]> git.zerfleddert.de Git - micropolis/blame - src/notes/Beyond-Intelligent-Machines
fix fire coverage overlay by iterating over the whole map
[micropolis] / src / notes / Beyond-Intelligent-Machines
CommitLineData
6a5fa4e0
MG
1Date: Tue, 24 Nov 92 22:25:59 -0500
2From: ben@cs.UMD.EDU (Ben Shneiderman)
3To: bam@cs.cmu.edu, ben@cs.umd.edu, callahan@cerc.wvu.wvnet.edu,
4 hopkins@bongo.garnet.cs.cmu.edu, weiser.pa@xerox.com
5Subject: Re: more pie menus!
6
7I couldn't resist sending you all this latest essay which is
8destined for IEEE Software...some readers expect it to generate
9some strong responses...Ben
10
11
12Beyond Intelligent Machines:
13 Designing Predictable and Controllable User Interfaces
14
15
16 Ben Shneiderman November 24, 1992
17
18 University of Maryland, College Park, MD 20742
19
20 Professor, Department of Computer Science,
21 Head, Human-Computer Interaction Laboratory at the
22 Center for Automation Research &
23 Member, Institute for Systems Research
24
25
26Who's in control?
27
28An important shift is occurring from the old vision of computers
29as 'intelligent' to a new vision based on predictable and controllable
30user interfaces that depend on direct manipulation of objects and actions.
31Appropriate metaphors and terminology are important since they shape
32the thoughts of researchers, designers, managers, congress-people,
33journalists, etc. Most of us have learned the importance of gender
34neutral terminology and similarly I have been strongly opposed to
35suggesting that computers are 'intelligent' or 'smart' for several
36reasons:
37
381) Limits to Imagination
39
40I think we should have much greater ambition than to make a computer
41behave like an intelligent butler or other human agent. Computer
42supported cooperative work (CSCW), hypertext/hypermedia, multi-media,
43information visualization, and virtual realities are powerful
44technologies that enable human users to accomplish tasks that no human
45has ever done. If we describe computers in human terms then we
46run the risk of limiting our ambition and creativity in the design
47of future computer capabilities.
48
49
502) Predictability and Control are Desirable
51
52If machines are 'intelligent' or 'adaptive' then they may become less
53predictable and controllable. Our usability studies show that users
54want feelings of mastery,
55competence, and understanding that come from a predictable and
56controllable interface. Most users seek a sense of
57accomplishment at the end of the day, not the sense that this
58'intelligent' machine magically did their job for them.
59
60
613) Human Responsibility
62
63I am concerned that if designers are successful in convincing the users
64that computers are intelligent, then the users will have a reduced sense
65of responsibility for failures. The tendency to blame the machine is
66already widespread and I think we will be on dangerous grounds if we
67encourage this trend.
68
69
704) Machines are not People AND People are not Machines
71
72I have a basic philosophical objection to the suggestion that machines
73are, or can ever be, intelligent. I know that many of my colleagues are
74quite happy to call machines intelligent and knowledgeable, but I prefer
75to treat and think about machines in very different ways from the way I
76treat and think about people.
77
78
79The lessons of history
80
81While some productive work has been done under the banner of
82`intelligent', often those who use this term reveal how little they
83know about what users want or need. The users's goal is not to
84interact with an 'intelligent' machine, but to create, communicate,
85explore, plan, draw, compose, design, or learn. Ample evidence
86exists of the misguided directions brought by 'intelligent' machines:
87
88 - natural language interaction seems clumsy and slow compared to
89direct manipulation and information visualization methods that use
90rapid, high-resolution, color displays with pointing devices. Lotus HAL
91is gone, AI INTELLECT hangs on but is not catching on. There are some
92interesting directions for tools that support human work through
93natural language processing: aiding human translators, parsing
94texts, and generating reports from structured databases.
95
96 - speech I/O in talking cars and vending machines is gone.
97Voice recognition is fine for handicapped users plus special situations,
98but doesn't seem to be viable in general office, home, or school
99settings. Our recent studies suggest that speech I/O has a greater
100interference with short term and working memory than hand-eye
101coordination for mouse menu selection. Voice store and forward,
102phone-based information retrieval, and voice annotation have great
103potential but these are not the 'intelligent' applications.
104
105 - adaptive interfaces are unstable and unpredictable, often leading
106users to worry about what will change next. I see only modest chances
107for success in user modeling to recognize the level of expertise and
108revise the interface accordingly - can anyone point to successful
109studies or commercial products? By contrast, user controlled
110adaptation through control panels, cruise control for cars, and
111remote controls for TV are success stories. While algorithms to
112deal with dynamic
113issues in network or disk space management are needed, the task domain
114and user interface issues of the application program
115should generally be under direct user control.
116
117 - Intelligent CAI (Computer Assisted Instruction) only prolonged the
118time (compared to traditional CAI) until the users felt they were the
119victims of the machine. Newer variations such as Intelligent Tutoring
120Systems are giving way to Interactive Learning Environments where
121students are in control and actively creating or exploring.
122
123 - intelligent talking robots with five-fingered hands and human facial
124features (quaint fantasy that did well in Hollywood but not in Detroit
125or elsewhere) are mostly gone in favor of flexible manufacturing systems
126that enable supervisors to specify behavior with predictable results.
127
128
129It seems that some designers continue to ignore this historical pattern
130and still dream of creating 'intelligent' or 'smart' machines. It is an
131ancient and primitive fantasy, and its seems most new technologies must
132pass through this child-like animistic phase. Lewis Mumford identified
133this pattern (Technics and Civilization, 1934) when he wrote about the
134Obstacle of Animism: 'the most ineffective kind of machine is the
135realistic mechanical imitation of a man or another animal...for
136thousands of years animism has stood in the way of...development.'
137
138
139An alternate vision
140
141My point in this essay is not merely to counter a popular design
142philosophy, but to offer a new vision that is more in harmony with what
143users want. I believe that the future will be filled with powerful, but
144predictable and controllable computers that genuinely serve human needs
145(Designing the User Interface: Strategies for Effective Human-Computer
146Interaction, Second Edition, Addison-Wesley Publ. Co., Reading, MA, 1992).
147
148In this vision of predictable and controllable (PC) computing,
149the promising strategies are rapid,
150visual, animated, colorful, high resolution interfaces built on
151meaningful control panels, appropriate preference boxes,
152user-selectable toolbars, rapid menu selection, easy to create macros,
153and comprehensible shortcuts. These enable me to specify rapidly,
154accurately, and confidently how I want my email filtered, what documents
155I want retrieved and in what order, and how my documents will be
156formatted.
157
158
159Our Human-Computer Interaction Laboratory has applied these principles
160to information visualization methods that give users X-ray vision to see
161through their mountains of data. Treemaps enable users to see (and
162hear) 2-3000 nodes of hierarchically structured information by utilizing
163every pixel on the display. Each node is represented by a rectangle
164whose location preserves the logical tree structure and whose area is
165proportional to one of its attributes. Color represents a second
166attribute and sound a third (B. Johnson & D. Turo, Improving the
167Visualization of Hierarchies with Treemaps: Design Issues and
168Experimentation, Proc. IEEE Visualization '92). Treemaps have been
169applied to Macintosh directory browsing (Figure 1), in which area could
170be set to file size, color to application type, and sound to file age
171(our TreeViz application is available from the University of Maryland's
172Office of Technology Liaison, (301) 405-4210). When users first try
173TreeViz they usually discover duplicate or misplaced files, redundant
174and chaotic directories, and many useless files or applications. Other
175applications include: stock market portfolio management, sales data,
176voting patterns, sports (48 statistics on 459 NBA players, in 27 teams,
177in four leagues), etc.
178
179
180Dynamic queries allow rapid adjustment of query parameters and immediate
181display of updated result sets. These animations enable users to
182develop intuitions, discover patterns, spot trends, find exceptions, and
183see anomalies. The Dynamic HomeFinder prototype (Figure 2) allows users
184to adjust the cost, number of bedrooms, location, etc. and see points of
185light come and go on a map to indicate a matching home. Users execute
186up to 100 queries/second (rather than one query per 100 seconds)
187producing a revealing animated view of where high or low priced homes
188are found, and there are no syntax errors. Clicking on a point of
189light brings up a description or image (videotape available, or for
190an empirical comparison with a natural language system, see
191Williamson, C. and Shneiderman, B., The Dynamic HomeFinder: Evaluating
192dynamic queries in a real-estate information exploration system, 1992
193ACM SIGIR Proceedings).
194
195Dynamic queries are very effective when a visual environment such as a
196map, calendar, or schematic diagram are available, but they can be
197easily applied with standard text file output (Figure 3). Dynamic
198queries exemplify the future of interaction; You don't need to
199describe your goals, negotiate with an intelligent agent, and wait for
200a response, you Just Do It! Furthermore, dynamically seeing the
201results enables you to explore and rapidly reformulate your goals in
202an engaging videogame-like manner.
203
204
205Open problems in information visualization include screen organization,
206widget design, algorithms for rapid search and display, use of color
207and sound, and strategies to accommodate human perceptual skills.
208We also see promise in expanding macro makers into the graphical
209environment with visual triggers based on controlled replay of
210desired actions - the
211general idea is Programming in the User Interface (PITUI) to
212Do-What-I-Did (DWID).
213
214
215I want to encourage the exploration of new metaphors and visions of how
216computers can empower people by presenting information, allowing rapid
217selection, supporting personally specified automation, and providing
218relevant feedback. Metaphors related to controlling tools or machines
219such as driving, steering, flying, directing, conducting, piloting,
220or operating seem more generative of effective and acceptable
221interfaces, than 'intelligent' machines.
222
223
224A scientific approach to user interface research
225
226Whether you agree with the design philosophy in this essay, and
227especially if you disagree, I hope that you will add to our scientific
228knowledge by conducting well-designed empirical studies of learning
229time, measuring performance time for appropriate tasks, recording error
230rates, evaluating human retention of interface features, and assessing
231subjective satisfaction. There's much work to be done to make
232computing accessible, effective, and enjoyable.
233
234
235Acknowledgements: This essay was prompted by the discussion between
236Mark Weiser and Bill Hefley, stimulated by lively email and personal
237discussions with Paul Resnick, Tom Malone, and Christopher Fry at MIT,
238and refined by comments from Catherine Plaisant, Rick Chimera, Brian
239Johnson, David Turo, Richard Huddleston, and Richard Potter at the
240Human-Computer Interaction Lab at Univ. of Maryland. I appreciate Bill
241Curtis's support for this vision. Thanks to all.
Impressum, Datenschutz