Alan Dix and Janet Finlay
|At time of publication:|| HCI Group and Dept. of Computer Science, |
University of York
|Currently:|| Alan Dix
| Janet Finlay
School of Computing and Mathematics, Huddersfield University
This paper is an (unpublished) extended version of:
A. J. Dix and J. E. Finlay (1989). AMO - the interface as medium. Poster sessions, HCI International'89, . Boston, 22.
See also: the ecology of information web pages
Various interface styles suggest paradigms for understanding interaction. Direct manipulation suggests the interface as a passive entity, tools for the user to control. Intelligent interfaces suggest instead an active interface, a colleague which (or even who) cooperates with the user on the task in hand. User interface management systems typically have a paradigm between the two where the interface is seen primarily as a translator, or mediator between the user and the application. Each of these paradigms seem useful in different contexts, but mixing them runs the risk of getting the worst, rather than the best, of all worlds. Viewing the interface as a medium allows us to make sense of the interplay between passive and active components of an interface.
Direct manipulation interfaces emphasise the passivity of the interface: the user is interacting with things in an artificial world (e.g. a desktop). This interface style is highly popular partly because of the naturalness of the physical metaphors used, partly because of the immediacy of response (if you want something done you do it, rather then tell the system to do it). In addition, the very passivity of the system gives the user a sense of control, the initiative lies with the user. The danger that such user controlled dialogues will be under-determined (Thimbleby 1980) is largely obviated by the graphical presentation of the objects of interest.
Successful as such interfaces are the concept becomes dangerously stretched when extended beyond those applications most well suited such as drawing and simple word processing. In an application such as statistical analysis one is clearly after the machine doing something for you rather than you doing it yourself. It is quite possible to use DM techniques to draw a diagram describing the statistical processes required, and enabling you to see, for instance, intermediate results at the various stages, thus encompassing many of the positive points of DM. However, in the end you want the machine to do the actual calculations, after all that's why you're using it.
Even the classic WYSIWYG word processor starts to fail when more complex facilities are demanded of it: style sheets are added to paragraphs, and alternative view may even be given containing essentially text formatting languages. Basically when the issue was simple page layout or simulated type-writing then DM is sufficient, however as soon as the focus changes to document processing then issues such as consistency of style make us demand that (again) the computer works for us.
Traditional UIMS come from almost the opposite extreme. They act as a bridge between the user and some set of underlying functionality (of which the statistical system would be the perfect example). The aim being that the application should know little about the actual user interface being addressed purely at the application domain itself. The job of the UIMS is to produce a means of accessing the functionality and presenting and manipulating the objects in the application domain. This philosophy of interface independent applications (and even sometimes application independent interfaces!) and the language models that underlies it, has been the subject of extensive critique which more modern UIMS are attempting to address. In particular, many try to be more DMish, but of course they are starting in precisely the application areas where DM has trouble.
One issue that has been a problem with older UIMS was that the UIMS's knowledge of the application was restricted to a type-syntactic description of function calls available. In order to produce sensible interfaces deeper semantic knowledge has often been needed. To address this, several recent UIMS proposals include knowledge bases and expert systems components which overlap to some extent with another strand of active interfaces, the intelligent and adaptive interfaces. These sit between the user and the system, trying to use their knowledge of the system semantic to help the user or perhaps use (assumed) knowledge of the user's intentions and abilities to actively alter the interface style.
In both the case of the traditional UIMS and the intelligent interface the user-interface is seen as mediator between the user and application. The user communicates intentions to the interface agent which then processes these and passes them on to the application. In the case of many UIMS, to allow semantic feedback with reasonable performance the user is also allowed to sometimes talk directly to the application, largely obviating the original intentions. In each, case we have a three agent communication.
So we see two very conflicting views of the interface. One, direct manipulation, emphasises the systems passivity, but has difficulty coping with active application. The other, has a far more active interface component, but lacks that feeling of directness and control that make DM so popular, especially among the computer naive.
Some systems have tried to bridge these two
Viewing the interface as a medium allows us to make sense of the interplay between passive and active components of an interface. The word medium here is taken to include the whole software/hardware amalgam, with both its functional and aesthetic attributes. In particular it is not limited to the information theoretic concept of a channel or the physical characteristics of a device, although these will both be facets of the media. In this paradigm, we can decompose systems into agents (human or machine), media and objects (AMO).
The motivating example for this approach is of course mail and conferencing systems. Clearly when viewed as theoretical communication channels most such systems are identical, they differ more in the qualitative aspects of the interface as a communication medium. The important thing to note is that these non-functional differences such as pace and ease of interaction can make profound differences to the content of communication. This is typified by the differences between face-to-face, telephone and paper mail communication.
In the mail example, the total system consists of the medium and the people who are communicating. The medium is (relatively) passive and the people active. In general systems have more active members than just the humans and we refer to both types of active member as agents. The final classification, objects refers to those components which are passive, but are not merely artifacts of the interface, for instance data-files.
As with many such classifications it can be carried out at various levels in the system description. For example, at one level the postal system is a media, whereas at another level it involves the letters as objects being manipulated by the postman, an agent. Another example is a washing machine, this is an object when it is being lifted out of a van, but an agent when washing clothes.
The distinctions introduced can be used purely descriptively or normatively in judging existing or putative systems. Various examples of interface phenomena can be addressed:
For computer conferencing, do we want a communication system or a system that allows participants to communicate.
Adaptive interfaces are often justified by analogy with human dialogue. When we enter a dialogue with other human agents we expect them to adapt the level and style of the dialogue to their perception of our abilities, knowledge and aims. We are evidently used to participating with adaptive dialogue partners and thus, it may be argued, by analogy computer systems should adapt their interface styles to their perception (sic) of our knowledge etc.
On the other hand, it may be argued that such systems are unpredictable to use (from a control theory perspective, perhaps fundamentally unstable) leaving the users feeling lacking in control as the system continually tries to second guess them and unsure of the response from (and even method of achieving) system operations.
We thus again see the conflict between a user controlled passive (but stupid) interface and a more active, intelligent, and independent one. Can the AMO distinctions help us to resolve this conflict?
If we examine again the analogous human-human dialogue, it consists of two agents, the participants and the medium through which they communicate. In the case of face to face conversation this would include the air through which the sound waves travelled, the language used, the non-verbal visual cues, and in a wider setting perhaps the room in which the conversation is taking place. The important thing to note is that it is the other participant (the agent) which adapts and the medium will remain relatively stable. We would find it disconcerting in the extreme if the air around us began to vary its properties in sympathy with its model of us, perhaps reducing the speed of sound to a few inches per second in order to slow the rate of dialogue down, or adding echoes if it thought we were missing things. The closest effect one could imagine would be chatting next to a sound sensitive disco light controller!
Of course, changes of medium do occur however these tend to occur at a low rate and/or under the mutual agreement of the parties. For instance, with an interspersing of phone conversations and written letters where the pace and context of the medium change is rather obvious. One can consider, more dynamic situations, such as a technical conversation swaps between use of direct speech, whiteboard and pencil and paper, however in this case the alternative media are supplementary rather than representing a total change in the medium of dialogue.
If we look again at the situation of adaptive interfaces, again we have two agents, the user and the application and the adaptive interface which is the medium between them. However in this case, it is not the other agent (the application) that adapts but the medium itself. That is, the supporting analogy breaks down on who and what is adapting. It is not surprising then that the user may feel out of control as the very means of communicating with the system is neither predictable nor stable (like trying to write home to mum using Egyptian hieroglyphics). It can be attacked on the grounds of unpredictability. By analysing this real world situation in terms of agents and media, we can make recommendations as to which parts of a computer system should be subject to adaptivity.
UIMS typically take a rather active role in the interface, they can be seen as an additional agent in the system, between the actual interface media and the application (another agent). Window mangers are usually far less interventionist leaving the application to shape the media. Presenter, a display manger designed at York, supplies more nearly a pure media.
Some of the metaphors used to describe hypertext (eg. stacks of cards) suggest a passive object to browse. However many systems involve computation and change even when not authoring. Depending on users' perception of the active and passive roles of parts of the system they will make different inferences about its possible behaviour and integrity. Clarifying the distinction between the hypertext as media, and computer agents that act on it simplifies the situation and suggests possible improvements to the conceptual design.
We have seen how the AMO framework can be used to analyse real world phenomena and their relationships to various interface techniques, and thus suggest directions for future interface developments.