Adaptive Interfaces and Agents
Anthony Jameson

1. Introduction
   • Concepts
     • Brief introductory example
       • System for personalized presentation of news stories on a handheld device
     • What types of system does this chapter concern?
       • Systems (S) that adapt to an individual user (U) on the basis of some nontrivial inferences from information about U
     • Terms commonly employed to refer to such systems
       • Adaptive interfaces
       • Intelligent agents
         • The term agents covers a wide variety of systems (and components of systems)
         • We consider here those agents that support an individual U by performing tasks for U in a way that is adapted to the goals, preferences, habits, etc. of U
       • User-adaptive systems
       • [A number of other terms]
     • Relationships with concepts of adaptability, anthropomorphism, and personalization
   • Potential usability benefits and problems
     • Note: These issues will be discussed in more detail in connection with specific types of system; the brief discussion here serves as an advance organizer
     • Potential benefits
       • Depending on the type of adaptation involved:
         • Faster and more extensive access to relevant information
         • Better comprehension of system output
         • Saving of time through task migration from user to system
         • More successful task performance
         • Greater tendency to continue to use S (benefit largely for the provider of S, as when S is an e-commerce site)
     • Central usability issue: Potential lack of controllability, predictability, and trust
       • Conditions under which these problems are most likely to arise
       • Strategies for preventing them from arising
       • Note: For further discussion of issues of trust, ethics, and privacy, readers will be referred to the handbook chapters that deal with these topics
   • Reasons for increasing importance of adaptive interfaces and agents
     • Where are such systems currently used?
       • Remarks on commercially deployed systems, other systems in regular use, and systems apparently on the verge of widespread use
     • Why are these systems becoming more important?
       • Increasing number, diversity and/or complexity of interactive systems and computing devices
       • Increasing diversity of users and contexts of use
       • Increasing feasibility of successful adaptation
   • Organization of rest of article
2. Functions of Adaptation
   • Introduction
     • In each of the following subsections, an example system will be presented briefly that illustrates one way in which a system can support the individual user through adaptation
     • Then the following questions will be addressed, with reference to the example system for concreteness:
       • How does the introduction of this type of adaptivity change the way in which U performs his or her tasks?
       • What are the potential benefits of this type of adaptation?
         • Examples of empirical results concerning these benefits will be presented briefly here; a more detailed discussion of empirical studies will be given in the final major section of the chapter
       • What are the potential limitations and drawbacks?
         • In particular: In what forms do issues of controllability, predictability, and trust arise?
   • Help U to find information
     • Example: A system for recommending web pages and/or downloadable documents
   • Tailor information presentation to U
     • Example: A system for personalized presentation of medical information
   • Recommend products
     • Example: An entertainment recommendation system that uses social recommendation ("collaborative filtering") methods
   • Help with routine tasks
     • Example: A system that takes over part of the routine processing of email
   • Perform tasks for U that require more complex planning and/or reasoning
     • Example: An web agent that finds and purchases products on U's behalf
   • Adapt an interface
     • Example: A system that adapts the layout of its menus on the basis of U's usage patterns
     • Note: This type of adaptation tends to cause the most serious problems of predictability and consistency
   • Give help
     • Example: An intelligent help system that tries to infer U's current goal on the basis of U's actions
   • Support learning
     • Example: A web-based learning environment that helps U to find the most suitable material to study next
     • Note: Many adaptive learning environments employ techniques that are specific to that type of system, but the example and discussion in this chapter will refer to principles that are of more general relevance to the HCI field
   • Conduct a dialog
     • Example: A speech dialog system that adapts its dialog strategy on the basis of the progress of the dialog that has been conducted so far
   • Support collaboration
     • Example: A system that helps U to find a person who can help U with U's current problem
3. User Properties Modeled
   • Introduction
     • Systems that adapt to their users usually model some long- and/or short-term properties of each user
     • First, we present a picture of the general relationships among different kinds of properties
     • Then, in each of the following brief subsections, we make some comments on one type of property, referring to one concrete example (either from a previously introduced system or from one introduced in the subsection in question)
   • Personal characteristics (e.g., profession)
   • General interests and preferences
   • Knowledge and skills
   • Current goals
   • Noncognitive abilities (e.g., motor and perceptual limitations)
   • Behavioral regularities (e.g., email processing habits)
   • Psychological states (e.g., emotions)
   • Context of interaction (e.g., location, social context)
4. Information Obtained About Each User
   • Introduction
     • The types of information that are obtained about each user influence:
       • the types of inference that can be made by S
       • the amount of additional effort that U must invest in order to make adaptation possible
     • As in the previous section, for each type of information about U, some comments are made with reference to a (previously seen or new) example
   • Self-reports on personal characteristics
   • Self-reports on knowledge, skills, and interests
   • Evaluations of specific objects (e.g., ratings of web pages)
   • Responses to test items (e.g., in a learning environment)
   • Naturally occurring actions
   • Low-level indices of psychological states (e.g., as transmitted by sensors on U's body)
   • Evidence about the current context of use (e.g., environmental noises captured by a microphone)
5. Approaches to Inference and Decision Making
   • Introduction
     • Reasons for including this section
       • Even in an HCI-oriented article it is worthwhile to provide an overview of the main approaches to inference and decision making that are used to realize adaptive interfaces and agents
       • The different paradigms tend to generate different types of system behavior and user experience
       • Moreover, potential designers of such systems will want to have an idea of what is possible given the current state of artificial intelligence technology
     • Structure of each subsection
       • Each subsection will introduce one broad paradigm with reference to one or more previously introduced example systems
       • It will summarize the main strengths and limitations of each paradigm
   • Classification learning
     • This broad category includes a variety of machine learning techniques
     • The most common pattern is that S learns, on the basis of observations of U, to make predictions about U given features of the current situation
   • Social recommendation
     • This paradigm is also known as "collaborative filtering"
     • S makes predictions about U on the basis of U's similarity to other known users
   • Decision-theoretic methods
     • This family of methods includes the most systematic techniques for dealing with uncertainty about the user and with tradeoffs among conflicting goals
   • Application-specific procedures
     • This broad category includes all methods that do not rely on general inference or decision making techniques
     • Instead, the designers specify qualitative rules and/or computational procedures on the basis of their domain-specific knowledge
   • Other paradigms for inference and decision making
     • Other paradigms that are less widely used are discussed more briefly
6. Especially Applicable Empirical Methods
   • Introduction
     • The various types of empirical methods that are discussed in Section VI of the handbook, The Development Process, are applicable in particular to adaptive interfaces and agents (though they have traditionally been employed less frequently than would have been desirable, for various reasons)
     • This section of the present chapter focuses on four types of empirical method, discussing how each one can address one or more issues that are especially important in connection with systems that adapt to their users:
       • The accuracy of S's inference and decision making
       • The acceptability of S's adaptive behavior, especially with regard to the issues of controllability and predictability
     • Each subsection summarizes at least one example of an application of the method in question to one of the systems that was introduced earlier
   • Wizard of Oz studies
     • This method is especially suitable for assessing the potential value and usability of an intelligent agent that would be hard to prototype but that can be emulated by a human
   • Simulations with existing data
     • Paradoxically, the accuracy of S's inferences can often be evaluated most effectively through the use of data collected from users working with a normal, nonadaptive system
   • Controlled studies
     • A typical controlled study compares an adaptive version of a system with one or more nonadaptive variants
   • Studies of actual system use
     • Whether a system's adaptation and intelligence are truly acceptable and useful can often be determined only through the study of actual use
7. Concluding Remarks
   • The main lessons of this chapter are recapitulated
   • Several key questions that call for the attention of researchers and practitioners during the coming few years are listed
8. References
   • The reference list will include not only the works that were discussed in some detail in the chapter but also a larger number of works that were cited briefly with the aim of providing readers with further entry points into the literature