Currently, this system has only been recently implemented and has not been explored fully. In the next weeks, it will be undergo an evaluation process where several users will interact with it in an attempt to generate ideas about the relevant HCI and design issues. Nevertheless, from our initial experiments, some HCI insights and questions can be gleaned.
One particular issue is that of intrusiveness. The system ideally is to have a peripheral interaction with the users. However, the presence of head mounted microphones was deemed intrusive and clip-on microphones were preferred (although voice recognition accuracy suffered). The presence of a video-camera was not problematic but the output modality (large screen projection screen) was distracting. We are currently installing a projector configuration that will allow screens to be projected onto a flat table or workspace. In addition, privacy concerns related to having microphones are still present and users were reluctant to discuss sensitive topics. It is difficult to be fully comfortable with audio devices unless guarantess about privacy can be made. Even though the users were assured that the system was not recording their words and only had a coarse understanding of the topics, users still constrained their interactions somewhat.
Another issue is revisiting the role of a real human mediator and the interactions they generate. Of course, part of the role of a human mediator is to ask questions but this might be counterproductive in certain meetings where other types of feedback are preferable. For example, the mediator or facilitator in a group meeting scenario has higher level tasks such as deciding when, and whom should convene in a meeting and managing the dynamics between people. In addition, there is usually an a priori agenda that is specified and guides the meeting. It would be interesting to provide the system with this context (the agenda, a list of people, etc.) to further shape its feedback. For instance, if it is known that the meeting's agenda is to discuss politics and health care as agenda items, the system could discourage other topics. Furthermore, certain users will exhibit more general behavior that hinders progress in the meeting and do not attain the group's ultimate goals. If this could be detected, the system could actively discourage this behavior.
Questions and prompts are not the only interaction that generates ideas and furthers a group meeting. The system could generate other output i.e. encourage brainstorming and related activities. For instance, the so-called ``Nominal Group Technique'', ``Brainwriting'', the ``Affinity Diagram'' and the ``Delphi Method'' are idea generation approaches used in the facilitation literature. Other types of general assistance that mediators provide include: correcting bad attitudes in the attendees, evaluation of the subjects and their roles, determining group needs, offering support, reflecting on what was said, confronting speakers in a provocative way, etc. It is evident that various levels of conversational context understanding are required for the above tasks which may be beyond the capacity of the current technology.