Opportunistic Planning and Direct Manipulation

Experiments

Experiment 1 - Planning the errands for an afternoon

Purpose

The purpose of this experiment is to show how the Opportunistic Planning Model can be used to explain the planning errands for one afternoon.

Description

The class was divided in three groups. Each group had to plan the execution of a set of errands. During the planning process, all the members of the group were asked to "think aloud" their planning ideas. One member of the group wrote down what the other members said.

Each group had a map of the campus and the following list of errands:

It is 11:30 and your class in Whittemore just let out. You now want to complete as man of the following errands this afternoon as you can:

• Go to the coliseum and get football tickets
• Get money from the ATM, available at both Student Centers and the bookstore
• Grab lunch somewhere on campus
• Attend a 2-3pm class in McBride
• Straighten out confusion with you financial support in Sandy Hall
• Pick up your paycheck in Burruss
• Find and copy an article from the library for a paper due in three weeks
• Get transparencies made at Kinko's for a class presentation next week
• Meet a friend for a racquet-ball game in he Memorial Gym at 4
• Stop at the bookstore to buy a new novel you've heard a lot of good things about

Results

The following text is an excerpt from the script of one of the groups:

1. Let us find some fixed points in the schedule
2. There is a class from 2pm to 3pm, meet a friend at 4
3. Get the paycheck
4. Let us see where on the map are the places we have to go
5. Maybe we can find a route
6. Last thing is the racquet ball game
7. Let us try to make a time line of the events
8. It is 11:30 now, we are in Whittemore
9. ATM and Hardee's are close, we can do them one after the other
10. ...

Analysis

The theoretical support for Opportunistic Planning is given in a previous section. In this analysis we assume that the reader knows the concepts used in the model. The analysis shows how the specialists in different planes and at different levels interact with each other:

1. "Get the paycheck" - This errand was in the attention of every group at the beginning of the process. We can assume here that, preceding this statement, the user prioritize the errands and this one has a "high priority" status. Several planes are involve to give priorities to errands: the meta-plan defines "importance" as one of the policies. Other levels in the meta-plan, such as scheduling and scenario are involved in this process. In the executive plan is the priorities level. In lower plans, it involves the intentions in plan abstractions and the list of errands in the knowledge base;
2. "ATM and Hardee's are close we can do them one after the other" - this step requires an evaluation of proximity in the knowledge base plan. It also uses a specialist in the plan abstraction plan that gives the scheme of clustering as a proposal for the design. In the plan itself the subject defines a set of actions to do even if it is not defined in absolute time. It also involves the executive plan because the focus is now on a low level design;
3. "It is 11:30 now, we are in Whittemore" - the planes involved here are the plan abstraction, the knowledge base and the plan plan. The users decides that the strategy to adopt is to work from the start, he identifies in the knowledge base, the position and the time at the beginning. From this he starts imagining him leaving the class room and going to a cluster of errands.
4. ... this interpretation of the script can continue, for every step being able to identify the plans, the levels and the specialists that interact with each other in the design process.

Conclusions

• The experiment gave clear examples of opportunism in planning. Planning is not a linear process. The experiment gave evidence of interleaved cognitive specialists at various levels of abstraction;
• The model proposed by Hayes Roth and Hayes Roth is able to explain the planning process;
• The effort to interpret the script is large. Even a few steps involve the interaction between many planes and levels;
• The experiment designed in the class was slightly different from the experiment described in Hayes Roth and Hayes Roth, 1979 because, in our case there was a group that had to plan, while in the paper each person had to make an individual plan. We are sure that this difference created an even more unstructured planning process.

Experiment 2- Comparison Conversational Model vs. Direct Manipulation

Purpose

The purpose of this experiment was to present the differences, the advantages and the disadvantages between two models for the interaction user-computer: the conversational model and the direct-manipulation model. We chose two applications from the same domain: information retrieval system about movies. One application is the Internet Movie Database, a conversational application used to obtain information about movies on the Internet. The other application is a stand-alone application developed at the University of Maryland that uses a direct-manipulation-style interface.

The class was divided in six groups. Each group had to accomplish a list of tasks:

Find:

1. all movies directed by Steven Spielberg
2. all mystery movies from 1991 to 1993
3. all movies featuring both Robert De Niro and Meryl Streep
4. all movies whose titles begin with the word `time' (not counting `a', `an', and `the')
5. all movies with Mary Elizabeth Mastrantonio in the 90's that are over two hours long
6. some movie you might like to rent this weekend

Results and Analysis

• The groups were able to accomplish the tasks in both environments with less or more difficulty.
• The main idea derived from the discussions about the experiment was that the option for the model itself (conversational or direct manipulation) is not the only factor in the users's performance. Many other aspects, especially usability, affect performance.
• One of the suggestion in the class was that a hybrid model, which allows both direct manipulation and conversational interactions, may be a better solution for the paradigm used in the application.