Why Build Adaptive Systems -------------------------- Browne, Norman and Riches Presented by Darin Krasle Friday, September 30, 1994 Class notes: Dawn Chappelle Jones gt9167a@prism.gatech.edu Text from overheads which Darin used in class: Overhead 1 ---------- How do users differ? According to Noah & Halpin: Users' model of general computer use Users' model of the system under consideration Users' physical and perceptual abilities Users' cognitive skills Users' social skills Users' ability to identify and perform tasks Users' abilities in certain domains According to Veer et al.: cognitive styles field independence vs. field dependence - being fixed in your ways impulsivity vs. reflectivity - haphazardness operation learning (vs.) comprehension learning - procedural knowledge vs. conceptual understanding personality factors introversion vs. extroversion - reaction to repetition or stress negative fear of failure - panicing perception of own competence - leads to less exploration Overhead 2 ---------- According to the Authors (Browne, Norman, and Riches) Psycho-motor Skills - typing abilities, impaired users Capability - prior experience, competence Learning Ability - reaching new levels Understanding - of domain problems and tasks Expectations - conceptual transfer from other systems Motives - seriousness, interest, data needs Requirements - accuracy or expediency type trade-offs Cognitive Strategies - solution oriented vs. problem oriented Cognitive Abilities - memory: semantic, spacial, episodic Preferences - control, presentation modes/characteristics Overhead 3 ---------- Temporal Changes - retention over lapses of use * Maskery Study: consistent use showed better performance, higher acceptance of errors, and faster refamiliarization than irregular use typified by higher cognizance and acceptance of adaptation. - personality traits excluded due to difficulty in measuring Situation Specificity - adaptation in relation to some differences may not make sense in some situations - casual/short-term//infrequent use - ATM vs. expert system requirements Overhead 4 ---------- Purpose of Adaptive Systems Extend Systems Life-span - maintainability Widen System's User Base - power expert to grunt Enable User Goals - compatibility, consistency, die-hards Satisfy User Wants - reduce intimidation, add appeal and flexibility Improve Operational Accuracy - human and system working together Increase Operational Speed - productivity Reduce Operational Learning - improve the learning curve Enhance User Understanding - how the system works vs. performing tasks Overhead 5 ---------- Dynamics - things that could be the targets of adaptation Help - various levels of detail, control, presentation Windowing and Scrolling - accommodate preferences Command Language - use terminology that makes sense the user Error Messages - meaningful cause explanations and recommendations Metrics - used for design, building, and evaluation Objective Metric - reason for adaptation; what we want to improve through adaptation - specifies goals and target aspects Theory Assessment Metric - what aspects we presume we have to affect in order to indirectly achieve our objectives when they cannot be directly affected Trigger Metric - what interaction cues, and in what form, trigger adaptive actions/changes Recommendation Metric - what gets changed in the interface/dialog due to triggering Generality Metric - scope of applicability both for users and tasks Implementation Metric - how much the process costs in terms of system performance Overhead 6 ---------- (Refer to Fig. 2.3, p. 45 of handout) Overhead 7 ---------- Design Methodology Requirements Analysis Requirements determine purposes and scope Variability what user variability, task variability will be modelled Acceptability costs vs. benefits Usability Viability Analysis Practicality of: - triggers of adaptation and how they are captured - changes to be made at the user interface - assumptions justifying the adaptation Levels of Adaptability determined Usability Design Architecture must provide acceptable performance Variability of the user interface determined form content timing Adaptive Behavior How is triggering mapped to change Overhead 8 ---------- (cont'd from #7) Build Construct Models User Model Task Model Application model Monitors for triggering, for evaluation Control Mechanisms Post-build Evaluation Effects of Interaction What are the benefits? How generally applicable are the benefits? System Operation Was desired adaptive behavior achieved? Did it result in what we wanted (were the assumptions valid)? Overhead 9 ---------- (Refer to Fig. 2.9, p. 56 of handout) Discussion Topics ----------------- 1. Adaptation seems to be geared toward improving positive qualities, what about reducing negative qualities such as injury, fatigue, stress, and other ergonomic concerns. (see Purposes) Refer to Overhead #4: It was proposed that another section be added to that called 'Ergonomics'. Injuries related to computer work (such as CTS) are becoming a big issue, and we need to develop ways of dealing with this. --2. How can the system identify and control side-effects? | | 3. How well does the proposed design methodology lend itself to the rest | of the system life-cycle, including maintainability? | | 4. How can prototyping be integrated with the design methodology? | | 5. How well can the design methodology be applied to modifying existing _ systems? The rest of the questions were dealt with as a group, as they were all related. Discussion focused on prototyping, and how it could be used, its advantages, and its disadvantages. It was generally agreed upon that prototyping is a necessity when dealing with user adaptive systems - if you do not have any feedback, you can't go back and change the requirements. Usibility needs to be taken into account when designing a prototype: for example, the amount of light on the screen, the noise level, etc. A system should be well tested in the environment in which it will be used. It would be possible to do usability testong on a prototype without building an entire system... separate modules could be constructed, with limited application. It is, unfortunately, impossible to have a prototype of something that measures long-time use. It was brought up that you cannot decide if a system is not working if it hasn't been built yet. However, the other side of the coin is that you can't know if the system will be adaptable to the user once the entire system is built. A simple prototype could be made to test some things, but these would be trivial, and not meaningful in the long-term... you can't know without building the entire thing how hard it would be to change things. It would also be extremely difficult to build an adaptive system using the change-test-change method, as there would be a number of people working on the modules, and there is no way to tell if they would all fit together once the system was finished. There would be so much interdependancy between the parts that the system would be difficult to adapt. The advantage to this is that the modules -would- be separate, and it would be possible to substitute one part for another, or simply switch them out. It is also difficult to determine on what basis you determine what policies go in the 'black box' of the system (eg. which triggers are used) - you can't get the metrics right if you don't have examples of successes and failures to go by. We also talked a little about Maskery's study of frequent use vs. irregular use... if you are an infrequent user of the system, you can't be really familiar with it, hence you might not notice changes - you would go down on the learning curve. It's hard to tell if this study was possibly measuring how wasily the user learned the system. For example, would an infrequent user still be able to accomplish the task even though changes -had- been made. It was decided in the process that individual personalities can't be a factor in adaptivity, unless you could take your preferences from one system to another (the example used was that of ATM cards that use your name differently from system to system). It was also pointed out that changes in technology have made it somewhat easier for us than it was for the authors.