Another resolution of the controversy is to accept user control at the interface,
but consider agent-like or multi-agent programming to automate internal
processes such as disk-space allocation or network routing based on
current loads. However, these are adaptations based on system features, not
user profiles.
Since agent advocates promote autonomy, it seems they must take on the
issue of responsibility for failures. Who is responsible when an agent violates
copyright, invades privacy, or destroys data? Agent designs might be better
received if they supported performance monitoring while allowing users to
examine and revise the current user model.
An alternative to agents with user models may be to expand the control-panel
model. Computer control panels, like automobile cruise-control mechanisms
and television remote controls, are designed to convey the sense of control that
users seem to expect. Users employ control panels to set physical parameters,
such as the cursor blinking speed or speaker volume, and to establish personal
preferences such as time! date formats or color schemes (Figs 2.2 and 2.3). Some
software packages allow users to set parameters such as the speed of play in
games-users start at layer 1 and can then choose when to progress to higher
levels; often they are content remaining experts at layer 1 of a complex interface
rather than dealing with the uncertainties of higher layers. More elaborate control
panels exist in style sheets of word processors, specification boxes of query
facilities, and information-visualization tools. Similarly, scheduling software
may have elaborate controls to allow users to execute planned procedures at
regular intervals or when triggered by events.