Where the Action Is.
Paul Dourish (2001)
As hard as it may be for modern computer users to believe in this era of smartphones, tablets, and touchscreens, early computers were not interactive. They were used to read files of input data, execute well-defined computations on that data, print or record results in new files, and then quit. A few examples of users actually interacting with a computer appeared during the 1960s as part of such projects as Ivan Sutherland’s Sketchpad, but the event most often credited with launching the interactive-user paradigm is Douglas Engelbart’s demonstration at the 1968 Fall Joint Computer Conference of work done at the Stanford Research Institute (SRI). The demonstration is often referred to as “the mother of all demos” in recognition of the number and range of innovations it showcased, including a variety of interactive applications, a mouse, a chorded keyboard, hypertext links, windows, and video conferencing (Engelbart and English 1969). The resulting paradigm shift paved the way for the Apple Macintosh, Microsoft Windows, and the WYSIWYG (what you see is what you get) paradigm that prevails today, in which the user continuously interacts with the machine to create or edit data, launch analyses, or view results.
From tools to accomplish specific narrowly defined computational tasks, computers have morphed into assistants with whom we communicate almost continuously about a wide range of daily tasks. This has been paired with sophisticated software and ever-greater use of computer-controlled devices, from electronic communications to robotic assembly and computer-numerically-controlled (CNC) fabrication devices. In this transformation, the “Run” command has been replaced by a complex interactive relationship with multiple virtual tools that demand, consume, and direct attention, mediate interpersonal relationships, and permit us to edit or control complex data sets and processes anywhere on the planet with the result that computers form an integral part of systems for air-traffic-control, manufacturing, subway systems, and buildings. We are increasingly aware of the importance, limitations, and affordances of the input and output mechanisms that constitute our interface with the machine, leading Paul Dourish, a prominent researcher from Xerox’s famed Palo Alto Research Center (PARC), to title his book on the subject Where the Action Is (2001).
As computers become more central to design processes, and as buildings are increasingly recognized as significant computer input and output devices, the relevance of these topics to architects and other designers becomes more and more clear. This chapter will review some of the fundamental concepts, challenges, and opportunities of current and emerging interfaces linking designers to data, examining the direct manipulation paradigm, limits and use of gesture and vision, intent and meaning in drawing, computer support for design teams, and the potential of buildings as interfaces to their own operations.
While Engelbart’s chorded keyboard, on which patterns of simultaneous keystrokes permit efficient typing with one hand, never caught on, most modern desktop computers do use a mouse and keyboard for input, and create visualizations on one or more high-resolution raster graphics displays. High-speed networks link machines together, and many devices now include cameras and microphones for human-to-human communication. Mobile devices add more sensors, such as accelerometers and global positioning system (GPS) antennas, in a pocket-sized form-factor with high-speed wireless and cellphone network connections that allow these devices to track and monitor as well as interact with us. But let’s focus on the interactive paradigm first. We will look at computers as communication platforms later in the chapter.
An interactive computer maintains some sort of representation that users manipulate (“view” or “edit”) by communicating commands and parameters using mouse and keyboard, using what is called direct manipulation (Shneiderman 1982). Results are viewed on the screen and available for additional manipulation. The representation might be used as input for a separate computational process (e.g., rendering), or simply displayed, stored, and retrieved (e.g., word processing). In recent years this definition has expanded with the addition of voice and gesture input and “displays” (outputs controlled by the computer) that include building ventilation systems and smartphone icons, as well as traditional desktop graphics.
The issues associated with human–computer interaction (HCI) impact design in two ways: Designers interact with computers during design, as we have discussed in earlier chapters, and building occupants interact with surroundings that are increasingly incorporating sensors and computer systems that respond to their presence, as we will explore in Chapter 13. How meaning (commands or results) is conveyed between human and computer elements of design systems is of critical importance.
Design computing is concerned with both types of interaction under the broad subject of HCI, a subdiscipline of computer and information science that overlaps design computing in areas related to interaction and interface. While much HCI research is focused on efficient user interfaces for programs, researchers are beginning to address questions such as: Can your assisted living apartment help assess your cognitive state or turn off the stove if you leave it on? If each light in your house can be individually computer-controlled, and your house has occupancy sensors in every room, do you still need light switches? If you do away with switches, is occupancy enough, or should users talk to the lights, snap their fingers, or clap? Can a smart house learn from your past behavior and deduce what you need, perhaps holding phone calls or email if you are resting, or calling for help if you fall?
During design we use the computer interface to manipulate a representation of the design artifact, whether that is a drawing or a BIM model. The intended meaning, or semantic payload, of the representation is never entirely explicit. Meaning is often left to the cultural norms of the ultimate data users, or used and abandoned in the process of data entry. Further, as digital fabrication becomes more common, ironically, the object is no longer the product of design, a data file is. In the even more abstract case of “mass customization,” a family of objects, represented by user-specified parameters and designer specified process, produces objects that may not have been foreseen. The design itself becomes implicit.
In the transition from hand-drawn to CAD-produced construction documents, one critic complained about the loss of “latent content” in the drawing of a detail. The claim was that a manually well-drawn detail with good line-weight control, lettering, and layout revealed a skilled hand, which implied a depth of experience and knowledge that the contractor could rely on. Because some physical attributes of a CAD-drawn detail (line-weight and lettering) no longer reflect the experience-level of the drafter, this critic felt they were inferior—the technology was artificially delivering something previously left to skill. However, while the observed change is real and CAD software does produce greater uniformity in drawings, skill and experience still reveal themselves in the organization of the drawing, the content shown, and in the operator’s control of parametric features such as pochés or dimensions. In both cases there is more information conveyed than the face-value content of the drawing in question.
Designers organize space. Drawings as visual objects are largely about space, so drawings are an important part of design, but as we’ve seen in earlier chapters, neither CAD drawings nor BIM models are terribly good at directly representing space. Space is what happens between data elements. Interestingly, the user interfaces of most drawing and modeling programs don’t directly support the organization of space either, though some emerging research is re-asserting the importance of such approaches (Jabi 2016).
Most drawing editors operate using direct manipulation, a term coined to describe the style of computer interface in which a user is able to select and manipulate elements of the display using only a mouse rather than typed commands (Shneiderman 1982). Most of us are familiar with direct manipulation in the form of word processor or drawing programs that highlight or display “handles” on objects as we manipulate them, even if we don’t know the technical term. Direct manipulation makes the human user into the central change-agent in the data, able to observe each change both before and during the transformation. It also creates a precision problem.
The precision of direct manipulation depends on how well the editing gestures can be tracked, both by the computer and by the user. The geometry of the human body and limited muscle coordination mean most of us cannot draw a perfectly straight line—that’s why there are straightedges. Nor can we draw easily two lines that end at precisely the same point. This is even harder when the screen you are viewing represents something the size of a building, and you require absolute precision. On top of this screen displays are pixelated, with each pixel representing a range of coordinates. Even if you have the hand–eye coordination to position the mouse cursor on the single pixel at the very end of an existing line, your point may not correspond to the same real-world coordinate as the existing endpoint. Because the lines mean more than the drawing, the intended result is often that the line needs to be constrained in some fashion above and beyond the direct sensing ability of the hardware.
Constraints reflect a designer’s deeper understanding and organizational intent. But while designers may speak of symmetry, regulating lines and alignment in a design (Kolarevic 1994), drawing and geometry modeling environments control geometry largely using individual points in space, following the direct-manipulation metaphor. The enforcement of organizing concepts is left to the designer, not the software.
A limited suite of “pointing aids” in most CAD software facilitates precise pointing, including “ortho” mode, “grid snap,” and the many “object snaps” (endpoint, intersection, etc.). These software filters (shown in Figure 1.3) correct for imprecise mouse, finger, or stylus input, constraining results and inserting meaning where hand–eye coordination fails. More advanced filters, or inference engines, allow software to offer hints or suggestions consistent with traditional organizing principles, displaying horizontal or vertical lines or endpoints related to points in the cursor’s neighborhood or elsewhere in the drawing.
These input aids are often transitory. That is, in many programs “snapping” the end of one line to the end of another makes their endpoints coincident, but it doesn’t create or encode the intent that they stay so. Similarly, making a line parallel or perpendicular to another, or offset a specific distance, doesn’t mean it will remain so in the future. Such drawing aids can be used to efficiently make new data from old but without adding structure to the new data. This limitation is implicit in the representation of a drawing as a list of primitive elements. Fortunately, there is increasing recognition that constraints are important and explicit constraint manipulation is now available in some tools, though visualization and editing of constrained systems remains a work in progress.
The theory of affordances (Gibson 1986) leads software developers to make editing options visible within the interface. As a result, modern user-interfaces are festooned with tool icons and command menus. In the absence of a fixed design process, the interface offers a very broad range of tools to the user, based on the software designer’s model of what the user is likely to do and the tools they will need. Well-designed tools incorporate disciplinary knowledge and culture in order to increase efficiency, but as they get more specific they also tend to be less flexible, and an inflexible tool cannot be used in unusual ways. The extent to which CAD or BIM software is guiding or limiting designer actions can also be a problem because users tend to do what is offered and may mistake action for forward progress in the design. Unlike the blank piece of paper demanding focus on unresolved problems, the software interface can invite action without focus.
Meaning in drawings arises from both conventions of architectural drawing (e.g., plans and sections are drawn as if cut horizontally and vertically parallel to primary elements of the building) and the specific contents of the image (a wall in plan is shown with a line on each side, at least in part because it is a vertical element and the plan-view line represents it well over its entire height)—a tilted wall would look quite different in plan. To the extent that a program seeks to enhance drawing productivity by assuming certain geometric relationships, or incorporating template graphics such as doors, toilet partitions, and cabinets that rely on convention, they subtly nudge designers in the direction of routine design. When using software such as BIM, that supports and enforces more complete architectural semantics and comes with many pre-defined parts including, perhaps, preset constraints, these benefits can only be realized if the design intent conforms to the expectations of those who created the constraint system, or if deviation is very easy to perform.
Even when an evolving design is very conventional, the designer may not have precise answers to wall-thickness or material-selection options at the time they sketch the plan. Tools that demand such answers before work can be done are frustrating, even when the answers are known, and can reduce productivity because they break concentration (Csikszentmihalyi 1991). Inappropriate nagging is one of the common complaints about BIM software today, contributing to the view that BIM is best applied to later stages of design, not during schematic design.
The field of design presents both one of the most difficult and promising opportunities to combine human and computer strengths. The complex mix of spatial details and cultural symbolism, of predictable pattern and delightful novelty, benefits from both a machine’s ability to organize and retain detail and a human’s integrative pattern making. The iterative process of design, operating across a variety of scales over a period of time, means that there is no tidy one-time handoff of responsibility from one partner to the other, or one representation to another. Instead, we see a complex interaction, a process of hypothesis, exploration, analysis, backtracking, and synthesis in which human and machine accept, process, and respond to information provided by the other. If the interchange can be organized well, it promises to be a powerful, fruitful collaboration that helps the designer to achieve the state of focused concentration and productive output called flow (Csikszentmihalyi 1991). Alternatively, designers could find their thoughts interrupted by ill-timed requests for information or offers of aid, a catastrophe illustrated by Microsoft’s 1990s experience with an automated assistant in Word—Clippy (Robinson 2015; Gentilviso 2010).
Representations appropriate to the cognitive state of the designer are one challenge, but it is also important to uncover and make available modes of interaction that fit well with the work being done and others who may need that data. Over the lifespan of a building project, multiple participants will need to be brought onto the team and their input will need to be integrated into the overall project. Managers, bankers, clients, code checkers, and maintenance crews each have their own view of the building. Some users will be infrequent, or concerned with a defect or operation at a specific location in the building. Others may be monitoring progress during construction, or curious about the current state of the various building systems. Most of these won’t find a BIM editor useful, but the information they consume and create might well be part of the BIM. Providing them with appropriate interfaces is an ongoing challenge.
During design, architects often straddle the boundary between digital and physical, frequently shifting back and forth between different representational media in order to gain insights and test conclusions about their designs. Sometimes these inquiries can be met by transformations within the medium, as when digital models are rendered, but sometimes they cannot. At such times it isn’t uncommon to build physical scale models or prototype full-sized components. At a small scale, digital fabrication offers an enticing option—3D printing a physical model directly from the digital one. While far from instantaneous, inexpensive, or problem-free, this workflow is increasingly robust. The resulting physical model can provide tactile feedback, can be appreciated by multiple participants in a meeting, shared without technological paraphernalia, and modified with traditional tools if necessary.
However, it is also desirable to move from the physical world to the digital one, to (re)integrate design executed or modified in the physical world into the digital for further work; this is harder. Experiments with scanning, or smart construction kits that communicate their physical configuration to a computer (Gorbet and Orth 1997), or “over-the-shoulder” tools that build digital models in parallel with the physical one (Hsiao and Johnson 2011) illustrate strategies for maintaining parallel models, while augmented reality and projection technology suggests useful strategies for including analysis results within real-world models (Ullmer and Ishii 1997; Hsiao and Johnson 2011), but there is more work to be done.
Even when a designer chooses to work exclusively with physical models, it may be desirable to execute simulation or analysis work on that model and view the results in the physical context (Ullmer and Ishii 1997).
In the future, when an architect is designing a house, their computer might use a camera to watch and interpret their hand gestures to indicate the overall shape, or let them consult “face to face” with a specialist half-way round the world, presenting both with a mutually controllable shared view of the design within which details of concern can be identified and discussed via video chat. As technologies develop, the designer might be able to give a client an immersive virtual-reality model of the design, but how do you represent different levels of commitment within the design? How do you distinguish between the “possible” and the “intended”? Immersive technologies—the hardware and software that enable them—are becoming commercially available, but the logic and cultural changes required to best fit them into both life and practice remains an area of research.
As we attempt to extend the utility of design systems beyond the editing of drawings to manipulation of designs, certain challenges and opportunities appear. Some have to do with representation of non-geometric qualities in designs, such as the distinction between unambiguous aspects of the design (property line locations, topography, site features, utility connections, etc.) and more tentative or propositional content, or emergent reinterpretations of prior work. Visual representation of behaviors, such as parametric constraints, is another realm where progress is needed. Can the unambiguous be visually distinguished from the ambiguous, the certain from the uncertain, or the related from the unrelated? More research and development needs to be done in these areas.
Another challenge has to do with memory, not as a design resource, but as a record. How can a design team remember the rationale for every decision as the design progresses, especially in the face of personnel changes? Now that computers can watch what we do and hear what we say, it seems only natural to include these modalities of interaction in our work with them. Can the rationales behind a design be easily captured and represented in digital form, to become part of the external representation, and searched or computed on in some fashion? Capturing time-stamped graphic and verbal components of design development in a recoverable and (more importantly) searchable form, without the tedium of typing descriptions or transcripts on a keyboard, might aid significantly with CBR, IBIS, and overall knowledge capture and review during a project (Gross et al. 2001; McCall et al. 1994).
Historically, architects have used renderings, models, and words to conjure up their design prior to construction. Now, virtual environment technologies enable a synthetic visual experience to be wrapped around the user, simulating the visual experience of being there. It seems only natural to adopt a technology that will allow others to preview design proposals, but the value goes beyond that. In an experiment conducted by an architecture student at the University of Washington in 1992, a virtual reality (VR) system was used to review models constructed in a traditional 3D modeling interface. The student found that the VR environment often revealed modeling and design errors simply because they were able to experience a wider range of views while using it (Campbell 1992, conversation with the author). As digital models become increasingly important in construction, what they “mean” becomes more important than what they “show.”
Beyond simply presenting design geometry during design review, VR offers the possibility that individuals or groups might construct or edit models from within the virtual space. Perhaps we can even realize the common fantasy and edit the design by pushing walls around and drawing windows with our fingers! It’s an interesting vision, and VR-based BIM will almost certainly appear, but there are a few problematic aspects that need to be sorted out—more design computing opportunities.
Virtual environments usually present the world at the same size as the user, but certain editing, such as shaping the terrain of a large site or laying out a large building, involves actions much larger than the human body. Virtual laser-pointers can be used to extend reach, but accuracy diminishes with distance, and without a “top” view it is hard to judge whether objects align, edges are parallel, etc. One possible solution is a “model-in-model” approach, in which a scaled-down version of the virtual world is located within that world, available to the user to view and edit (HITL 1996).
In first-person video games the game-controller provides relative movements to the protagonist’s avatar or in-game representation—move forward, turn left, jump. To turn all the way around to our left we need only turn left enough times, all-the-while sitting in the same physical position in the world. This is fortunate, as the display screen that shows what our character sees remains solidly in one place as well. In immersive VR, on the other hand, viewpoint motion is usually extracted from the user’s actual movements in the real world. Problems arise when the participant’s virtual universe is larger than their physical one, causing them to bump into walls, or the virtual space includes stairs that don’t exist in the real world. Various strategies to virtualize motion have been applied, ranging from placing participants inside gimbaled “bucky-balls” that rotate to accommodate “virtual movement” in any direction, to placing them in the seats of stationary bicycles or golf-carts whose controls update the in-world viewpoint (Mine 1995).
The technology of VR necessarily involves obscuring the participant’s view of the real world. This means they can see neither their own hands nor the facial expressions of other participants, whether physically co-located or not. While systems allow multiple participants to share a space and see each other in the virtual world, using avatars, the range of facial expression is limited, as is the range of motion that the avatar is able to sense and reproduce from their host user. This makes it very difficult to read a client or colleague’s subtle emotional cues.
Similar to VR technology, augmented reality (AR) interfaces begin with a camera’s live video stream and then digitally superimpose information over the top, as is commonly done with the “line of scrimmage” and “ten-yard line” in television broadcasts of American football. The video signal is monitored for specific colors. If they occur in certain recognized patterns, a secondary, usually synthetic, image stream is superimposed over the top of the live stream. This allows a graphic “marker” printed on a piece of plain paper to be replaced with correctly oriented and positioned real-time renderings of all or part of a design. Other marker-based tools can provide localized transparency or magnification of the model, access to layer visibility controls, overlay of analysis results, etc. (Belcher and Johnson 2008).
While possibly very useful during design presentations, this technology is also appealing to those who need data in the world, and is currently used in certain high-tech fields such as jet engine maintenance, or emergency response. It is also easy to imagine it in use during construction or remodeling projects in complex environments such as hospitals. Such applications are currently hindered by the difficulty of establishing position accurately and flexibly within a building where GPS signals are generally unable to penetrate.
While interface issues are important to efficient and productive design processes, one reason we construct buildings is to support other human activities, including education, business, and housing. All of these activities are increasingly mediated by computing tools, tools which look less and less like free-standing desktop computers and more like elements of the overall environment. Researchers at the famous Xerox Palo Alto Research Center (PARC) were among the first to see this trend and coined the term “ubiquitous computing” in recognition of the shift from conscious (often expensive) computer use to casual background service (Weiser 1991). The continued decline in the price of computing, the increasing ubiquity of network connectivity, and commodification of advanced sensors like cameras, flat-panel displays, and touchscreens offer designers a new palette of tools through which to embed interfaces between people and tasks. With regard to interfaces between people and tasks, or interfaces between people and people, all this tech allows such interfaces to be embedded into the fabric of the building—into the architecture itself.
The challenges of the interface are not limited to human-to-computer interaction. Computer networks allow people to interact, but they often limit those interactions in important ways. Gone are the casual encounters at the elevator, the overheard conversation that reminds you of a shared discussion topic, or the document seen in passing on someone’s desk (Johnson 2001). Working in a group can be invigorating or frustrating, depending on how the members of the group are able to exchange ideas, interact to solve problems, and collaborate on shared tasks. Research in computer mediated communication (CMC) and computer supported collaborative work (CSCW) offers architects both a set of tools with which to practice and an area of research that might benefit from architectural metaphors and insights into human interaction and work.
Nor must the design computing researcher leave the office to find other opportunities to improve on the designer–computer interface. Traditional design presentations, with multiple drawing sheets pinned to a wall and a panel of reviewers, operate in parallel. Each viewer is able to peruse the entire presentation in any order, assembling understanding and questions from the information presented. If available, a 3D model may be examined from all sides—maybe even held in the hands. In contrast, projected digital presentations, even with high-resolution projectors, are usually built on the slideshow model. Serial and controlled from a single point, they come nowhere near their predecessors in terms of the amount of simultaneously available data, nor do they permit self-directed browsing of the information. If there is a 3D model, its presentation is either automated (as an animation) or requires use of the appropriate viewing software, a notoriously difficult proposition. Designers need new and better tools for this kind of interaction, tools that can be shared more easily than the single mouse or keyboard and which are simple and intuitive to use (Fritz et al. 2009).
Modern-day CAD and BIM work involves both human and computer. The efficiency or accuracy of analysis algorithms and the sophistication of representations only matter if the human and the computer can communicate effectively. HCI issues touch on fundamental cognitive issues of both individual designers and groups, as well as designer and client interactions with the design representation. As design data shifts from drawings produced for construction to BIMs produced to guide construction and operation, and as the digital design medium spans ever more of the building lifecycle, the amount of information in digital form and the number of stakeholders that need access to the data increases, while the suitability of existing authoring/editing interfaces to provide it for all users declines. We need richer models of what a BIM can be, and how it can be fluidly accessed and manipulated by these many stakeholders.
At the same time, the opportunity to embed sensors, displays, and actuators into the built environment is changing, and promises to continue to change, the form of buildings and the affordances they offer their occupants—areas of concern to architects.
Dourish, Paul. 2001. Where the action is: The foundations of embodied interaction. Cambridge, MA: MIT Press.
Shneiderman, Ben. 1982. The future of interactive systems and the emergence of direct manipulation. Behaviour & Information Technology 1: 237–256.
Ullmer, B. and H. Ishii. 1997. The metaDESK: Models and prototypes for tangible user interfaces, in Proceedings of UIST ’97, the 10th Annual ACM symposium on user interface software and technology, 223–232. New York, NY: ACM.
Weiser, Mark. 1991. The computer for the 21st century. Scientific American Special Issue on Communications, Computers and Networks, 265 (3): 94–104.
Belcher, Daniel and B. Johnson. 2008. ArchitectureView: An augmented reality interface for viewing 3D building information models, in Proceedings of eCAADe 2008, 561–567.
Csikszentmihalyi, M. 1991. Flow: The psychology of optimal experience. New York: Harper Collins.
Dourish, Paul. 2001. Where the action is: The foundations of embodied interaction. Cambridge, MA: MIT Press.
Engelbart, Douglas and William English. 1969. A research center for augmenting human intellect. AFIPS Fall Joint Computer Conference 33, 395–410.
Fritz, Randolph, Chih-Pin Hsiao, and Brian R. Johnson. 2009. Gizmo & WiiView: Tangible user interfaces enabling architectural presentations, in Proceedings of ACADIA 2009, 278–280.
Gentilviso, Chris. 2010. The 50 worst inventions: Clippy. Time, May 27, 2010.
Gibson, James J. 1986. The theory of affordances, in The ecological approach to visual perception, 127–146. Hillsdale, NJ: Lawrence Erlbaum Associates.
Gorbet, M.G. and M. Orth. 1997. Triangles: Design of a physical/digital construction kit. Proceedings of DIS ’97: The 2nd conference on designing interactive systems—processes, practices, methods, and techniques, 125–128. New York, NY: ACM.
Gross, M., E. Do, and B. Johnson. 2001. The design amanuensis: An instrument for multimodal design capture and playback. Computer aided architectural design futures 2001, 1–13. Dordrecht: Kluwer Academic Publishers.
HITL. 1996. The Greenspace project, Phase 2. Human Interface Technology Lab. www.hitl.washington.edu/projects/greenspace/
Hsiao, Chih-Pin and Brian R. Johnson. 2011. Combined digital & physical modeling with vision-based tangible user interfaces: Opportunities and challenges. Computer aided architectural design futures 2011, 785–799.
Jabi, Wassim. 2016. Linking design and simulation using non-manifold topology. Architectural Science Review. DOI: 10.1080/00038628.2015.1117959.
Johnson, Brian. 2001. Unfocused interaction in distributed workgroups: Establishing group presence in a web-based environment, in Computer aided architectural design futures 2001, 401–414. Dordrecht: Kluwer Academic Publishers.
Kolarevic, Branko. 1994. Lines, relations, drawings and design, in Proceedings of ACADIA-94. Edited by A. Harfmann and M. Fraser, 51–62. ACADIA.
Meyer, Robinson. 2015. Even early focus groups hated Clippy. The Atlantic Online, June 23.
Mine, Mark. 1995. Virtual environment interaction techniques. UNC Chapel Hill CS Dept.
Shneiderman, Ben. 1982. The future of interactive systems and the emergence of direct manipulation. Behaviour & Information Technology 1: 237–256.
Ullmer, B. and H. Ishii. 1997. The metaDESK: Models and prototypes for tangible user interfaces, in Proceedings of UIST ’97, the 10th Annual ACM symposium on user interface software and technology, 223–232. New York, NY: ACM.
Weiser, Mark. 1991. The computer for the 21st century. Scientific American Special Issue on Communications, Computers and Networks, 265 (3): 94–104.