Computing devices are being used by an ever increasing variety of users in an increasing variety of contexts. It’s therefore becoming harder to design a system that will be suitable for all. Studies show that less than 5 per cent of people adjust default settings which makes the use of adaptive systems an attractive propostition for both user and designer.
An adaptive user interface (AUI) is one that adapts its behaviour to individual users on the basis of processes of model acquisition and application that involve some form of learning, inference or decision making.
AUI's are also known as user modelling systems, software agents and personalisation systems. They are different from adaptable systems as these can be tailored to users prefs.
They have many functions whihc can be loosely categorised into support system use or information acquisition. There are also a number of usability challenges.
Help find information, adaptive search engines, essentially.
Aid browsing. Pinterest is a relative newcomer to this field: users are presented with a one-of-a-kind visual interface based on their tastes. The items they see are curated through people and topics they've identified as interesting and what is shown to them improves the more they interact with it.
Query based search. With personalized search, a AUI keeps track of search history. Google introduced the concept, however surveys indicate that the public needs convincing.
There are also ethical issues. One of Google’s newest tools, Search plus Your World, is the most direct attempt yet, by Google, to use its core service to help it make up lost ground in social networking. However, favouring its own Google+ network at the expense of rivals could heighten regulatory concerns
Another interesting example is Thoughtwire: an adapted call centre which uses software “agents”- essentially carefully planned scripts - to extract the data it needs. It's particulalry well-suited to heavily regulated industries, like health-care, where confidentiality concerns can prevent backend databases from being fully integrated.
Spontaneous Provision of Information. Methods used are of the utmost importance. Pop Ups, for instance, risk being intrusive, but if the method is too subtle, it will be ignored.
Search engine specialist Wolfram Alpha is considering the prospect of pre-emptive search. Managing Director, Stephen Wolfram envisions a search engine that can — through data maps of personal history — provide reports automatically when they’re needed without an explicit query. It can tell you, for instance, which planes are flying over your head.The company has, in its technological arsenal, tools for analytics and visualisation, linguistic understanding, image processing and a way to deal with diverse data in uniform ways — all ingredients Wolfram suggests will be key to the future’s preemptive search engine.
TV is another area where AUI's are likely to flourish. Samsung's Smart TV’s can "predict what you want to watch and also to share special moments and events with your family and friends", however they have as yet failed to capture the publics interest in a major way.
Recommending products. Amazon is best known for providing this service, also known as collaborative filtering or cross-selling
Supporting Collaboration - particularly useful for learning or for help with complex tasks, a service provided by Quora.
Supporting Learning: Recent work has been concerned with continuous adaptation while a student learns, in the belief that many systems are too static. US company, Knewton, claim to be developing the industry’s “most powerful adaptive learning engine” with “continuous adaptability and the ability to customise educational content to meet the needs of each student on a daily basis.”
Learning agent, eDoorways is launching a product called SmartONE - an AI adaptive learning system. It uses Natural Language understanding which is designed to handle anything the user says, even if it cannot be properly interpreted within the learning context. The idea is that students will be able to chat with the app, just as you would with a human tutor.
Taking over routine tasks such as email or shedulung meetings. For these basi duties, different degrees of success have ocurred with different users. The most important factor is a minimal consequence of error
Adapting the interface - the idea here is to adapt only the things - menus, icons - the user needs displayed.
Research conducted on Word 2000's Smart Menu indicates users preferred it over a static menu. An additional study indicated that interfaces which duplicate rather than move frequently used but hard to access functionality are most satisfactory. Accuracy had a stronger effect on performance, utilisation and some satisfaction ratings than the improvement in predictability.
Certain classes of users such as the motor impaired, studied within the SUPPLE project (pdf) were shown to benefit most from adaptive interfaces.
However, in spite of major progress in AUI research, we still lack a methodology for determining when and how adaptivity should be implemented. Context is vey important. The preferred type of system depends on a number of factors, such as the frequency at which tasks are performed, a users age, the difficulty of the task and the level of involvement.
It may be beneficial to consider intermediate levels of adaptivity, rather than seeing the introduction of adaptivity as an all-or-none decision, Intermediate levels keep users involved in the task and help them become more proficient when performing both routine and non-routine tasks.
But a technique such as animating a transition scenario, showing the evolution from the user interface before adaptation to the user interface after adaptation, has met with mixed results, mainly because the time the animation took.
Helping with system use.
It can be difficult to recognise the user’s goal. Microsoft’s Office Assistant, for example, has had mixed views. However, Mitsubishi's washing machine system, DiamondHelp based on dialogue with a user, met with reasonable success.
Complex Systems: A study by Bezold and Minker, 2010 (pdf) worked on adaptation as a solution for coping with the increasing complexity of user interfaces in smart environments, such as interactive TV systems in the living room or infotainment systems in cars. They found an age difference: users older than 45 prefer a static interface, whereas younger users like the idea of a system that selects the most appropriate version for them.
Energy: Providing mechanisms, such as Smart Meters, to gracefully reduce the energy consumed when its full functionality is not used offers the promise of dramatically reducing display power consumption without compromising user acceptance.
Mediating Interaction with the world by working with a user’s cognitive and emotional state.
Lilsys is a system which assesses whether a user is moving or speaking, it can consult a calendar and inform callers on status. Callers have been shown to change the nature of their communication as a result. This area promising, status often changes and users are often too busy to update.
The technology is still young. The ability to pick up nuances, for example at meetings or subtle flaws in sales presentations remains a hard computational problem.
Products have begun to appear in cars. A study by Tchankue, Wesson and Vogts (2011) looked at at the impact of an AUI on reducing driver distraction. The level of distraction is detected by a neural network using the driving speed and steering wheel angle of the car as inputs. Results obtained showed that the adaptive version provided several usability and safety benefits, including reducing the cognitive load, and, crucially, users preferred it.
Controlling a dialogue. Natural language advances could alleviate the need to learn how to use an interface, guiding the user through a process and providing different explanations for novice and professional users and a filter process specific to gender and age.
In Kyoto, a successful experimental bus information system (pdf) was developed to adjust on the basis of: levels of skill in using the system, knowledge of the domain and urgency.
Recognition of negative emotions is a further challenge for systems. For example, to expedite the transfer of certain calls to a human operator within a calling system
These include threats to predictability, controlability and data privacy issues.
Common methods of obtaining Information about about users include: Explicit self report or tests which involve tedious menu selection. There are privacy issues so requests need to be restricted and explained. A solution is simply to see what users click on. However, these naturally occurring actions can be difficult to analyse
Previously stored info and user modelling servers are other alternatives. Privacy is also an issue here.
Methods of accessing user data are becoming more powerful with advances in technology the booming field of data mining. An issue for adaptive systems is that the user data they need to adapt is the very data which users are being warned to withold
A number of groups worldwide are attempting to create thought-controlled robots, which learn from people and from each other. The application potential for such technology is limitless and is becoming increasingly easy to replicate, using even commercially available materials, such as Lego Mindstorms.
To conclude, adaptive technologies are coming thick and fast and have been galvanised by new developments such as more accurate voice recognition and advances in AI. However, a consumer backlash about data provision may hold the development of AUI’s at bay unless adequate provision for privacy of data is implemented and observed.