Allen, J.E., C.I. Guinn, and E. Horvtz. 1999. “Mixed-Initiative Interaction.” IEEE Intelligent Systems and Their Applications 14 (5): 14–23. https://doi.org/10.1109/5254.796083.

Notes

• In human control, human always initiates the interaction

Levels of Mixed Initiative

Unsolicited Reporting

“Agent may notify others of critical information as it arises” (Allen et al., 1999, p. 15)

Subdialogue initiation

“Agent may initiate subdialogues to clarify, correct, and so on” (Allen et al., 1999, p. 15)

Fixed subtask initiative

“Agent takes initiative to solve predefined subtasks” (Allen et al., 1999, p. 15)

Negotiated mixed initiative

“Agents coordinate and negotiate with other agents to determine initiative” (Allen et al., 1999, p. 15)

“a creative integration of direct manipulation and automated services could provide fundamentally new kinds of user experiences” (Allen et al., 1999, p. 17)

• Mixed initiative interactions support joint activity under uncertainty
• the system should know
  • when to engage users with a service
  • how to best contribute to solving a problem
  • when to pass control of problem solving back to users for refinement or guidance
  • when to query a user for additional information in pursuit of minimizing uncertainty about a task
• Systems with the better knowledge of the user's expected utility should perform better

In-text annotations

"Mixed-initiative refers to a flexible interaction strategy, where each agent can contribute to the task what it does best." (Page 14)

"Furthermore, in the most general cases, the agents’ roles are not determined in advance, but opportunistically negotiated between them as the problem is being solved." (Page 14)

"Mixed-initiative interaction lets agents work most effectively as a team—that’s the key." (Page 14)

"The secret is to let the agents who currently know best how to proceed coordinate the other agents." (Page 14)

"The best way to view interaction between agents is as a dialogue, and thus mixedinitiative becomes a key property of effective dialogue" (Page 15)

"In mixed-initiative interaction, the situation can be more complex. Because different agents might take the initiative at different times, an agent must be able to tell when it should appropriately start an interaction by taking the turn." (Page 15)

"Other researchers have voiced concerns that efforts focused on automation might be better expended on tools and metaphors that enhance the abilities of users to directly manipulate and inspect objects and information." (Page 17)

"a creative integration of direct manipulation and automated services could provide fundamentally new kinds of user experiences" (Page 17)

"I shall use the phrase to refer broadly to methods that explicitly support an efficient, natural interleaving of contributions by users and automated services aimed at converging on solutions to problems." (Page 17)

"we need to develop machinery for gathering information and making inferences about the intentions, attention, and competencies of users—and for ultimately making decisions about the nature and timing of automated services" (Page 17)

"Mixed-initiative systems must consider a set of key decisions in their efforts to support joint activity and grounding. These include when to engage users with a service, how to best contribute to solving a problem, when to pass control of problem solving back to users for refinement or guidance, and when to query a user for additional information in pursuit of minimizing uncertainty about a task." (Page 17)

"Thus, building effective mixed-initiative systems requires the consideration of key uncertainties both at design time and in real time." (Page 18)

"More specifically, access to beliefs about a user’s goals give a mixed-initiative system the ability to take information-gathering and problem-solving actions that have the highest expected utility, taking into consideration the expected benefits and costs of attempting to participate in problem solving." (Page 18)

"The large space of design opportunities for mixed-initiative interaction includes • developing automated services that are performed in line with a user’s activity, allowing users to take advantage of contributions provided by a system while they work in a natural manner, • identifying elegant metaphors that promote efficient grounding by providing efficient means for users and computers to communicate information about intended or ongoing contributions to a solution, and • developing automated services that can provide solutions at varying levels of precision or completeness, giving mixed-initiative systems the flexibility to scope the precision of contributions in accordance with the uncertainty about a user’s goals or the competency of an analysis." (Page 18)

"The latter class of design opportunities is motivated by the notion that, as uncertainty grows about a user’s intentions or about the quality of the result, a system should gracefully degrade its contribution so as to “do less, but do it well.”" (Page 18)