Seminal Ideas

From CS260Wiki
Jump to: navigation, search

Bjoern's slides

File:Cs260-slides-02-seminal-ideas.pdf

Extra Materials

Discussant's slides and Materials

File:02-seminal-ideas-discussion.pdf

Reading Responses

Airi Lampinen - 8/31/2010 10:23:33

These two articles that date to the eighties are outlining the early steps towards an applied user psychology and a systematic study of how users perceive and work with computer interfaces.

Hutchins, Hollan & Norman outline a framework in which it is possible to consider the costs and benefits of increasing the sensation of directness in a user interface. The article on direct manipulation interfaces discusses the phenomena underlying the feeling of directness: distance and engagement.

The authors refer by distance to factors that underlie the generation of the feeling of directness, whereas engagement is used to depict the qualitative feeling of directly manipulating the objects of the feeling of interest. As a rough analogy, the difference between them can be considered similar to that between usability and user experience. While factors that determine distance are more objective, the optimization of them alone is not enough to ensure that users will find the interface a pleasant match to their needs. Engagement, like user experience, is a more subjective, qualitative issue, that is hard to capture.

I found especially interesting the observations of how experienced users can consider even complex interfaces as direct due to having reached a level of automation in their use practices and how interfaces that give a feeling of directness for novice users may in the longer run feel clumsy and limiting. Also, overall, Hutchins, Hollan & Norman make thoughtful remarks of the pros and cons related to direct manipulation interfaces.

The second article, Card & Moran's "User Technology: From Pointing to Pondering" describes the authors' experiences of trying to understand users and their interaction with workstations. They consider different levels of interfaces: ranging from physical to cognitive and conceptual interfaces, and, to their then current understanding of the topic, the task interface.

The article is, in essence, an attempt to outline a vision of the form of an applied science of the user. The authors go through their key findings throughout their years at Xerox PARC. For instance, they describe how they figured out that user performance with pointing devices is limited by the information-processing capacity of the user and how routine human-computer interaction does not involve problem-solving but rather the execution of known methods.

They conclude by stating that the most interesting problems are likely to be found on the task level, that is, in understanding what complex intellectual tasks are like and, based on that, finding ways to build tools for structuring and browsing ideas. The authors are, furthermore, marking an important shift in their line of research by noting that instead of using existing systems to understand HCI, they have turned the table by using what is known of users to drive the creation of new experimental systems.


Kurtis Heimerl - 8/31/2010 13:59:26

Direct manipulation interfaces

This paper strongly argues for "direct manipulation interfaces"; interfaces that provide models that are directly interacted with, as apposed to programming-like "interdirect" models.

I'm not sure what to say about this paper. From a reader's perspective, it was muddled, verbose, and wandering. It seems as though they avoided the topic of when and why direct interfaces are valuable, and this is the interesting question to me. Direct interfaces greatly reduce the learning curve, and this is something they explicitly fudged their way around. However, there are tasks that are complicated or deep, and in these the learning curve is not the issue. I feel like they tried to address these points, but failed. The discussion on the UNIX/lisp model being complicated, for instance. I'm strongly in the "worse is better" camp, but I recognize the utility of having every layer available to users. This seems like the perfect solution, you give people the complexity they need, and no more.

As far as contributing to HCI, I don't buy it. There were too many thought experiments and too little clarity. Measuring this stuff is hard, users quickly optimize and the feedback loop there is a bitch. You have to measure long enough for people to become experts, and even when this is the case the debate continues (emacs vs VI, for example). This is an opinion piece in my mind, and not one of great value to me or to my research.


Kurtis Heimerl - 8/31/2010 14:09:00

I accidentally submitted just one review. Here's the other:

User Technology: From Pointing to Pondering

This paper discusses the HCI research initiatives at PARC, their progress, rationale and goals.

I found this paper immensely interesting. The vast majority of HCI "laws" came from this small group of researchers, and now I know why. They wanted to formalize HCI and create a set of rules based on cognitive reasoning. I would say they have failed, GOMS and Fitts law are not commonly used (so far as I know), and implementation is still king. However, I empathize. HCI is one of the few fields that still has user-facing metrics and all the hassle that comes with them.

There were a number of interesting discussions in the paper. The first was the notion of mental models, and having the user create them correctly. I'm unsure of where I stand on this issue. To a point, I agree. However, I find no fault in the user creating their own, different model, so long as it works for them. It may be best to design a system capable of supporting many models, rather than trying to shove the user into the one the programmer has decided is best. This could improve the generalizablity of the application as well. This is a common idiom, I feel, as users attempt to "hack" the models given to them by programmers. For instance, networked games often try to present a world where the connection is constant and without lag. Clever users often make use of this for cheating.

I also thought some of their understandings of mental models were flawed. They argued that "model-based problem solving seems mentally intensive", which is incorrect (or at least depends strongly on the model). I'm quite certain that they were seeing users who didn't trust the model that was built. Models are an experimental science in a way, you build it and run it until the models fails, then plug holes. Building models is definitely mentally intensive, but not using them.

Lastly, I feel like this paper fudged on the novice vs expert debate. They never strongly placed their intuition on either group, and I think the two build very different models. Expert users are loathe to abandon their working (perhaps complex) models, while novice users are loathe to build complicated models without testing. A thought.

This paper contributed a lot to HCI, showing systems that work, and don't. It was interesting to see the early thoughts of researchers in the field.


Thomas Schluchter - 8/31/2010 15:36:54

  • Direct Manipulation Interfaces*

The paper describes a framework for critically evaluating interfaces in which the symbolic representation of system output can be interacted with, an interaction that users experience as "direct manipulation" of the structures underlying the symbolic representation. To this end, the authors engage in an analysis of what constitutes "directness" in an interface.

The paper is relevant to basically any graphical user interface that we know today. At the time of writing, the first successful commercial GUI-based operating systems were just springing up and making their way into a mass market. They have since become the dominant way we interact with computing devices, and this paper contributes to our understanding of some critical aspects in thinking about them.

For example, although the authors don't explicitly use the term "mental model", the gulf of execution and the gulf of evaluation hint at a similar concept. The abstraction that any interface introduces into human-computer interactions relies on metaphors that the user must decipher. By framing this as a language problem, the authors make the compelling point that the user is involved in two different translation processes each time (s)he interacts with the system. Thus, the measure of "directness" becomes the cognitive effort that the user has to expend to perform the translation. Since low cognitive load seems to be universally accepted as a measure for well-designed systems, this gives us a solid understanding of how to conceptualize good design both for implementation and evaluation purposes.

Another valuable aspect of the paper is the clear analysis of the problems with direct manipulation. By illustrating the cyclical model that underlies interactions with direct manipulation interfaces -- translating system output and comparing it to the user's intentions, then formulating a course of action and translating that into system input -- the paper reveals that they are most useful for non-linear, exploratory tasks. This raises the important point that ease of use is not identical with directness, but rather a measure that must be calibrated to the task at hand.

It would be interesting to see how the authors would grapple with the more recent developments in user interfaces that add to the conceptual directness described in this paper a physical directness. Touchscreens, token-based interfaces, spatial sensor installations introduce another, or rather a different symbolic layer to the interactions that could possibly dramatically change the way we have to think about directness.

  • User Technology*

The paper reviews the efforts at Xerox PARC to develop an applied science of the user in order to support design and engineering of computer systems that take advantage of human cognitive ability.

While I have great respect for the depth and breadth of the research presented here, I am uncomfortable with many of the conclusions and underlying models. As in much of the literature on behavioral psychology to which this research seems to owe some of its strategies of inquiry, I find the reductionist tendencies disturbing. As much as the Model Human Processor might be useful to calculate performance times for set tasks, the larger question is whether processing time is an valid measure for performance. While I have no good answer to that myself, I suspect that there is a difference between an engineer's assessment of effectiveness and the user's experience of his/her own effectiveness. What the model doesn't cover are the internal feedback loops that occur in the process of learning how to operate a device/a piece of software. For expert users, there might be a sense of self-efficacy and empowerment that completely differentiates their experience from that of a first-time user. Performance would then be an evolving concept that is hard to standardize.

Another interesting observation is how strongly the motivation to found an applied science of the user seems to be tied to the desire of designers and engineers to control the user's experience. The discussion about conceptual models raises the question: Do users have to have elaborate, 'correct' ones or rather functional ones? The results of the calculator experiments seem to indicate that the most important thing for a user to have is experiental knowledge about the system that allows him/her to make inferences. Whether this experiental knowledge translates to a model that completely and accurately covers the domain of the product, seems to me an ancillary consideration when success rates of no-model users aren't dramatically lower even in "invention" mode

The huge effort invested in getting a working model out there sheds light on the importance of development practices. Although there is little evidence in the paper itself, it seems like this concern might stem from a waterfall-like design and deployment process. Prototyping as a way to test explicit and discover implicit assumptions wasn't, at least to my knowledge, common practice in software engineering at the time.

Apart from these philosophical issues, the paper remains a fascinating insight into the development of some of the heuristics that are still influential in HCI. Nevertheless, I disagree with the fundamental premise that the role of computing technology is to compensate for the "cognitive limitations" of the human mind.


matthew chan - 8/31/2010 16:58:45

//User Technology: From Pointing to Pondering

Written by Card and Moran, this paper aims to explore the history of how the authors explored the user profiles and their interaction with workstations. In particular, there are lots of details about the physical, cognitive, conceptual, and task interfaces, which the authors use to describe the origins of user psychology, methodology, etc.

This paper is pretty important because, in my opinion, it's the first of its kind to point out the importance of users and technology, namely the workstation. It has details about Card and Moran running user experiments to see how effective they perform certain tasks, whether they have a conceptual model, behavior and more psychological aspects. In fact, it seems this paper pioneered the start of the design/prototyping/evaluation model where the authors note that now learn "about users to help drive the creation of new experimental systems." (pg 197)

The paper offered several break-through results, techniques, AND methodologies. First, they re-examined Fitt's Law and modified it to better understand their experiment with Engelbart's mouse and how to improve it. In fact, this paper also highlights why the mouse has only 2 buttons instead of 1 or several more! The only limiting factor to the mouse boiled down to hand-eye coordination. Next, the authors considered the way how humans/users process information, which led to the Model Human Processor model. Furthermore, experiments were done through text editing tasks to explore the cognitive structures of users. One result was that the degree of resolution did not affect the accuracy of predictions. Also, the GOMS Model was created to define the user's cognitive skills for performing tasks and how a computer can compensate for human cognitive limitations. Next, the Keystroke-:eve; Model is introduced bc the authors wanted to assist system designers to make predictions "with a 'back-of-the-envelope' style of calculation," and the model proved to be useful in a real system design (this was used to determine the # of buttons on a mouse)! Conceptual models were also explored in which the results revealed no difference btwn two groups that had routine/complex problems; several of these experiments enhanced the creation of Xerox's new workstation. Lastly, Card and Moran heavily emphasized the ability of a user to structure and manipulate their ideas. The program "NoteCards" was important in demonstrating how users would structure and manipulate their ideas, but noted its limitations about scaling.

This paper relates to today's technologies in every way because it looks the it became the foundation of how to design workstation and devices around the user, the prevalent use of the mouse with only two buttons, the use of windows on the monitor, and many more.

This paper does not really relate to my current work. The closest would be the Model Human Processor for Speech Improvement via Games. Blind spots that i see is the use of hand-held, mobile devices where we deal with smaller screens, audio and visual entertainment, and speech interaction.

//Direct Manipulation Interfaces Written by Hutchins, Hollan, and Norman, "Direct Manipulation Interfaces" explores the advantages and disadvantages of direct manipulation through the semantic distance and articulatory distance, the gulf of execution and evaluation, and higher-level languages. More importantly, this paper scrutinizes the underlying basis for direct manipulation.

This paper is intellectually heavy and boggled my mind alot since there are lots of abstract terms and definitions. As an example, we have the conversation metaphor and the model-world metaphor to describe how users interact with computers and objects within. Next, the author explores the importance of distance defined by gulfs of execution and evaluation; systems that achieve qualitative feelings of direct manipulation are systems that makes the user feel they are directly in control of the objects. The authors also explore the higher-level model where a user might want to perform one task and the machine operations to shuffle or change bits in memory. Semantic and articulatory distance is also examined to explore input and output interface language, to figure out if it is possible for a user to say what they want and if it can be done concisely. Finally, direct engagement isn't the best for everything, especially a task that requires repetition and a script can handle that task, or accuracy. Moreover, this direct manipulation model may impede innovation.

This paper relates alot more to today's technologies due to Apple products such as the iPhone and iPad that relies solely on multi-touch and the use of fingers for direct interaction. We're beginning to reach a point where Card and Moran failed: improving the mouse. How? maybe by getting rid of it entirely!

This paper doesn't relate to my work in any way, but it is an area of interest (multi-touch, that is). Blind spots that i see relate to speech interfaces. Also, i recall reading about MIT having a sensor to track eye movement to move the mouse. Seems like "direct" is limited only to the hand or fingers and less on physical interactions in other ways.


Arpad Kovacs - 8/31/2010 17:03:08

The Hutchins, Hollan, and Norman paper is concerned with direct-manipulation interfaces, which are characterized as having continuous representation of the objects of interest, interaction through physical actions, and rapid, incremental operations with immediate visible impact. In particular, the authors examine how direct-manipulation interfaces reduce the cognitive resources required to to interact with the interface to create a feeling of direct engagement. I found the most useful part of the article to be its prescriptive solutions for reducing the gulfs of execution and evaluation by bridging the articulatory and semantic distances between the user's goals and the state of the system.

We can analyze the advantages and disadvantages of direct manipulation interfaces by comparing WYSIWYG image manipulation software such as Photoshop or The Gimp with indirect-manipulation command-line tools such as ImageMagick. I think that one of the advantages of direct-manipulation systems is that due to their graphical representations, they allow the user to rely on recognition (eg, a user must recognize which icon/button to press), rather than recall as in text-based input systems (what keyword do I need to type in order to perform a specific command?). This helps with aspects mentioned in the article, such as novice users being able to learn from a mentor, easier retention of operating concepts, and self-explanatory interfaces which require fewer error messages, since for example selecting a specific portion of the image using the lasso tool engages the user has a much smaller gulf of execution (the commands and mechanisms match the thoughts of the user) than the more abstract textual specification of coordinates and thresholds through the command line.

Another advantage of the direct-manipulation systems is immediate feedback, which diminishes the gulf of evaluation. For example, the dragging the lasso tool immediately leaves behind a "marching ants" line showing the path of the cursor, and the area to be selected, which can very easily be augmented or reduced, for example using a threshold slider. In contrast, the command-line image manipulation tool provides no feedback of the intermediate steps, and several iterations of trial and error is required to determine the correct selection parameters.

However, I disagree with the article's premise that direct-manipulation interfaces are faster for experts to use. A simple example is copying or renaming multiple files through the file browser. If I had a folder containing 1000 images of various filetype, and had to copy all of the PNGs a different directory and renumber them sequentially, then doing so through a GUI would be tediously slow compared to writing a trivial batch file or shell-script to perform the same task. Likewise, if I am trying to setup a customizable piece of software and already know which options I need (and what they are called), I find it faster to edit the configuration file directly rather than navigate through a menu hierarchy and set each individual option graphically (I can search a text-file for the line where I can set a particular option, but I'm not sure what the equivalent of grep would be for a GUI). And as for the claim that it is possible to easily define new functions using direct-manipulation interfaces, I think that most people who have tried to record a macro using direct manipulation ended up having to edit the textual instructions afterwards anyway, because it was too hard for the software to understand their intentions.

So in summary, I think that direct-manipulation is a wonderful approach for rapid-prototyping and lowering the barriers to entry for a system (eg most desktop software). However for customizable systems that have preconfigurable/known execution paths and thus can be highly automated (such as server/database administration, etc), indirect-manipulation command-line interfaces still have a role to play.


The Card and Moran paper is concerned with the evolution of user-models developed using the psychological approach to human-computer interaction by the researchers at Xerox PARC. Finding mere empirical human experiments to be deficient, the researchers decided that a better approach to understanding the physical level would be data-driven mathematical modeling such as Fitt's law for describing the motion of a computer mouse. After their success with this method, they decided to apply this modeling approach to the conceptual level, by uniting the contradictory findings of the psychology literature into the Model Human Processor architecture which can predict human performance.

I think that the most useful part of this paper is the introduction of the GOMS model, for scientifically and quantitatively evaluating expert performance based on specific goals, operators, methods, and selection rules. This system allows designers to roughly estimate how efficient it would be for a skilled user to use their interface. For example, the techniques described in the paper can a designer help determine how many options to show a user at once (as in the example decision of how many buttons to put on a mouse). On one hand, the keystroke-level model predicts that a flatter options-tree, with a high branching factor would be more efficient, since it allows experts to accomplish the task in fewer steps. However, this high level of complexity may also be a hindrance to novices, who would need more time to make a multi-option decision rather than a binary choice. In addition, it is possible that the benefits of a high branching-factor decision-tree predicted by the keystroke-level model could be negated by the need to store more chunks of information than working memory can handle at once. In this case, the Unit Task model would be more applicable, since now the behavior must be structured into a series of unit tasks separated by cognitive pauses, rather than the purely mechanical continuous flow assumed by the keystroke-level model.

Finally, I thought that the authors' mention of the tension between the users' mental models, and the designer's intended user model was an excellent observation. This potential for misunderstanding means that it is critical for the designer to provide clear, unambiguous documentation, that abstracts away the irrelevant (to the user) portions of the documentation, while exposing the important concepts in the underlying architecture. This is most apparent in the cryptic error messages that various operating systems display when something goes wrong (eg "Invalid memory access at 0x7ff815cc"). A less technical explanation would both let the user understand what is going on, and how to resolve the problem (eg "Two programs tried to modify the same file at once. Please close one of the programs and try again"). As the reading shows, even many non-technical expert users have shallow understandings of the systems they are working with; they may have a mental model of how files and folders work, but the programmer should not expect the user to understand the inner workings of virtual memory. By providing and reinforcing a simplified and possibly graphical model that shows which files are open, and which programs are trying to access each file, the model can be externalized and used to represent and manipulate ideas. Using this approach, system designers can give users a better understanding of the system, which will provide the flexibility to utilize innovative model-based problem solving when the user's cognitive skills fail to accomplish the task at hand.


Charlie Hsu - 8/31/2010 18:08:09

Direct Manipulation Interfaces

This paper describes direct manipulation interfaces and their characteristics: small semantic and articulatory distances, and direct engagement. Semantic distance is the difficulty an user has expressing his/her goals in the interface language. Articulatory distance is the relationship between the meanings and physical forms of expressions. Direct engagement is characterized by inter-referential IO (output can be part of an input expression), immediate system feedback, and an unobtrusive, seemingly invisible interface.

The paper provides a logical breakdown of the cognitive properties that direct manipulation interfaces have. The concepts of distance and engagement provide a consistent metric and clear definition for "directness". The gulfs of execution and evaluation are important to keep in mind for all interfaces, and I feel that it is always beneficial to keep these gulfs small. Consequently, semantic and articulatory distances should always be kept small as well. Task analysis is also critical, as the paper mentioned, since two interfaces that have the same domain of tasks may have different semantic distances to different tasks.

But high level programming languages, one of the paper's main examples of low semantic distance interfaces, are sometimes criticized for abstracting too much away, hiding and automating functionality in a tradeoff for smaller semantic distance. Furthermore, high level languages often have worse performance than low level languages, a problem with direct manipulation interfaces the paper overlooked: sometimes the creation of bridges across the gulfs of execution and evaluation for the user requires more resources and engineering power to implement. One clear case of this is in the area of videogames: as games move from text-based MUDs to realistic 3D virtual worlds, more and more graphics power is required, and sometimes the direct engagement feel of the game is interrupted due to performance issues.

I found the problems with direct manipulation mentioned by the paper interesting and pertinent. Direct manipulation, as the paper mentioned, requires precise control from the user. Although the accuracy issue can be addressed by constraining input (perhaps, for example, by supporting arrow-key movement of objects on a locked grid instead of drag-and-drop), repetitive manipulation can become tiresome and is much better handled symbolically.


User Technology

This paper concerns Card and Moran's attempts to develop a "science of the user", applied user psychology, that explores how to advance human-computer interaction. Card and Moran focus in particular on users performing tasks on their individual workstations, looking at the physical (pointing), cognitive (task execution), conceptual (models), and task (idea structuring) levels of the interface.

Card and Moran's analysis of the mouse as a physical interface made me think of alternatives to hand-based input. They claimed that "designing a device faster than the mouse would be difficult" because of the limiting factor of eye-hand coordination. But are there other body parts that perhaps have a faster data transmission rate than the mouse? In CS160 we briefly discussed research involving optical tracking and even brain tracking interfaces. Could these be faster? Or perhaps is the freeflow 2D pane of our typical graphical displays the real delimiter? What if our typical graphical display was more grid-like, perhaps reducing the distance between targets in the Fitt's Law analysis? What sorts of effects do things like custom mouse acceleration curves have on Fitt's Law?

I found that the analysis of the cognitive interface opened up a lot of interesting observations and methods. Using quantitative analysis with the KLM is a great way to estimate something directly useful (time taken to complete task with interface) in an economical way (measuring component times instead of full task times). I also found the need to consider working memory load during a task important. Breaking tasks down into unit chunks to avoid working memory overload is important to consider when designing interfaces, perhaps by providing constant and up-to-date feedback on the state of the system and displaying the progress the user has made on a task.

The idea structuring tool described in the paper, NoteCards, brought to mind the memex from Vannevar Bush's "As We May Think." I thought the idea of the organizational chart could use some improvement. Certainly having one primary card of focus and many others in a sort of background would be an efficient use of screenspace, but the interface would still, as the paper mentioned, be limited by a physical quantity of screen real estate. Even given an unlimited amount of screen space, an user could only absorb and be aware of a finite part of it. I feel that a better method would be to actually compress the "note cards" themselves, perhaps keeping a few in focus, but, as foreshadowed in "As We May Think", having each notecard contain hyperlinks and "trails of association" to other notecards.


Aaron Hong - 8/31/2010 18:14:07

The first reading on Direct Manipulation by Hutchins, et al. tried to net out what it meant to have this feeling of "directness." There are 2 main aspects: distance and direct engagement (metaphor of the world).

I think this is a far sighted article about direct manipulation as we know it now. In the article, they even predicted corrected that "In the end, many things done today will be replaced by direct manipulation systems. But we will still have conventional programming languages." Contrary to what Kay (1984) believe. There are still limitations to direct manipulations and that dealing in descriptions is still important.

In the second article, User Technology: From Pointing to Pondering by Xerox Park researchers talked about the history of Human computer interaction from the perspective of their history.

It's interesting how they were able to evaluation what the limitations of humans were and were able to find that the mouse was close to optimal. I think that is critical in evaluating how current user interface designs are, but it seems like a lot of it is done from intuition or black-box experimental data. More in depth research like it was done at Xerox Park would be helpful for evaluation.


Thejo Kote - 8/31/2010 18:28:21

Direct Manipulation Interfaces:

This paper is a cognitive account of direct manipulation interfaces. The authors define the factors that contribute to the "directness" of an interface in terms of the distance between what a user wants to accomplish and the facilities provided by an interface to do so, and the input and output languages involved in manipulation of the interface.

The main contribution of this paper is a framework that can be used to determine how direct a user interface is. The authors argue that to accomplish any task through an interface, a user has to cross a gulf of execution in translating their intentions into something that the computer can understand. Similarly, they need to cross a gulf of evaluation to understand and successfully interpret the output or response from the interface.

Crossing the gulf of execution and evaluation involves a number of trade offs on the part of the user or the interface. Thinking in terms of this framework allows an interface designer to consciosuly make decisions about where the boundaries of the trade offs and effort lie and the desired levels of generality and flexibility that needs to be supported.

The authors also discuss the disadvantages of direct manipulation interfaces and how they're not a panacea for all interface design challenges. They point out how such interfaces are cognitively tighly coupled with the domain of operation, which may result in the prevention of new ways of thinking that a more general system would allow.

What I found interesting while reading the paper was that plain text may not have been the most natural language to communicate the ideas, i.e, the semantic distance may have been lesser if it was possible to convey the same ideas in a different "language" involving live examples.

User Technology - From Pointing to Pondering:

This paper is an account of the creation of an applied science of human computer interaction. The goal was to create repeatable models based on cognitive theory that could be used by designers. It also explains the models like GOMS and Keystroke-Level model created to apply the science to the study of HCI. The authors identify multiple interfaces between a user and a computer, like physical, cognitive, conceptual and task based, and discuss how the activities in each span from pointing to more intellectual pondering.

Since the work was conducted in a commercial research lab, the paper contains many applications to real world usage like the Star system. It successfully makes the case for why a scientific model of HCI is important and can have business value too.

In order to model the information processing characteristics of humans the authors propose a Model Human Processor, which they use predict human performance and ability. While it may be a limitation of the medium through which they communicate these ideas, I wish they had gone deeper into this model. How does the knowledge from psychology and congitive theory translate into the model they propose? When they talk about working memory limitations in humans and use it in the models they propose, it would have been helpful if some background had been provided on these concepts.


Shaon Barman - 8/31/2010 18:39:23

Direct Manipulation Interfaces

Hutchins et al discuss direct manipulation interfaces, and analyze what factors lead to the feeling of directness. The two factors they discuss are distance and engagement.

The strengths of the paper are the ways in which they try to decompose the abstract thought of directness. By breaking down the barrier between a computer and a human into distance and engagement, they provide a framework which can be used to evaluate and compare different interfaces. The "gulf of execution" and the "gulf of evaluation" are the barriers which cause difficulty and Hutchins analyzes how information is converted in order to get past these gulfs. By minimizing distance, the user is quickly able to understand the computer. Hutchins also discusses how the user can adapt to the system, which is another way to bridge these gulfs. This concept can be seem today in most complex programs. Also, the idea of having the input and output in the same format is quite clever. This one concept is able to create a "virtual" world to the user and is essential to the feeling direct manipulation. This framework of evaluating interfaces seems quite novel for the time period and is quite applicable to modern software.

My main critique with this paper is the lack of real examples where this analysis is applied. The authors discuss a few simple examples dealing with graphs, but do not analyze a complex system. It seems like it would be quite complicated to apply these concepts to a real system and it would have been more convincing if such an example was presented. Also, they do not provide any suggestions on how to improve an interface based on these analyses.

User Technology: From Pointing to Pondering

Card and Moran discuss there experiences with user interactions when developing personal workstations. In particular, they created models of human-computer interaction in order to quantify and predict different models of interaction.

The authors drew from existing work in cognitive science in order to evaluate human-computer interfaces, which both is clever and makes their experiments more credible. I especially liked the experiment with Fitt's Law which showed that the mouse is an almost optimal pointing tool, which has been shown by its continued use today. It was also interesting to see that they used their models for real work, like using the Keystroke-Level Model to figure out the optimal number of buttons on a mouse. Another important aspect in this paper is the use of empirical studies. It was quite surprising to see that no-model and model users had the same success rate when dealing with complex problems and only differed with novel problems. Another good idea in this paper is using the computer to compensate for human cognitive limits. By creating models, they can pinpoint where humans have difficulties and then tailor the systems to compensate. The mix of theory and empirical studies seems to have been successful in this domain.

The paper lacks in the types of programs used in the evaluations. It only uses simple repetitive programs in the user studies, such as text editing or calculator programs. These programs are simple enough that all the internal operations do not need to be understood in order to use the program. It would be interesting how their models with the mouse and keyboard would work when doing more intensive work. Also the Model Human Processor seems like an arbitrary model with little empirical support. Giving more information about the theory of the model would have strengthened the paper.


krishna - 8/31/2010 18:39:43

Direct Manipulation Interfaces

In the entire paper the authors try to provide answers for the following questions: What does it mean for an interface to be direct ? Is such an interface the optimal design solution for all user interface problems ? Towards answering the first question, the authors come up with two dimensions that can qualitatively judge the directness of an interface: distance and engagement and then try to categorize interfaces based on these dimensions. As for the second question, the authors mention potential issues with such direct manipulation interfaces and the cases when they may not be the optimal choice.

A way of defining direct manipulation interfaces would be to call those interfaces direct that enable users to explicitly converse with and manipulate information while solving problems. We see that any such naive definitions do not provide insight into terms like “explicit” and “manipulate”. The authors take a more cognitive approach and characterize interfaces as direct or indirect based on the amount of cognitive resources needed to use the interface. From this perspective, more the cognitive resources the user uses, more indirect the user interface becomes. In the rest of the paper, the authors build on top of this definition. They identify and discuss those factors that affect the usage of cognitive resources and thus effect the feeling of directness (or indirectness).

The first factor the authors mention is the notion of distance between what the user wants to achieve and the requirements of the conceptual model represented by the system interface. The authors note that there are primarily two ways in which communication happens between users and interfaces - input and output. The distance factor can be seen in both directions and give rise to gulfs which characterizes the extent to which the user's goals and the systems knowledge or response differ. Specifically, the gulf of execution happens in the input direction and the gulf of evaluation happens at the output direction. The authors claim that cognitive effort is directly proportional to the extent of these gulfs. This would mean that a majority of interfaces which are developed using a conversation model require more cognitive effort than model based interfaces. Though this seems true for identically skilled users, it may be argued that these distances may shrink with skill level and that an expert user using a conversational interface might feel more direct than a novice starting to use the same interface - the authors seem to concur with this in their vi example (pp. 327)

The authors mention two types of distances : semantic and articulatory and use several examples to illustrate the same. In the Violin and Piano example they argue that the Piano is more semantically direct provided the user wants to play a single note, and Violin is more direct when the user wants to play a continuous melody. In general, from the examples provided in these sections we can garner that to make interfaces more semantically direct, we need to define a vocabulary that enables users to precisely perform a task requiring little conceptual mapping. However, in doing so we restrict the capabilities of the user and thus effect enormous semantic distance on creative and innovative tasks requiring more cognitive resources - this is later reported by the authors when they talk about the limitations of direct manipulative interfaces (pp. 336)

While discussing articulatory distance, the authors provide an example by comparing a table of numbers and graphs and argue that the graph is more direct in an articulatory sense. The authors seem to mention that reducing articulatory distance is more of a technological, economical problem than a design problem. Given the availability of touch sensitive and force feedback capable hardware these days, it would have been interesting and useful if the authors had provided formal guidelines to achieve high levels of articulation assuming the availability of these technologies.

The second factor mentioned by the authors is engagement. They describe it as the feeling the user gets when he is involved directly with objects of interest than through an intermediary. To effect the same, the authors mention creating and supporting an illusion (pp. 323) thereby implicitly stating that manipulating the user's psychology might play a crucial part in effecting direct engagement. For example, in Card and Moran we see the Billiards game simulation where it was necessary to program the mechanics of the game in accordance with cognitive expectations to create a more engaging experience.

It is interesting to note that while discussing the space of interfaces, the authors do not take skill level and training of a user as a parameter. It could be argued, as before, that expert users might find an interface to be direct even when it is conversational. For example, UNIX administrators tend to form strong associations with scripting languages over a period of time and might feel more direct writing scripts than using a graphical interface to administer systems.

The authors suggest that in systems where users perform repetitive tasks, direct manipulation systems may not be ideal - a single term in the vocabulary of the system might do the job. The authors note that in general direct manipulation systems are not accurate, this could be immediately related to the Violin example - as much as direct Violin is, it is incredibly hard for a novice to play it compared to a Piano. The authors also mention that such interfaces have the dangerous potential to over fit, not generalize and not encourage creative thinking from users. Here we see parallels in thought between the authors and Card and Moran. The overall insight is that interfaces should be direct, natural and yet capable of providing flexibility and generalization.


User Technology: From Pointing to Pondering

The crux of the paper has to be how the authors try and leverage results from cognitive psychology to try and build generic behavioral models of users. The context and timing of this work is imposing when we note the state of HCI research at that time, which the authors succinctly mention as "one sided" (pp 183). Throughout the paper, the authors discuss simple, cognitive framework(s) for modeling users and show how they can be effectively used to generalize user behaviors under varying contexts and hence productively design and evaluate user interface systems - a striking alternative to the predominantly qualitative and human factors based evaluation techniques seemingly common in those days. They provide examples of how such generalizations can be computed and discuss evaluation of models and systems using them.

In the mouse evaluation experiment, the authors show that evaluation of physical interfaces can go beyond human factors experiments. They prove that there may be upper bounds for the efficiency of an interface and that these bounds are governed by the physical and cognitive limits of humans. As a researcher, it is interesting to note the way the authors leverage on their success with the mouse experiment and start experimenting with more generic models. What we see through these discussions, models and experiments is that the authors are trying to provide an intellectual bridge between HCI designers and the mass of literature in cognitive psychology.

While discussing the cognitive model of users, the authors provide an alternate theory of how users accomplish their task. Specifically, how practice on a task would gradually change problem solving behavior into skilled behavior(pp 187). The authors suggest that for familiar problems, user behavior can be effectively predicted by their simple, abstract GOMS model, where the task is characterized by already seen goals and known methods. They provide an example - using the GOMS paradigm, the authors develop the Keystroke-Level model and use it to predict the time taken by a user to replace words in a hypothetical text editor. They atomize the replace task into a set of subtasks and assign constant times for each of the subtasks. Then, they compute the time taken for different variants of the replace task as an aggregate of the time taken by each of the respective subtasks. The authors compare their theoretically derived results with actual experiments and show that simple, abstract models of users can efficiently predict actual responses. What is not clear in the GOMS approach, however, is how to atomize a given task into subtasks. It would have been interesting if the authors provided insight on how they decided the building blocks the user might use to solve tasks in a given context.

The authors' discussion on the conceptual interface reflect their concerns on how users map atomic tasks to the concepts represented by the system. They argue that a good interface system is the one which makes this mapping explicit and easy to learn. They mention about the success of the user interface in the Xerox Star system where the design of the system was done under constraints that the conceptual model is explicit and consistent. But it is not clear as to how a consistent model could be effected on a user as users typically develop individual conceptual models that evolve with their skill level. This is somewhat clarified in the authors' experiments with a stack based calculator with trained and untrained users where, depending on the skill level of the users and the problem in question, they show that the users navigate either through a space of tasks, methods or models. Thus interface systems should be conceptually explicit and at the same time flexible such that users can map the conceptual model to spaces of tasks and improvise on them as needed and when solving unseen problems.

In the final section of the paper, the authors focus on issues in designing interface systems that enable users to be creative. The key issue, the authors say is to externalize users' ideas and then organize, give structure to them so that the users can directly manipulate them. They mention an important point when they say unlike most interfaces where the target users are controlled and skilled, users of interfaces that aid creative work will be exploratory and unconventional; in other words, the interface should be capable of handling a lot more uncertainty than in normal circumstances. They also mention that these interfaces should try and compensate for the cognitive limitations of the users - for example short term internal working memory(pp 189). However, as the authors concur, this might lead to large information structures and would need effective visualization and retrieval techniques.

An important aspect of the paper, apart from its contribution in bridging ideas from cognitive psychology and human computer interfaces, is its implicit emphasis on the importance of not just developing a theory but to provide a generic framework and give example usages so that others, specifically designers and engineers, can use it. Another important learning as a designer would be to have parallel thoughts on cognitive (or) psychological limitations of users while designing interfaces, to evaluate them on these dimensions and critically think on how we can overcome them by providing alternate physical, cognitive, conceptual or task models.

Finally, what is most striking about the models discussed in the paper is that they are all linear with no notion of uncertainty or randomness. We see a plethora of HCI literature which try to model user and system behavior using complex random processes, this paper serves as an example that simple, abstract linear equations do work if modeled and used in the right way.


Luke Segars - 8/31/2010 18:49:40

Luke Segars

User Technology: From Pointing to Pondering

The paper by Card and Moran explores the emergence of the field of human-computer interaction from the desire to strengthen the means of communication between humans and machines. The authors differentiate between different types of interfaces and provided a survey describing how each type of interface evolved over time.

This paper provides tremendous value to the field of HCI given the year it was written in. Among its most important contributions is the summary of quantitative techniques that can be used to determine how effective a particular machine interface is (mouse as an input device, simulated expertise). In addition the paper provides a historical view regarding the motivations for establishing the field of human-computer interaction and why it was necessary to distinguish it from engineering and psychology. Although Card and Moran investigate particular devices, they also manage to produce a number of models of human behavior that, if found to be consistent over time, can be used to evaluate any interaction method between man and machine. These models alone are tremendous resources for HCI research that can outlive a particular technological tool or trend.

While this survey doesn't make any particular arguments in itself, it does touch on a number of important and interesting ideas that have become central in modern HCI projects. First, it is very significant to be able to quantitatively determine that a particular device is optimal for a particular human-based task, such as the mouse as a hand-guided input device. Many activities, including those not related to technology, suffer from having rough definable limits due to their heavy dependence on the particular subjects being tested. Discoveries like Fitt's Law, however, allow us to place bounds on the possible performance of a hand-guided device and to determine that a mouse is actually quite close to that barrier. This statement is a powerful testament to the usefulness of the computer mouse as an input device, but even more importantly it gives us a means of measuring when a particular problem (or subproblem) is essentially “solved.”

Card and Moran also discuss the problem of measuring the effectiveness of a design on an expert population; this is, both obviously and interestingly, a very difficult thing to measure directly since most new systems will not have experts available. Simulating, even calculating, how experts may perform based on systems like the Keystroke-Level Model adds the same quantitative strength that Fitt's Law provides to measuring mouse effectiveness. In this case, however, it is predicting how an expert's experience will be affected before experts are actually around to test the system.

The idea of establishing and changing mental models are perhaps one of the most relevant restrictions on modern design that Card and Moran mention. As a generation that has now grown up surrounded by computers, many younger people have well-established ideas of how particular types of technology should look and operate. Mental models are, in some ways, incredibly difficult to bend and designers are often held within the overlap between the realms of what is useful and what people are willing to use.

Card and Moran's survey is a helpful summary of the field of human-computer interaction and does a good job of justifying both the importance and difficulty of the field.


Bryan Trinh - 8/31/2010 18:51:51

In "Direct Manipulation Interfaces", Hutchins et All provides a framework for designers to create more user friendly interfaces. This paper describes the various qualities of an interface that increases the directness, or more generally, usability. With these qualitative descriptions they also defined useful abstractions that are used to qualitatively measure the directness of an interface.

Like all enduring design principles, the validity comes from a myriad of anecdotal evidence that proves its validity. By evaluating interfaces of the past and present in this lens, designers have a useful framework to help them create and evaluate different user interfaces.

These principles will become even more relevant as digital interfaces spill over into all parts of our daily lives. The desktop will no longer be the dominant domain of digital interactions, instead we can expect it to surround us. There the computer will meet users who know very little about using a computer, making the need for well designed direct manipulation interfaces even more necessary.

If we take a look at the various web programming APIs and mobile applications APIs today, we can already find high level tasks that point designers in the direction of creating interfaces with the features of direct manipulation. As these technologies become more mature, I think we'll see more and more computing devices that are built for the masses being created with direct manipulation features.


Aditi Muralidharan - 8/31/2010 18:52:47

n a scholarly article, Hutchins, Hollan and Norman explore the concept of direct manipulation interfaces. Specifically, they ask what makes an interface feel direct in cognitive terms, why these interfaces can sometimes seem "tedious", and ways interfaces can be made to feel more direct.

Their starting point is the assumption that lower cognitive effort for use means a more direct-feeling interface. They then explore different ways in which cognitive overheads insert themselved into interfaces. They set out two constraints on the interface that are necessary in order for it to feel direct. First, the items manipulated by the interface must be represented in a way that allows the user to believe that they are the things themselves (an not representations). Second, the representations of the results of some action (output) must be capable of being further interacted with (i.e. being used as input). In order for an interface to feel direct, it must also have a low distance, and have direct engagement.

Distance refers to the gaps between the user's mental model, and the actions and outputs allowed by the interface. Semantic distance is the disparity between how a user concieves a task, and how the task is formulated in the interface. Articulatory distance is the disparity between the actions the user takes in the interface, and the physical form they take. For example, using a mouse to move a cursor around the screen has low articulatory distance, because both input and resut are the same kind of movement.

Engagement is a more fuzzily defined concept - it is when the user feels as if they are manipulating an object directly. ".. we know little about the actual requirements for producing it", say the authors. The hypothesize that it occurs when system's model and the user's model are close, and the level of interaction is not too low-level.

Card and Moran write another scholarly article that emphasizes how important it is to have a small distance between how a system presents itself and a user's mental model of how it works. They come to this conclusion from a scientific, empirical point of view in contrast to the Hutchins et.al. article, which is declarative.

"human factors", which are physical characteristics of humans' compuatonal and perceptive powers are at at the center of their investigations. They ask how these factors should shape the design of both user interfaces, and of studies of user interfaces. They conclude that the growing power of personal computers should support users in tackling complex mental tasks, but that the key is:

   "to find ways by which a user can act on his ideas as objects, just as current text editors allow him to act on words as objects."

or, in the words of Hutchins et.al. article, to provide them with direct manipulation interfaces.

They present the NoteCards system for 'idea processing' (an example of a complex mental task) in which their empirical principles were applied at every stage of the design process. Nevertheless, it is hard to believe that their solution was a good one because there have been many attempts at building note-taking and -organizing software in the 15 years since this article was written, but no clear winner has yet emerged.

These are insightful articles, not only because they gives UI designers a new way to constrain and classify the space of interface ideas they generate, but also because they gives us concrete new ways with which to think about our creations: "I think this input interface is semantically less distant than that one", "Mouse dragging is a good idea here because it reduces the articulatory distance" etc.

As the number and sensing capacities of input devices increase, direct manipualtion interfaces have the power to become more and more direct. Some tasks for which good interfaces have not yet been developed, such as fact-finding in unstructured text (an area of research for me), direct manipulation could offer solutions. But in what sense can one directly manipulate a collection of news articles? The question sounds interesting, but also difficult to answer.


Prahalika Reddy - 8/31/2010 18:53:12

Direct Manipulation Interfaces 1. In the paper “Direct Manipulation Interfaces” the main argument is how direct manipulation can be very advantageous. It argues that users benefit from a feeling of directness that is achieved by representing system objects as the objects themselves. This reduces the effort required of the user to use the system and thus increases the feeling of direct manipulation. The paper then talks about the possible disadvantages of direct manipulation. 2. I think the concept of direct manipulation is a very good one. It outlines what the trend in history has been so far, going from the earlier text-based machines to the graphical user interfaces, and now to multi-touch devices. With each new device that is created, the user is given a stronger feeling of being in control through direct manipulation. With the newest multi-touch devices that are created, the user directly interfaces with the objects of the system and this greatly reduces the distance that the paper talks about.

I also agree with the disadvantages that are put forth in the paper, especially those about repetitive actions. As a programmer, I understand the ease with which some tasks can be done through scripts instead of having to do it manually through an interface. 3. I believe that the paper is reasonably sound. It brings up the arguments about the “gulf of execution” versus the “gulf of evaluation” and I feel this is a very relevant and important discussion to the topic of this paper. Direct manipulation can only be argued about in terms of how well the user interacts with the system, what he expects and what he gets. As the paper mentioned, direct manipulation is similar to the concept of “what you see is what you get” in that what the user can directly interact with represents the functionality of the system. In some ways this can be very helpful for the user, but in other ways is lacking. From Pointing to Pondering 1. In the paper “User Technology: From Pointing to Pondering” the main idea is the research that has been conducted in the area of human’s personal interaction with workstations. It discusses areas of applied science that it focused on and develops 4 interfaces: the physical interface, the cognitive interface, the conceptual interface, and the task interface. 2. I liked the research and experiments presented in this paper, although a lot of it was more advanced than I have read before. I especially like the research about the “Model Human Processor.” It’s very interesting to see how the human brain can be examined to see how users think about tasks and how they go about solving them. 3. I think this paper was strongly presented. The experiments explained were well researched. It was also interesting how the paper was organized into four separate interfaces and the research was presented as such as well. It made the paper stronger and the argument was better presented.


Richard Shin - 8/31/2010 18:53:30

The first paper describes "direct manipulation interfaces"; what they are, how they are "direct", and how this directness creates a better-to-use interface which lets users better express their intentions to computer systems as well as receive feedback about the state of the system. The authors define a notion of "semantic distance" and how direct-manipulation interfaces reduce it in the user-computer direction ('gulf of execution') as well as in the computer-user direction ('gulf of evaluation'). The second paper provides a survey of prior studies in cognitive processing as related to human-computer interaction, and presents previous explorations into the human-computer interface in its different facets: the physical interface (moving the mouse pointer and typing with the keyboard), the cognitive interface (decomposing tasks into subgoals, methods, and ultimately operations like keystrokes), and the conceptual interface (mental models that users form to explain the system's behavior).

Since these papers are both relatively old (~25 years old), I thought that current user interfaces in commercial use reflect some of the results explained in these papers, especially the first one. For example, for most users, interfaces using the WYSIWYG principle dominate most of computing, whether in browsing the web, writing documents, creating presentations, or drawing pictures. As we haven’t yet read any other papers in this class, I don’t have specific examples to compare these papers with, but their main contribution seems to be in defining and refining ways of thinking about, and evaluating, user interfaces. The papers don’t present new empirical data or describe new explorations into creating user interfaces. Instead, especially in the second paper, the authors distill the results of previous work into coherent frameworks for measuring user interfaces.

Overall, I felt that the conclusions drawn in the papers were logically sound. For the first paper, I found myself generally agreeing with the identification of what exactly direct-manipulation interfaces excel at, and why these factors create a good user interface. The distinction from the interface as a conversation seemed clear, as well as the importance of being close to user goals. It seemed that experimental data to validate this reasoning was lacking, however, and I thought that if such data had also been presented, then that the arguments put forth in this paper would be more convincing. The second paper, in contrast, provides a lot of concrete data to validate the authors’ arguments. I found the models of how long a certain task would take, ranging from estimating the amount of time required to move the mouse to dividing a task into the individual user actions, helpful in understanding the conclusions presented in the paper. However, the relationship between the different ‘interfaces’ seemed unclear, and there was little explanation of why these levels of abstraction are the most useful.


Aditi Muralidharan - 8/31/2010 18:53:46

In a scholarly article, Hutchins, Hollan and Norman explore the concept of direct manipulation interfaces. Specifically, they ask what makes an interface feel direct in cognitive terms, why these interfaces can sometimes seem "tedious", and ways interfaces can be made to feel more direct.

Their starting point is the assumption that lower cognitive effort for use means a more direct-feeling interface. They then explore different ways in which cognitive overheads insert themselved into interfaces. They set out two constraints on the interface that are necessary in order for it to feel direct. First, the items manipulated by the interface must be represented in a way that allows the user to believe that they are the things themselves (an not representations). Second, the representations of the results of some action (output) must be capable of being further interacted with (i.e. being used as input). In order for an interface to feel direct, it must also have a low distance, and have direct engagement.

Distance refers to the gaps between the user's mental model, and the actions and outputs allowed by the interface. Semantic distance is the disparity between how a user concieves a task, and how the task is formulated in the interface. Articulatory distance is the disparity between the actions the user takes in the interface, and the physical form they take. For example, using a mouse to move a cursor around the screen has low articulatory distance, because both input and resut are the same kind of movement.

Engagement is a more fuzzily defined concept - it is when the user feels as if they are manipulating an object directly. ".. we know little about the actual requirements for producing it", say the authors. The hypothesize that it occurs when system's model and the user's model are close, and the level of interaction is not too low-level.

Card and Moran write another scholarly article that emphasizes how important it is to have a small distance between how a system presents itself and a user's mental model of how it works. They come to this conclusion from a scientific, empirical point of view in contrast to the Hutchins et.al. article, which is declarative.

"human factors", which are physical characteristics of humans' compuatonal and perceptive powers are at at the center of their investigations. They ask how these factors should shape the design of both user interfaces, and of studies of user interfaces. They conclude that the growing power of personal computers should support users in tackling complex mental tasks, but that the key is:

   to find ways by which a user can act on his ideas as objects, just as current text editors allow him to act on words as objects.

or, in the words of Hutchins et.al. article, to provide them with direct manipulation interfaces.

They present the NoteCards system for 'idea processing' (an example of a complex mental task) in which their empirical principles were applied at every stage of the design process. Nevertheless, it is hard to believe that their solution was a good one because there have been many attempts at building note-taking and -organizing software in the 15 years since this article was written, but no clear winner has yet emerged.

These are insightful articles, not only because they gives UI designers a new way to constrain and classify the space of interface ideas they generate, but also because they gives us concrete new ways with which to think about our creations: "I think this input interface is semantically less distant than that one", "Mouse dragging is a good idea here because it reduces the articulatory distance" etc.

As the number and sensing capacities of input devices increase, direct manipualtion interfaces have the power to become more and more direct. Some tasks for which good interfaces have not yet been developed, such as fact-finding in unstructured text (an area of research for me), direct manipulation could offer solutions. But in what sense can one directly manipulate a collection of news articles? The question sounds interesting, but also difficult to answer.


Linsey Hansen - 8/31/2010 18:55:15

User Technology from Pointing to Pondering

In Card and Moran's article, they point out that the purpose of technology is to make tasks undergone by users easier. However, many researchers still focus on the device more than the person who will be using it, and in order for the user's interactions with the device to be successful, scientists must have a better understanding of the user and the user's mental processes.

The article generally talks about how methods from cognitive psychology and AI can be and have been used to improve the interaction experience in various interface types. While in modern times most of these methods are now common, at the time this article was written, the methods were just being used for the first time, which helps me appreciate how the usability of many devices is approached today. I found it interesting that in most of the examples given, the results of testing a user's interaction countered the researchers' initial beliefs. For instance, cognitive skill proved to be the subject's go-to method for dealing with text editing as opposed to navigating through a “problem space,” and creating more fined tuned GOMS models did not really affect prediction accuracy- thus in some cases, a user's interaction with something was much more simple than designers would have originally thought. I would imagine that if a designer considered the user's interaction process to be more complicated, the method for interacting with the interface would be more complicated, resulting in a difficult-to-understand device.

Another breakthrough mentioned in the article is the Keystroke level model, which allows researchers to predict how long various keyboard tasks will take, and thus give the researchers a better of idea of how long certain interactions would take. Knowing the time a task takes in general is useful since it allows the designer to cut out overly tedious methods before ever presenting the product to a user for testing. The authors also discussed the difference between user models and intended user models, and how even though the user might not have an accurate/detailed model of what they are working with, they can still solve most problems (though they had difficulty with the “invention problems”). For this part, I was rather surprised that the non-model users did so well, though in the end I suppose that it kind of makes sense because they had their own problem space.

As far as blind spots go, I could not think of many while reading the article, since the theme did appear to be “listen to the target user's before designing some big elaborate product.” One thing I did think of however is that this article definitely encourages testing users, but it does not discourage asking too many of the wrong questions. Personally, I feel like it would be very easy for an all to eager researcher to cause a subject to over think their experience, thus providing the more “logical” answer over the more natural, correct one, which I feel like is still an issue most researchers run into today if they are not careful- I suppose that this can also arise from the researcher trying to give the user “something they don't know they want.”


Direct Manipulation Interfaces

The Hutchins, Hollan, and Norman article discusses the strengths of direct manipulation interfaces, as well as what requirements must be met to create one. In creating a direct interface, the authors elaborate on two gaps: one between the user's intentions and the machines abilities, and another between the input and output methods of the interface language.

In summary, the purpose of creating a direct manipulation interface is to have an interface that the user can easily 1) relate a task to, thus allowing the user to learn the interface and work quickly, and 2) see instant results of actions, thus allowing the user to decide whether or not an action will assist with meeting a goal. Having an interface without direct engagement just seems weird to me now-a-days since I am so used to games where if I press an arrow key I am moving a character in the direction of that key, while I know in the past one might have instead typed in a number representing some position that a character would eventually move to, or one might have selected a sentence describing an action that would then result in the device telling (not even showing) the user that the character made some sort of move after it- I suppose that in a way this example also involves some distance between my idea of moving and the machines idea of how a move is done. While the more primitive versions might have been better for the imagination, it could still be really hard to get as engaged in something if actions do not have instant gratification. In the event that there is a large time gap between the user action and the device response, it could also be more difficult to keep track of what is going on, or become easier to allow one's mind to numb over just long enough to miss some critical mishap and unknowingly continue.

As far as blind spots go, I feel like there were a few decade-based ones in the conclusion. The authors claim that direct manipulation interfaces have problems with things such as variables, accuracy, and providing new ways to think about things. While that may have been true at the time this article was written, I feel like these problems have mostly been overcome (maybe not perfectly), even though the authors seemed to imply they would be permanent road blocks. For example, looking at interfaces such as the Xcode interface builder, variables and objects are handled quite well in my opinion- this is mostly because most variables are physical on screen objects, but even updating the individual variables in an object, such as length or width, is straight forward. As far as accuracy goes, in most drawing programs the user can choose from various levels of autocorrection for their strokes, or not use any at all, and users are also given grids and other accuracy-improving tools. Then, with all of the touchscreen and movement-based technologies available, I feel like the gestures available through those allow most people to think differently about how they do a task, but that could be debatable I suppose.

With technology like the 3D tvs, touchscreens, iEverything, recent gaming devices (wii, move, kinect), modern interfaces are embracing direct manipulation more and more. As the authors stated early on in article, technological constraints, seem to be the primary thing that prevents all interfaces from being direct manipulation, and as technology improves, so does the direct manipulation of an interface. Especially with programs such as Milo (or I guess it is going to be a game now or something), interfaces are even beginning to talk back to, mimic and watch people in their attempts to better simulate the real world.


Anand Kulkarni - 8/31/2010 18:55:52

Direct Manipulation Interfaces 1. The paper argues that direct manipulation interfaces provide a feeling of directness by reducing the "semantic" and "articulatory" distance between the goals and representations of the user and those representations used in the interface. It also gives a cursory examination of disadvantages of direct manipulation systems.

2. The notion of direct manipulation interfaces existed before this paper; the paper's main contribution is an examination of two phenomena that make such interfaces effective, as well as an examination of some disadvantages of these interfaces. Understanding these phenomena can help inform the design of better direct manipulation interfaces.

The two phenomena that the authors argue make direct manipulation interfaces successful would seem to characterize the vast majority of graphical user interfaces available today, particularly in desktop managers -- in these settings, vocabulary with low semantic distance, such as "files", "folders", and "windows" have become commonplace. Moreover, actions with low articulatory distance, like dragging and dropping files to be deleted into a trash can have become the norm. The ease with which we can identify such examples throughout modern graphical interfaces indicate that the concepts presented in this paper are an accurate and useful way to characterize successful direct manipulation interfaces. Moreover, the notions of reducing these distances to bridge the gulfs of evaluation and execution (as identified by the authors) can serve as an excellent guide to the design of interfaces, even non-graphical ones such as programming languages.

3. Reduced articulatory distance and semantic distance, as the authors define them, are intuitively plausible as a source of increased feelings of direct engagement. While these statements are presented primarily through a series of examples rather than the authors' own experiments, the examples are well-sourced and believable. In particular, the authors cite enough well-studied experiments to thoroughly justify their case. However, I wish that the authors had chosen to construct their own experiment -- however, this might be asking too much.


User Technology: from Pointing to Pondering 1. The authors discuss issues around the user's cognitive and conceptual relationship to the personal workstation based on a series of experiments at Xerox Parc, presenting several models for understanding the user's behavior based on a cognitive analysis, a conceptual analysis, and a task analysis. Less time is spent on analysis of a physical interface.

2. It appears that this paper is one of the earliest to formalize the study of a user's cognitive, conceptual, and task interfaces when interacting with a personal computer; continued formal study in this area would seem to be the basis of a substantial body of HCI work and tremendously important to informing design of the personal computer. The specific contributions presented include the Goals/Objects/Methods/Selection model for representing the user's cognitive behavior, a comprehensive study of how users construct an internal model for representing the machine's workings, and an idea structuring/idea browsing model for how a user chooses to use a personal computer. In particular, this last model serves as a proposal for how future personal computers can be designed to be more useful and seems to have been incorporated into a good portion of personal computing since then.

3. The authors support their physical pointing, GOMS, and conceptual models with user experiments, although the physical pointing experiment is given only cursory treatment. The proposed task interface is not supported by experiment, but the authors describe a future project that will be able to validate these ideas. The GOMS and conceptual models are well-supported by the experimental data provided; it would have been better for the authors to extend the other two models with similarly extensive experiments.


Matthew Can - 8/31/2010 18:57:40

Direct Manipulation Interfaces In their paper, Hutchins, Hollan, and Norman describe the direct manipulation interface and provide an explanation of what accounts for an interface’s directness. Central to this are notions of distance (between the language of the task domain and the language of the interface) and engagement (level of manipulation of objects in the interface world). Though the authors note many benefits of direct manipulation, they also acknowledge that there are problems with direct manipulation interfaces. This paper makes a contribution to HCI by explaining the phenomena behind direct manipulation interfaces. In other words, the paper is an analysis of an established interface framework in HCI. There are at least two big reasons why this contribution is useful. The first benefit of understanding what causes directness is it that provides a lens with which to view and analyze a user interface. This, in turn, can lead to a way to diagnose poor interfaces. The authors give an example of a task where the user tries to control how fast the water rises in a tank where only the current water level is shown. Applying the analysis in the paper to this example, it is clear that the interface lacks directness because there is a large semantic distance between the output language of the interface and the language of the user, also known as the Gulf of Evaluation. The second benefit of a nuanced understanding of direct manipulation is that it provides a neat subdivision of direct manipulation for further research. For example, additional research could focus solely on the study of semantic distance or articulatory distance (and ways of reducing each). Or, one could examine similarities and differences between the Gulf of Evaluation and the Gulf of Execution. The paper provides a convincing argument for the benefits of direct manipulation interfaces while also acknowledging some of their shortcomings. More important perhaps is the explanation of what factors lead to directness. This breakdown of directness seems sound and principled, but it is not apparent that the paper contains an argument for why such a breakdown was chosen. One can only wonder what other aspects of directness exist that aren’t captured by the framework in the paper.

User Technology: From Pointing to Pondering In this paper, Card and Moran give a short account of how PARC has developed a science of the computer user to get a more comprehensive and thorough understanding of the human component in HCI. This, they argue, has led to the development of improved user interfaces. As the authors detail, PARC developed user models at four levels of abstraction: the physical interface, the cognitive interface, the conceptual interface, and the task interface. This research done by PARC has directly contributed to HCI because it led to the development of several models of user behavior and user cognition. As stated above, this yields a deeper understanding of the human component in HCI, whereas before this understanding was lacking. As the authors note, these models have directly influenced decisions regarding the design of user interfaces. For example, in the physical model of the user, the studies of pointer device movement revealed that the mouse was basically an optimal device because the limiting factor was in fact the user’s hand. The research presented in the paper was conducted thoroughly and validated with quantitative methods. For example, when assessing the predictive power of the keystroke-level model, the authors describe that a large set of experiments with expert users revealed a good fit between the model and the empirical results.


David Wong - 8/31/2010 19:00:10

1. The "Direct Manipulation" paper discusses the advantages/disadvantages of using direct manipulation interfaces. The paper goes into detail on how to define of the feeling of directness: the distance between the user's intentions and the capabilities of the machine and the representation of the output by systems which are intuitive to the end user. The paper also gives examples of differnt types of distance, inputs, and outputs. The "User Technology" paper discusses the advent of modeling human computer interaction and how that translates to the design of computer interfaces. The paper looks at the physical, cognitive, and conceptual interface levels and illustrates these models through experiments.

2. I enjoyed the scope of the "Direct Manipulation" paper. It thoroughly covered and analyzed the components required of direct manipulation and brought up many insightful ideas. One interesting quote was how Kay claimed that the "development of dynamic spreadsheet systems gives strong hints that programming styles are in the offing that will make programming...obselete". I found that statement particularly intruiging because things have not seemed to have progressed much since then, in regards to direct manipulation. The paper is inspiring as it mentions most interfaces have been limited due to limits in hardware and other technology. Given the rapid advances in graphics and hardware since 1985, I believe that current HCI research can implement many of the ideas brought up in the paper.

I really enjoyed the "User Technology" paper. It showed me how I have taken for granted the research into HCI. I never realized that it was so thorough and that there was a model like the Human Model Processor. The paper illustrated how human computer interaction has progressed significantly from the 1980's, for instance, with more advanced text editors and idea generation software. However, it also highlighted how HCI has not developed from that era--we are still using windowing in operating systems and the mouse. Most importantly, the paper lays the foundation on how human computer interaction can be researched in developing models of user behavior and using those as evaluation benchmarks--an idea I had not considered in the past.

3. I believe that the arguments brought up in the "Direct Manipulation" paper are sound. Although they did not cite any explicit experiments, all of their conceptual models made sense. They address a problem that was prevalent in the days before graphical user interfaces when text based input and output were the only means of manipulation.

I believe that the argument from the "User Technology" paper is sound. Their models were supported by data and their experiments seemed to account for confounding factors. The problem was well-motivated, at least for their time, and their solution was both novel and revolutionary. I wish, however, to hear about more related work. Where their approaches completely novel, or was there similar research going on at that time. Also, how did their results compare to the results of other researchers.


Brandon Liu - 8/31/2010 19:00:43

First Reading

The author’s main point is to explain why direct manipulation is desirable, because it goes against the belief that computer systems should be increasingly general and abstract. The article is valuable in helping to understand the trade-offs made in direct interfaces and to better understand what makes them powerful. The result of the paper is an intellectual framework in which to understand whether or not an interface is direct.

From a historical perspective, the authors are overly optimistic when they quote Alan Kay on programming practices, since we aren’t all programming by connecting images together. Since projects like this always pop up (See Google Android’s app inventor and David Pogue’s review) it would be a worthwhile research direction to understand what keeps these projects from getting more attention.

One thing that could have been better is a discussion of the advantages of conversational interfaces vs model-world interfaces. The major weakness of the paper is in addressing ease of use. The authors acknowledge that greater directness does not imply greater ease of use. They are not clear on whether directness or ease of use is to be optimized. For example, in the real world, it takes roughly twice as much effort to draw a square with area 4 than a a square of area 1. In a direct manipulation interface, this doubling of effort is present, while in a less direct interface, the relationship is more flat. I would have liked to see more discussion of the efficiency of user interfaces.

Second Reading‚Ä® The authors contribute to our understanding of HCI by developing a framework for evaluating the performance of an interface, and explaining the relevance of conceptual models. They are surprised that conceptual models are never explicitly documented, and do some investigation into the range of models that actual end users have.

Something I liked about the paper was it’s willingness to use experiments to back up its claims about mental models. Another part of the author’s support for their arguments is from their experience with the Alto and Star computer systems. They provide real world examples of how their ideas fed into the design of a product.

Another part of the paper I liked was the discussion of the ultimate goal of these interfaces. The authors envision human-computer interaction as making up for human deficits in information processing. They present brainstorming, and an example of a lawyer looking at ideas, as an example. One present day system that implements this that wasn’t around at the time of this article is version control systems for computer code. By making it very cheap to branch code, programmers can do “exploratory programming”. A research project related to this is creating version-control like systems for writers or artists and other content creators to minimize the cost of doing exploration.


Drew Fisher - 8/31/2010 19:01:24

Card and Moran analyzed what issues exist that cause slowdowns for computer use by both familiar and unfamiliar users. They found that the mouse generally operates as quickly as the user's brain can keep up with, approaching the limits of Fitt's Law, making it nearly perfect. They modeled a human as a computer, and found how the human dealt with "hardware" limitations. Finally, they showed quite clearly that the users that understood the model of a system were able to use it far more effectively, but prefered to learn rote action sequences that solved particular known tasks.

I was particularly pleased with the experimental data describing time to complete tasks. The GOMS model stuck out at me as trying to cram a human into a computer model, rather than adapting the model to properly fit the task at hand, but it seems to have worked well enough for this study's purposes, as supported by their experimental data. By thinking about people in this way (processors with separate motor control, I/O, and limited memory), we can analyze what tasks may be more efficient than others in an automated and quantitative fashion. Such a model could be applied to almost any design problem to predict user time consumption.

I would have loved to see data comparing skilled users with novices at the manuscript-editing task, although the paper did indicate that the instability of the behavior would have made it difficult to draw any clear conclusions. I also found valuable the proof that a correct mental model lead to greater efficiency and success when working with computer systems. This can and should be taken one step further - tell interface designers to ensure that the intended user's mental model reflects the model that users will construct of the system on their own, by exploring the system. It also shows that a small amount of instruction (describing the model) can make users vastly more effective, suggesting that if users want to be maximally productive, they should try to obtain sufficient documentation on the system to understand the intended user's model. Since this is rarely how people actually use software anymore, it emphasizes the importance of making intuitive interfaces that imply an accurate user model.


Direct Manipulation Interfaces is a paper on interfaces that allow to the user to easily express his goals to a computer system. It provides an extensive set of vague, qualitative metrics that do a poor job of highlighting the central point: that good interfaces present exactly the functionality the user wants at any given moment in a given system.

The paper reads like theoretical psychology - the authors propose a variety of ways of thinking about problems, but then appear to fail to back them up with any data whatsoever (perhaps data is available in the papers they cited). I see nothing quantitative in the entire paper, nor any examination of the relative importance of any of the "gaps" they mentioned. Since the purpose of the paper appears to be analyzing what makes interfaces intuitive, it seems the value in the paper is in providing a starting point for other researchers: a list of metrics for other researchers to analyze on a quantitative basis.

While this paper discussed some interesting ideas, I would have liked to have seen some quantification or even relative comparison of any of the "distances," backed by quantitative data, say, showing that greater semantic distance results in poorer user performance or experience. The paper makes a great deal of arguments, but they are either 1) taken from other work (not novel), or 2) purely logical and unsupported by data. When theorising about illogical human beings, we cannot rely on logic alone - experimental data must be provided.


Pablo Paredes - 8/31/2010 19:32:51

Direct Manipulation Interfaces

The paper describes the overall definition of directness as an inverse function of the cognitive resources required to accomplish a goal by the use of a (computational) system or tool.

I found extremely relevant the view of defining the interaction from an I/O perspective, defining the execution (input) and evaluation (output) gulfs that separate the user from the fulfillment of a task, showing the level of “intrusion” an interface has into this process.

Concurrently defining the semantic and articulatory distances that define the previously mentioned gulfs helps characterize the interface language via the inverse relationship between learning/planning time versus its generality, and how there is always a trade-off that has to be assumed and measured when defining new interface languages.

Finally, understanding that the overall experience could be defined as an illusion, and the necessity to maintain this illusion alive through a clear form and speed of feedback as well as a continuous representation of the system state, puts in evidence the technological needs to advance technology (and its abstraction level) to evolve to new forms of interfaces that define a closer direct engagement model.

I wish there was an evaluation in the paper on potential behavioral variations on the way people perform similar tasks, i.e. considering potential disabilities (mental or physical), as well as other types of behavioral altering situations such as culture, environmental limitations, etc. This would have provided a more complex yet complete analysis of direct manipulation under other levels constraints originated in the user side.


User Technology: From Pointing to Pondering

The paper describer the circular notion that experimentation with user interface systems renders potential definitions of enhanced or evolved new systems, and encompasses this in a more formal definition of user technology, defined such as the usage of hardware and software that help make effective user interfaces.

I find interesting that the paper describes the advantages to model humans via a cognitive psychological model, in order to improve the objective evaluation of the effectiveness of a new user interfaces, in terms of parameters such as speed and effectiveness of task execution (i.e. Fitts’s law).

I like the notion of acknowledging different levels of interface (physical, cognitive, conceptual and task) and how they relate to different underlying models (human processing ability, engineering model, mental model, nature of intellectual tasks) in order to use this holistic view to understand the design trade-offs attached to different user interfaces.

I find interesting to see how understanding the specific tasks emerges as the most relevant design aspect needed to correctly map a user cognitive abilities/limitations in order to define the path for evolving towards new user interfaces.

It would have been interesting to discuss the effect some other extrinsically aspects, such as economical or cultural constraints could affect the design of user interfaces, i.e. how could design trade-offs could be chosen/altered given some boundary conditions for actual technology implementation.


Siamak Faridani - 8/31/2010 19:39:27

Article 1: Direct Manipulation Interfaces

To me the paper seems to be one of the original papers on defining a mutual vocabulary to HCI research. Through the paper and while they talk about direct manipulation interfaces they end up introducing a number of interesting concepts. Perhaps many of the ideas that we now see as common industry best practices (like "don't make people think") might have been originated in this paper.

The authors start with examples of direct manipulation. And briefly review the works by Shneiderman (1974, 1982, 1983). They present why the idea took many years to start showing up in modern user interface. They then go ahead and define the three properties for a direct manipulation system. Two very first terminologies/concepts that they define for directness are distance and engagement. Through which they introduce the idea of gulf of execution and the gulf of evaluation as a distance between the physical system and the goals.

I was particularly surprised how well they were able to explain some of the modern trends in UI practices by just defining how we can bridge the gulf between the intention of the user and the specifications required by the computer. For example one solution that they provide is to use higher level languages that abstracts the complexity and contains structures for frequently encountered problems. Another one was providing quick and direct feedback. But finally they mention that the user is the one who ultimately bridges the remaining gap. Sometimes users have to change their own conceptual understanding of the system in order to span the gulf and think the way the system is thinking.

Another concept that was introduced in the paper and I enjoyed was the direct engagement concept. They introduce it as a powerful tool to take the computer out of the system and give the feeling that the user is manipulating the objects in the physical world. Rapid feedback rises again and is mentioned as a control mechanism that allows the user to correct her task while they are being executed.

Figure 7 is particularly very telling and it defines the properties of the system as a function of engagement and the distances from user goals.

As concluding remarks authors conclude with pointing out the deficiencies for direct manipulation. For example they remind us that repetitive tasks can be better done via scripts than visual interfaces. Perhaps this might be why many of us like bash scripts, or perhaps this might be the reason Microsoft has released its powershell programming script.




Siamak Faridani - 8/31/2010 20:01:10

Article 2: User Technology: From Pointing to Pondering

The article is a report on the inception of a new science. Authors report on their efforts and struggles in formalizing user studies. They argue that computers have been studied for many years but users, who are the actual audiences for these computer programs, are yet to be understood. They combine applied user psychology, human factors, mathematics and computer science to come up with models that explain human behavior.

Through the paper authors introduce a number of interfaces: The physical interface, the cognitive interface, the conceptual interface and the task interface. The rest of the paper falls into either of these interfaces and they use these concepts to explain different aspects of user behavior. What I really liked about their paper was their scientific view about mathematical foundations of user interface, for example they use the Fitts model for the pointing device and argue that the mouse is performing very close to the optimal model. They conclude that this optimality is the reason why we cannot develop a device faster than a mouse. While I believe it is an interesting argument I also believe it is inaccurate and this inaccuracy is an inherent property of their categorization. They never consider that the physical interface may be combined with the cognitive interface (interfacing between brain and computers) and result in a faster and more optimal interaction.

Another interesting aspect of the paper is their view of the UI design as a procedural task. In the mental models they make an effort to establish a framework for thinking about designing user interfaces and they introduce ways to measure the performance of the UI. Another aspect of the paper that seemed to be original and has become a trend in CHI practices is that the way they conduct their user studies. I was surprised that in the 80's they had already have procedures for watching users interact with systems, and for collecting precise behavioral data from these interactions. For example in Figure 7 they do a quantitative comparison between 3 different methods that accomplish the same thing and provide a quantitative framework to determine the best method.

The paper offers an historical review of how this "Applied user psychology" science was established and provides the fundamentals that an HCI researcher may need to know (methods for designing of UI, user studies, user data analysis and even design of experiments)


Dan Lynch - 8/31/2010 23:34:26

Direct Manipulation

The first article written about Direct Manipulation Interfaces attempts to characterize the space of interfaces. The article uses the terms distance, engagement, articulation, intentionality, and semantics to delineate what the meaning of directness means in the context of human computer interaction.

This is a very important article for obvious reasons, one being that computing today is ubiquitous and all tasks require human-computer interaction. The idea that we have in our mind and the task that we wish to perform must be as close as possible to what the computer can display as output (the gulf of evaluation), and the action of carrying out these tasks must be executed in a fashion that best represents the tasks at hand (the gulf of execution).

The gulf of evaluation is the distance that spans both semantic and articulatory distances, and can be used to analyze the usability and functionality of the overall system.

I had problems with an interface today using Eagle CAD to design a printed circuit board. My intentions were to simply connect a capacitor to a ground component, but it would not connect. The semantic distance was huge! I was clicking the wire on the actual component where I should and it did not work as expected. This was a major problem and after some googling I found that changing the grid size can cause results of this nature.

Semantic distance is a term the author uses to describe "the relationship between user's intentions and meaning of expressions, articulatory distance has to do with the relationship between the meanings of expressions and their physical form." I interpret semantic distance as being how far away the semantics of the implementation is from what the user wants to do. The articulatory distance can be viewed as how far away is the actual interfaces physical presence from the meaning of semantics of the operation.


User Technology

This article codifies the human in a way that enables us to characterize and classify tasks and their respective cost during human-computer interactions. On the highest level, there is the user, the system, and the task. Then there are four interfaces that the author mentions, which are relationships between these: the physical interface the cognitive interface, the conceptual interface, and the task interface.

The best part of this article in my opinion is the part where he mentions that the mouse was supposed to be an interim device, a device that we are not supposed to be using anymore. I completely agree with this and at times I find myself using the touchpad on my laptop more and more. Its way more intuitive and sensitive. However, there are times where the mouse is still required.

Onto the topic of the paper, I think that the issues brought up in this paper are important because they can apply to many areas of research. The author discusses the human thought process while interacting with computers in terms of goals, actions, methods, and a selection process.

These characterizations allow us to study the human thought process quantitatively. This is a novel concept and method for studying the mind and computer science. In addition, it was interesting to find that during a study of an electrical engineer working with a CAD program, you can see the trend of the thought process: the human must think about his/her task, and then performs the task---this is a realization of working memory and the limitations of the human brain and thought process.

This article is relevant in that it not only provides a vocabulary for characterizing the efficacy of a user interface but also a method for quantifying the interface.



Luke Segars - 9/1/2010 0:30:40

Luke Segars

Direct Manipulation Interfaces

This paper discusses a class of interfaces that are aimed at making interacting with computers more natural for users. The authors mention a few theoretical projects that would exhibit high usability scores and then analyze the theoretical interfaces in terms of their semantic and articulatory distances from a user's real-world experience.

This is actually a very exciting time for discussing the these ideas given today's culture's recent emphasis in direct manipulation interfaces. The authors describe how minimizing the “distance” between a virtual experience and everyday experience increases the usability and usefulness of a particular tool or language. Modern culture has fallen in love with Nintendo's Wii, perhaps primarily because of its unique sensor-based input method; the Wii takes the common experience of playing video games or virtual sports and brings it closer to the real-life experience that the game is simulating by allowing players to use physical actions that are then translated into the virtual world. A staggering number of similar devices are now emerging from Nintendo's competitors to strengthen the connection between playing on their systems and performing real-world actions. Apple's iPod Touch devices also provide touch screen technology that allows players to flip through lists, change songs, and otherwise control their device through realistic gestures instead of mouse clicks or spinning wheels.

The principles of direct manipulation are also becoming a big issue in the data management realm. Microsoft Surface is a specialized table that is equipped with a pressure-sensitive screen and the ability to visually represent a wide range of data. John Underkoffler is a researcher in the MIT Media Lab and described groundbreaking gesture-based navigation of visualized data in a TED talk earlier this year.

What is it that these major companies are striving for? Their efforts all seem to be targeted at making direct manipulation a reality. Although many of these tools haven't become mainstream yet, there is little question that the articulatory directness provided by tools like Microsoft Surface will be appealing for a wide range of uses. Interestingly, the importance of data visualization and manipulation has become increasingly difficult as the quantity has increased; until recently we've been using a reasonably antiquated system of “files” and “folders” to organize information --- I personally wouldn't be particularly surprised if the presentation of information lost this structural definition in the coming years. In As We May Think, Dr. Bush describes how this “binning” approach to organization is not how the human mind tends to catalog knowledge; as the distance between human and machine continues to shorten, useful presentation of information will become just as important as the information itself (for reasons described by Dr. Bush).

One interesting observation that the authors make is that people are most capable of dealing with new tasks when they are presented with an interface that makes the operation as close as possible to a non-virtual experience (resembling what the authors call a “model-world”). However, they can learn to use even unintuitive interfaces, even begin to think they're intuitive, after they have performed the particular task a couple of times. Changing a person's mind about how a particular task can, in some cases, be difficult given that it requires a readjustment of a mental process that had become near-automatic in their mind.

I believe that this paradigm shift from point-and-click interfaces to direct manipulation interfaces will require time for users to make. Nevertheless, the arguments in the paper seem to be very well-founded, and we have already seen success in a number of products that use subtle direct manipulation techniques. As newer and cheaper hardware become available, the move to direct manipulation may very well be a critical technique for dealing with floods of information. “Direct engagement” will also become more and more desirable as larger numbers of people involve the digital world in their daily routines.

It would be interesting to read a 2010 account of the progress of direct manipulation interfaces. I found it a bit disappointing that the authors didn't include any results from actual experiments; I assume that this was due to the lack of available hardware and processing power given the time the article was published. The theory seems to make a lot of sense, but nevertheless it is just that: theory. Even today, I think that most people consider direct manipulation techniques to be tools of science fiction writers, certainly not something that they would ever use. Nevertheless, many people are already using similar interfaces in tools like the Wiimote and the iPod. A modern-day account including both theory and experimental evidence would be even more powerful than this paper was.

Overall, the authors make a fantastic case for the promise of direct manipulation interfaces. As new interfaces are introduced into the general public, people tend to agree with exactly what this paper states: make it feel like they are actually doing something (whether it's flipping through an album list or playing tennis) and they will enjoy doing it more. Twenty-five years after its publication, some of the seminal ideas included here are finally ready to be experimented with. So far their words are ringing true.