Response to [Principles of mixed-initiative …] by Horvitz

One Sentence

This paper presented LookOut, a calendaring and scheduling agent that synthesizes two directions of developing user interface: direct manipulation and interface agents.

More Sentences

“We first presented a set of principles for designing mixed-initiative user interfaces that address systematic problems with the use of agents that may often have to guess about a user’s needs. Then, we focused on methods for managing the uncertainties that agents may have about users’ goals and  focus of attention.”

Key Points

• Principles for mixed-initiative UI:

1. Developing significant value-added automation;
2. Considering uncertainty about a user’s goals;
3. Considering the status of a user’s attention in the timing of services;
4. Inferring ideal action in light of costs, benefits, and uncertainties;
5. Employing dialog to resolve key uncertainties;
6. Allowing efficient direct invocation and termination;
7. Minimizing the cost of poor guesses about action and timing;
8. Scoping precision of service to match uncertainty;
9. Providing mechanism for efficient agent-user collaboration to refine results;
10. Employing socially appropriate behaviors for agent-user interaction;
11. Maintaining working memory of recent interaction;
12. Continuing to learn by observing.

Take-Away

• First of all, I don’t quite think LookOut synthesizes direct manipulation and interface agents – it looks to me more like adding interface agents to an existing direct manipulation -based interface;
• Even thought the example discussed (the LookOut) is indeed a mixed-initiative interface, it remains fairly simple, with only very few direct manipulations involved (sometimes zero – just reading the email). This is weakness when it comes to exploring the synergy between the two streams of work.
• Hunches:
• Agents should choose to do tedious, undesirable tasks (not MSNing or playing the video game for the user);
• Agents should emphasize on ‘offering’ something, instead of promptly getting something done.
• What is really worth learning is the construction of the agent reacting model. The model is logically deduced, but remains quite descriptive to allow for flexible applications. What seems lacking exploration is how to know the ‘utility’ under various circumstances and contexts.