordination device.

Figure 68. Roles of Mutual-Location Awareness in the three experiments.

The first use of MLA was to understand the situation and therefore to have a shared understanding. We observed a more elaborate role for MLA in that knowing others’ whereabouts allowed participants to gain an implicit awareness of their partners’ activities by making them manifest and comparable to previous actions. As a consequence, players made decisions about what they should do next based on the information supplied by this awareness tool: it was used a resource for division of labor in terms of choosing the next relevant course of actions for individuals involved in collaboration. Though MLA meant players communicated less about certain information because it was automatically conveyed by the tool (e.g. the location of others in the environment), the MLA was also a resource that fostered discussion within groups. The location of a player could indeed be used in a verbal interaction to acknowledge a partners’ utterance or to give an order to a partner. And finally, asynchronous versions of the tools had an additional role since traces of past positions

were important in drawing hypotheses about the future behavior of partners. This was of importance especially regarding directions because past traces enabled players to determine possible future trajectories. All those interpretations of MLA attest to its importance in performing the mutual modeling acts depicted on the figure.

However, though the range of inferences enabled by MLA is important, the quantitative analyses of the three studies elucidates the positive influence of MLA on coordination.

The first experiment in a virtual environment showed the importance of Mutual Location-Awareness in the coordination process. We found positive impacts of the tool on group performance but the results regarding the influence of MLA on the mutual modeling process are intricate. We only find the positive influence of the MLA on groups who effectively used the awareness tool. The results supported the idea that the MLA tool can be beneficial to mutual modeling if two conditions were met: (1) if users properly understand its meaning and use, (2) if users mutually recognize this information in completing their joint action.

The second experiment in physical space also supported the idea that in some situations the MLA influence could be detrimental to collaboration in the presence of a rich synchronous communication medium. Results of this experiment showed no differences in terms of task performance between groups who had the awareness tool and groups who did not have it. Moreover, the presence of the MLA tool proved to be detrimental to collaborative processes such as communication, division of labor and mutual modeling. The players without the MLA were indeed more verbose, especially regarding strategy planning during the task, which led them to better mutual modeling. As a matter of fact, this experiment highlighted an important characteristic of coordination devices such as MLA: their automation could be harmful to successful collaboration. The information about others’ whereabouts indeed inhibited the exchange of other coordination devices and subsequently lead to a poorer common ground in the groups who had the MLA tool. This diminished the possibilities to perform mutual modeling acts. Furthermore, the presence of that information conveyed by the awareness tool substituted less pertinent information (others’ locations) for more relevant information that players had to discuss (such as strategy changes for example). The third experiment showed that suppressing the plan definition phase is more detrimental to completion of a joint task than removing the MLA interface. In this third experiment, players’ performance was diminished when they had no planning phase because they had no opportunity to set up the local conventions of how to behave as a team. Nevertheless, the underwhelming effect of the MLA tool on communication disappeared: suppressing the most important coordination device (in the form of a plan) fostered more communication within the group.

In addition, our investigation of two kinds of MLA tools in the second experiment did not show any significant differences between synchronous and asynchronous location- awareness. However, this result is limited as we only did one experiment contrasting synchronous and synchronous location-awareness.

Overall, one of the main results of our research is that the manipulation of the availability of coordination devices can lead to important changes in collaboration processes. This was exemplified by the visualizations described in Chapter 7, which are tangible representations of the consequences arising from little changes such as adding a MLA interface or suppressing the plan.

8.1.2

The inter-dependence of Mutual Modeling

This contribution was not directly connected to the research question but it emerged from the studies we conducted. The first and the second studies indeed revealed a correlation between the models peers build about each other’s behaviors and intentions. Simply stated, if team member A builds an accurate model of his or her partner B, partner B also tends to build an accurate model of A. The conclusion we draw at this point is that the activity of modeling the partner is not reciprocal but mutual. A reciprocal relationship means that modeling is mainly an individual activity where A infers a model M (A,B) from B's actions and utterances. A mutual relationship implies that M(A,B) and M(B,A) are jointly constructed through interactions. The term 'mutual' may mean that not only A builds M(A,B), but he also builds M(A, M (B,A)). We will not enter here into the extensive debate on the possibility of an infinite regress of nested models (discussed in Smith, 1982 or in Clark, 1996). Another interpretation is that team members actually build a model of the group-in-interaction, something like M (A, AB). We are not able to choose among different hypotheses at this stage, since the reported experiments were not designed for exploring these issues.

This correlation was observed in two different contexts: virtual space in study 1 versus real space in study 2 and groups of 2 in study 1 versus groups of 3 in study 2. Moreover, this correlation between how A model B and how model A was also observed using different methods: on-task in study 1 versus off-task in study 2, subjective validation (comparing A's model to B's answer) in study 1 versus objective validation in study 2 (comparing A's model with B's behavior). This diversity of settings consolidates our results but we still face serious methodological difficulties.