Buxton, W. (1993). HCI and the inadequacies of Direct Manipulation systems. SIGCHI Bulletin, 25(1), 21-22.

HCI and the Inadequacies of Direct

Manipulation Systems



Bill Buxton

Alias | Wavefront Inc.
210 King St. East
Toronto, Ontario
Canada M5A 1J7


The Direct Manipulation (DM) style of user interface made popular by the Macintosh is becoming a de facto standard. Rather than being taken as a point of departure, it appears to be taken more as a standard to achieve. Using the specification of scope as an example, DM interfaces are shown to be deficient in supporting a transaction fundamental to word processing, information retreival and CAD. This essay is a plea for designers to break out of the complacency that surrounds the DM approach. It also calls into question the methodologies of HCI for the very limited degree to which they have challenged the DM approach and their paucity of ideas for generating strong new alternatives.


Direct Manipulation (DM) interfaces are demonstrably inadequate for supporting transactions fundamental to applications such as word processing, computer-aided design, and database queries.

From one perspective, there is nothing earth-shattering in this observation. DM interfaces - as typified by that of the Apple Macintosh - were a significant improvement over what preceded them. But nobody claimed, nor should expect them to be perfect. However, if we look more closely at what has happened since the wide-spread introduction of such systems, I believe that the observation gives rise to a number of questions and issues concerning contemporary user interface research and design.

What concerns me, and what is behind these comments, is a perception that the style of user interface made popular by the Macintosh is becoming some sort of de facto standard. The strongest evidence for the validity of this perception is the herd of "me too" copy-cat interfaces such as Open Look, Microsoft Windows, and the Presentation Manager. Can this rush be justified in business, scientific or ergonomic terms? I think not. As a designer, researcher and user, I believe that we can do better. Much better. And the fact that we haven't is not only disappointing, it begs some fairly serious questions of the user interface community.

These are strong words. To lay the basis for justifying them, therefore, let us look at at least one fundamental inadequacy of DM interfaces.


A Problem of Objectivity

In virtually any application, there are two components of every command: the verb, and that which the verb acts upon: the direct object. This is true if we are copying text, deleting components in a circuit diagram or finding an entry in a data base.

In DM systems, the normal modus operandi is to select and then operate. This is the noun-verb syntax defined in the Macintosh User Interface Guidelines, for example.

In such systems, the direct object is generally selected using the mouse or some other pointing device by demonstrative gestures such as pointing, circling, or dragging through. The directness of this indication is a key component of DM.

But there is a problem. By itself, this form of direct selection and specification is inadequate in nearly all applications.

If we step back one level of abstraction, we see that there are at least two ways to specify the direct object of a verb: by demonstration, as already discussed, and by description. The problem with current DM systems is their near universal failure to adequately support the latter: the descriptive specification of operands.

To support this claim, let us look a bit more closely at what we mean by descriptive specification, why it is important, and establish that it is, in fact, woefully undersupported. We can do so by working through a short example.

Let us assume that we are are working with a simple program for designing digital gate-level logic. Consider the case where we have a single And gate followed by an inverter which we want to convert to a single Nand gate. Using a visual-based DM system, we would typically select the components in question using a mouse and issue the command.

However, what if we are working on a fairly large logic array, and want to perform the same operation on all And gates followed by and inverter? Now we have a problem, since demonstrative forms of interaction are woefully weak at supporting the articulation of the type of Boolean or conditional expression required. Of central importance here is the fact that the demonstrative and descriptive forms are both required in this single application. Furthermore, they are both needed in most applications.

A shortcoming of existing design is that this need is not accommodated by any widely available system. (To be balanced, it must also be stated that the command-line interfaces that preceded DM were fairly good at descriptive scope specification, but poor at demonstrative.) A shortcoming of existing analysis is that this fact has not surfaced as a significant issue in contemporary research or design.


Questioning Methodology

Repeatedly, one seems to hear the user interface research community criticized as being too steeped in analysis and spending too little effort on design to be of much practical relevance. Now this may or may not be true. That is not an issue that I want to debate here.

But what is clear is that there is a large emphasis on analysis in the community. That being the case, and given the shortcomings of existing DM systems, where is the proverbial little boy crying out, "The emperor has no clothes!"?

If we are to argue for the value of analytical methodologies, then the lack of formal analytical criticism of DM systems, - despite shortcomings such as those outlined above - brings the effectiveness of our techniques into serious question.

The current world of user interfaces is analogous to a bicycle propelled society into which a unidirectional automobile has been introduced. The society is so amazed and impressed by how much better they can go forward with this new technology that they appear totally blind to the fact that there is no reverse gear.

For naive consumers, this may be understandable. But for serious analysis of transportation systems, especially those purporting to be doing research to support better design, this is completely unacceptable - yet that is where HCI is today.


In Summary ...

The Macintosh made an important contribution to the evolution of user interfaces. It introduced a new, significantly improved standard of design and usability for the end user.

As Alan Kay has said, it is one of the few computers worthy of serious criticism. But such criticism is too lacking in the research and design communities. Too often Macintosh-like interfaces are being take as a point of arrival rather than a point of departure. The gaggle of "me too" copy-cats have totally missed the worthy challenge that the Macintosh threw down. And that is too bad.

Research can do better, designers can do better, and the end user deserves better.