Summary:
The paper pointed out the theory that there is no universally ideas structures for effective teamwork. The best structure for teamwork is determined by different team members, tasks, surroundings, etc. Thus, this paper presents a system that investigates the optimal team structure by adapting different team structures to teams and evaluate the efficiency based on team performance and teamwork feedback. However, the combination of diverse dimensions of team structures and the arms (the different values of dimension) of each dimension is a large set. To avoid overwhelming group testers with these values, the paper also leverages a model called multi-armed bandits with temporal constrains which set the constraints of the number of arms selections based on several factors. The paper tested the platform with AMT workers, and evaluate the performance of the system with designed task and performance evaluation. The system confirmed that there are no two teams had the same optimal team structures, and this structure even different for the same group when completing the different tasks. The results also indicate the platform DreamTeam can promote high-efficiency teamwork.
Reflection:
First, I highly agree with the opinion that there are no universal idea structures for effective teamwork. Besides, searching the optimal structure through adapt different dimensions and evaluating each fit seems also reasonable. However, I think the experiment could gather more valuable information if they test the platform with a real group instead of a randomly formed group. Because I think the premise of becoming a group and completing a task together is that the group members are familiar with each other. Thus, this platform should be more effective in the early stages of team formation. Before the team members are familiar with each other, they can use this system to find the optimal team structure temporarily, so that they can quickly cooperate and work as a team even though they do not know each other. Nevertheless, as the familiarity among the group members raises, using this method to determine the optimal structure may be inefficient. Because they may have already find the best structure for some of the dimensions as they get along and getting more experience of working together.
I also considered another situation that is more suitable for this system, when a long-established team is assigned to working on a new type of task. Then maybe the working mode of the teamwork needs to be switched so that they can complete the new task most efficiently. At this time, the system support is demanded to find this new optimal structure.
Finally, I think the constraints method mentioned in the article also very inspirational. Maybe we can improve the effectiveness of the DreamTeam platform by allowing users to pre-delete some dimensions that they would not like to change. For example, the hierarchy or interaction pattern. In this case, the reduction of the combination is more conducive to exhaustive testing, and the adapting structure should be more fit for the teamwork.
Question:
- What do you think using the computation power to decide the optimal structure for teamwork?
- In this paper, the author finds random tester to form a group and complete the task, do you think this will influence the results?
- Under what condition do you think this platform would most benefit.
Word Count: 544
Hi Nan, good point! I agree with you that the experiment has a lot of room to improve. Randomly dividing the recruited crowd workers into five conditions does seem to be a good way to design persuasive control experiments. We have no information about the members in each team, such as their age, educational level, experience in such games, etc. Skewed grouping may lead to skewed results, which makes it even more difficult to evaluate what factors in the process make one team works better than others. The factors that make it work could be the knowledge level, experience with alike games, diversity of team members, and so on, not necessary the team structure. To answer your questions 1, I feel using the computational power to generate the optimal structure provide a unique angle, but it is just one of the means and the term “optimal” is ill-defined in the problem setting. The evaluation of the influence that team structure plays is kind of tricky and we cannot guarantee using a AI system like DreamTeam would always precede the decision coming out from the discussion among team members and leaders. I think the paper to discuss more about under what situations DreamTeam works better based on the observations in their experiments.