Direct Manipulation for Comprehensible, Predictable and Controllable User Interfaces
by Ben Shneiderman
Review by Jeffrey P. Bigham

Read Paper on CiteSeer

My Review and Thoughts

In this paper, Shneiderman claims that the best user interfaces are comprehensible, intuitive and predictable. He argues that these goals are most easily realized when objects of interest are directly manipulated as opposed to when computerized agents are relied upon. He urges interface designers to step away from the allure of the fancy robotic friends that have failed so often in the past and stick to simple interfaces that obey and reflect our intuitive notions of how objects should react and conserve state.

An underlying assumption that Shneiderman makes in this paper is that the only predictable aspect of intelligent agents is that they always fail. Indeed, it seems likely that very few would argue for them given the track record of the most popular. What is more disturbing, however, is the notion that he seems to be advocating that there is something fundamentally flawed with the idea of an intelligent agent guiding a user in his task.

Contrary to Shneiderman, I claim the power of computers in dealing with information and making our lives easier extends far beyond the ability to repeat actions quickly. It lies in the ability to efficiently abstract away complexity while retaining the ability to execute complicated queries and spawn intricate procedures. His examples may have provided good goals for the end interface that should be presented, but were hardly convincing that such functionality can be achieved by such interfaces unless they employ underlying intelligence.

Certainly there are bad user interfaces out there and many of them try to harness intelligent agents that just aren't quite yet up to the task, but that doesn't mean the idea is bad in principle. For most tasks a more direct and natural way to interact spatially with the data is a worthy aim; unfortunately, the arguments Shneiderman offered to support his notion that agents are a bad way to go only convinced me that most of the agents don't work that well and shouldn't be deployed until they have been shown to work well.

The most important contribution to HCI offered by this paper is simply that interfaces that are more complicated, less intuitive and unproven should not be preferred. I would suspect that this Occam's razor principle has long been a component of solid HCI design. Until the intelligent agents that Shneiderman so despises come into their own, perhaps the best compromise is to present their "helpful" guidance alongside the underlying objects and in a way that remains intuitive and consistent with Shneiderman's principles of direct manipulation interfaces.