Professional Documents
Culture Documents
Direct
B vs
BEN SHNEIDERMAN
Manipulation
en Shneiderman is a long-time proponent of direct manipulation
for user interfaces. Direct manipulation affords the user control
and predictability in their interfaces. Pattie Maes believes direct
manipulation will have to give way to some form of delegation
namely software agents. Should users give up complete control of
their interaction with interfaces? Will users want to risk depending on
agents that learn their likes and dislikes and act on a users behalf?
Ben and Pattie debated these issues and more at both IUI 97
(Intelligent User Interfaces conference - January 69, 1997) and again
at CHI 97 in Atlanta (March 2227, 1997). Read on and decide for
yourself where the future of interfaces should be headedand why.
n vs Interface
Agents
PAT T I E M A E S
I think we morning when hes been gnawing away at the university has licensed to more than 100 orga-
would do best agent peoplefrom the University of Mary- nizations around the world. QUIS consists of
to focus on the land. On my left, of course, fresh for a fight is 71 items, from low-level details to higher level
remarkable Pattie Maes from the MIT Media Laboratory. concepts in the interface (Office of Technolo-
human Let me just explain. Its 15 minutes from gy Liaison, +1-301-405-4210).
capabilities in each speaker to place their position, and then We think accommodating individual dif-
the visual 5 minutes allowed each for rebuttal. Then we ferences is importantnot just mentioning
domain, which I open it up to the floor for about 30 minutes experts and novices, but understanding quan-
think are largely or so. Could you please use the microphones? titatively what performance differences we
underutilized If you want to line up behind the micro- anticipate. Do we expect experts to perform
by the current phones, I will select you to make your com- twice as fast or twenty times as fast as novices?
designs with ments. Then at the end, there will be two Would men perform differently from women
40 icons in 2-3 5-minute summing ups. So, let us commence in the use of interfaces, or prefer different
windows. I the debate. interfaces? And then we try to deal with
think we should broader cultural issues that are even more dif-
have two or Ben: First, my thanks to the organizers of this ficult to measure.
three orders of Intelligent User Interfaces workshop for dar- For me, the future is most clearly moving
magnitude ing to arrange this debate. It was Angel Puer- in the direction of information visualiza-
more: 4,000 or tas careful attention to my concerns that tion. I think we would do best to focus on
more items on made me feel comfortable in coming to speak the remarkable human capabilities in the visu-
the screen in an here, and so I want to offer him a souvenir al domain, which I think are largely underuti-
orderly way that from our labthe usual T-shirt. And to Jim lized by the current designs with 40 icons in
enables people Alty, a cup of tea for when he needs to relax, 2-3 windows. I think we should have two or
to see all of the from our group at the lab. three orders of magnitude more: 4,000 or
possibilities and I am pleased to represent the University of more items on the screen in an orderly way
navigate among Marylands Human-Computer Interaction that enables people to see all of the possibili-
them. Lab, which is now 14 years young. Over the ties and navigate among them.
years weve explored a host of user-interface I will show you three brief videotaped
design issues in an interdisciplinary way examples. They all show applications and
involving computer science, psychology, and extensions of the strategy of direct manipula-
the library school. Our goal is to create envi- tion, a term I coined in 1982 to describe the
ronments where users comprehend the dis- existing successful systems. These systems all
play, where they feel in control, where the had rapid, incremental, and reversible actions,
system is predictable, and where they are will- selection by pointing, and immediate feedback
ing to take responsibility for their actions. To (100-millisecond updates for all actions). I
direct me, responsibility will be the central issues in believe that this strategy reduces errors and
vs
manipulation this debate. encourages exploration. The current manifes-
My pitch over 20 years has been to make a tations of direct manipulation are the visual
science out of user interface research. I want to ways of searching in databases and on the Web
get past the notion of user-friendly, a vague, accompanied by visual presentation of results.
and misleading term, and to be really clear Lets take a look at an example that goes
interface about specifying who the users are and what back 4 years in the first videotape called the
agents
their tasks are. Then we can make a theory FilmFinder.
Users have
great control
over the display
and as they
select items, the
details appear
in windows on
the sides.
FilmFinder video (see Figures 1a-c, CHI94 long, and then we can use the category button
videotape or HCIL 1995 Video tape reports): to show only drama or only action films. We
This display shows thousands of films can zoom in on more recent pictures and take
arranged on the x-axis from 1920 to 1993 only the more popular ones. And when there
and on the y-axis from low to high populari- are fewer than 25, the titles will appear auto-
ty. We can use the length slider to trim the set matically. When we select one of them, we get
of films by the number of minutes in the film a description of the film and information
so we do not have to see films that are too about the actress and actor. We can also hunt
for films organized by actors. In this case, you It would be hard to see how to program
might be interested in Sean Connery, and his any kind of agent tool to anticipate all of the
films will appear on the screen. possibilities that your eye would pick up. We
show the age of the youthful offenders on the
Ben: Okay. I think you get an idea that the x-axis. There are 4,700 of them, from 10 to
controls are visually apparent as you drag 19 years old. The number of days until a deci-
them. The updates occur in 100 milliseconds sion was made on their treatment plan is
and users get a clear understanding of what shown on the y-axis. The rules of this organi-
the situation is. This work goes back to 1993, zation say that decisions must be made with-
and the 1994 CHI conference has a couple of in 25 days, but you can see a lot more than
papers describing it [1, 2]. A general purpose they anticipated are well above the 25-day
version of that software was used for the limit.
direct Department of Juvenile Justice project, which Interesting things pop up whenever you try
vs
manipulation you will hear about shortly. a visualization. I hope you will spot these yel-
Here is a FilmFinder done in the UNIX low linesthose were a surprise. We thought
version of the product called SpotFire (Figure there was a bug in the program, but it turns
2). Chris Ahlberg has made a commercial out that if you start clicking on them to get
product out of this and you can download the details-on-demand, youll find out they all
interface the demo version off of the Web (http:// occur in a Hartford County. They were all
agents
www.ivee.com). brought in on a certain day. They all have the
Figure 3: Youth Record prototype using the Lifelines display to show a case history for the Maryland Department of Juvenile
Justice. (CHI96 videotape or HCIL 1996 Video Tape Reports; 4) ftp://www.cs.umd.edu/projects/hcil/Research/1997/patientrecord.html
Figure 4: Visible Human Explorer user interface, showing a reconstructed coronal section overview
(on the left) and an axial preview image of the upper abdominal region (on the upper right). Drag-
ging the sliders animates the cross-sections through the body (North et al., 1996).
ftp://www.cs.umd.edu/projects/hcil/Research/1995/visible-human.html
Pattie: Im not going to bribe the moderator tive because it knows what your interests are. It
with tea or T-shirts or anything. I hope that can, for example, tell you about something
the work will speak for itself. The word agent that you may want to know about based on the
is used in a lot of different ways. I want to start fact that you have particular interests. Current
this presentation by explaining what I mean by software, again, is not at all proactive. It does-
the word agent, and in a particular, software nt take any initiative. All of the initiative has to
agent. Basically, software agents are a new come from the user. A third difference with
approach to user software, a new way of think- current software is that software agents are direct
vs
ing about software that end-users have to use. more long-lived. They keep running, and they manipulation
In particular, the way in which agents differ can run autonomously while the user goes
from the software that we use today is that a about and does other things. Finally, software
software agent is personalized. agents are adaptive in that they track the users
A software agent knows the individual users interests as they change over time.
habits, preferences, and interests. Second, a So, you can ask, well, why do you call it an interface
agents
software agent is proactive. It can take initia- agent? Why call it an agent given that the term
vs
manipulation ment that we have complete control over. with because of our limited attention span or
Instead, our computer or the screen is a win- limited time, and we need other entities to be
dow onto this vast network, this vast network able to represent us and act on our behalf.
of information and other people. That net- Some examples to make this more concrete (I
work is continuously changing. It is dynamic. didnt bring any videos because of the limited
interface Something may be in one place today, and the amount of time), but most of you have seen at
agents
next day it may be in another place or may be least one of these agents. These are some of
the ones we built in our lab. Letizia, built by sent e-mail or whether you already replied to
Henry Lieberman, who is here at the confer- a certain e-mail message. It may proactively
ence, is an agent which continuously watches remind you of information related to the
you as you browse the Web, analyzing and information you are currently looking at. It We need to be
memorizing all of the pages that you visit. It works in EMACS. When I am, for example, able to delegate
extracts from those pages the common key- looking at an e-mail message from a particular to what could
words. person, it proactively reminds me of the previ- be thought of
Whenever you are using your Web brows- ous e-mail messages from that same person, as extra eyes
er, Letizia always looks ahead and checks which is very useful because I may have for- or extra ears
whether within a certain depth of the current gotten to reply to one of them. that are on the
page, there happen to be any pages that you Firefly, some of you may have tried that lookout for
may be interested in. So, for example, if I am agent, is basically a personal filterer for enter- things that
interested in scuba diving, my agent may have tainment, not unlike the movie application you may be
picked it up because I look at a lot of pages that Ben talked about, except that this agent interested in.
about scuba diving. If I go to a particular will again keep track of your interests, your
entertainment site, it may look ahead and say, preferences, and proactively tell you about
hey did you realize that if you follow that link new movies that you may be interested in which
that there are some pages about scuba diving you even forgot to ask about in the first place.
in the Florida area? The Remembrance Agent is Yenta is another agent that we built which
another agent that continuously tracks the tracks what the users interests are by looking
behavior of the user. It helps you remember at your e-mail and files and extracting key-
certain things. It helps you remember who words. It talks to other Yenta agents belonging
vs
manipulation those interests are very rare, then it introduces CD from you. Again, it is sort of acting on
you to that other user. It may say hey did you your behalf. You dont have to waste any time
realize that at this IUI conference there is trying to make 10 bucks, but the agent will do
another person who is interested in going this for you.
scuba diving in Florida so that maybe then I think its important to address some com-
interface we can decide to go scuba diving together. mon misconceptions about agents: First of all,
agents Again, its suggesting something that you sorry to say so, but agents are not an alterna-
extent its true. If I dont fix my car then Im not so I was delighted with her opening remarks
Opponents of going to learn about fixing cars. However, this that rejected intelligent and anthropomorphic
agents typically does not constitute a problem. As long as theres designs. The old Pattie Maes wrote agents
argue that well- always an agent available or I can call one by a will appear as living entities on the screen,
designed visual- motor association like AAA, then thats fine. Its conveying their current state of behavior with
ization too bad that I will never learn about cars, but I animated facial expression or body language
interfaces are want to learn about other things instead. rather than windows text, graphics, and fig-
better. Like I A third criticism expressed is that using ures. So weve got two Pattie Maes. I will
said before, you agents implies giving up all control. Thats incor- choose the newer one that demonstrates
still need a well- rect. I think you do give up some control when movement in my direction including her last
designed inter- you deal with an agent. I tell the car mechanic to slide which might have been written by me:
face when fix my car or to fix this or that part of the car. I User understanding is central, and user con-
incorporating dont know how exactly he or she is going to do trol is vital for people to be successful.
agents in an that. I dont mind giving up some control, actu- In fact, I have other ways of celebrating
application. ally, and giving up control over the details as long Pattie Maes. I encourage you to look at her
However, some as the job is done in a more-or-less satisfactory Firefly Web site, which is an interesting appli-
tasks I may just way, and it saves me a lot of time. cation. Collaborative filtering, I think, will
not want to do Okay, just very briefly, I want to say that I become an important approach for many
myself even if think where the true challenge lies is in domains. But as a user, I cant find the agents
the interface designing the right user-agent interface. In on the Firefly Web site. In fact, as I searched
was perfect. If particular, we need to take care of these two to find the agents, all I came up with was that
my car had a issues: understanding and control. Under- the company had previously been called
perfect interface standing means that the agent-user collabora- Agents, Inc. and is now called Firefly. If you
for fixing the tion can only be successful if the user can read the Firefly Web site, you will not find the
engine, I still understand and trust the agent, and control word agents in the description of this sys-
wouldnt fix it. means that users must be able to turn over tem. In fact, the interface is a quite lovely,
I just dont want control of tasks to agents but users must never direct manipulation environment, allowing
to bother with feel out of control. I believe that this is a won- users to make choices by clicking in a very
fixing cars. I derful interface-design challenge, and we have direct manipulation way.
want someone come up with a lot of solutions to actually So, I think weve made progress in clarify-
else to do it. make sure that the agents user interface has ing the issues in the past year of our ongoing
these two properties that the user feels in con- discussions. For example I think we can sepa-
trol or has control when he or she wants it, as rate out the issue of natural language interac-
well as that the user understands what the tion, which as far as I can see, has not been a
agent does and what its limitations are. Let me success. The systems that were offered com-
save that for later, maybe. Thanks. mercially even a few years ago, like Q&A from
Symantec or Intellect from AI Corporation to
Jim: Thanks very much. Okay, Bens going to do database query, and Lotus HAL for spread-
go up to 5 minutes to say whatever he likes. sheets, are gone, and direct manipulation is
the surviving technology.
direct Ben: How interesting. We are debating, but A second issue is anthropomorphic inter-
vs
manipulation part of me is drawn to the idea of celebrating faces such as chatty bank tellers and the Postal
Pattie Maes and encouraging you to follow Buddy or the proposed Knowledge Navigator
her example. I want to draw the audiences of Apples 1988 video. Microsofts playful
attention to her transformation during the attempt at a social interface in BOB is also a
months weve had these discussions. As I go failed product. As far as I can see the anthro-
interface back to Pattie Maess work and I read her ear- pomorphic or social interface is not to be the
agents lier papers and her Web sites, she promotes future of computing.
I am concerned it deceives the users; it increases anxiety about about other users that you may want to talk
about the computer usage, interferes with predictability, to. It will warn you when there is something
confusion of reduces user control, and undermines users that has changed somewhere that you may be
human and responsibilitywhich I think is central. I interested in. There is still an agent there
machine think anthropomorphic representations monitoring all of your preferences and proac-
capabilities. I destroy the users sense of accomplishment; I tively making recommendations to you, but
make the basic think users want to have the feeling they did that doesnt mean that there has to be this lit-
assertion that the jobnot some magical agent. tle cute character on the screen.
people are not Finally I am concerned about the confu- Now, I think one of the reasons that Ben
machines and sion of human and machine capabilities. I and I disagree is actually that we are focusing
machines are make the basic assertion that people are not on completely different problem domains. In
not people. I do machines and machines are not people. I do pretty much all of the problem domains that
not think that not think that human-to-human interaction is Ben looks at we are dealing with a user who is
human-to-human a good model for the design of user interfaces. a professional user, and we are dealing with a
interaction is a task domain that is very well structured and an
good model for Jim: Okay, Pattie, would you like to respond information domain that is very well orga-
the design of to that? nized, so that it lends itself to visualizing all of
user interfaces. the different dimensions. The kind of prob-
Pattie: First of all, I should clarify that lems that we have typically been dealing with
autonomous agents or the word agents has a are very different because we are dealing with
much broader meaning than the words soft- end users who are not necessarily very trained.
ware agent, and my group at the Media Lab They may use the Web for a couple of hours
does research on autonomous agents more per week, but that is about it. We are dealing
generally, as well as software agents. So when with a very different information domain, an
Ben was quoting from our Web site, hes actu- information domain that may be very ill struc-
ally quoting other work that we do, for exam- tured and very dynamic. For example, the
ple, work on synthetic characters that can World Wide Web is actually sort of one of the
interact with people in a virtual environment, key domains that we do all of our research on.
which doesnt have anything to do with the Finally, to illustrate that these approaches
software agents work. In fact, it has less and arent necessarily incompatible, I could envi-
less to do with it than it may have at one sion a version of Bens movie finder which uses
point. So, if you go to autonomous agent con- Bens nice visualization interface, where an
ferences, for example, the Agents Conference agent is continuously monitoring what movies
direct in Marina del Rey in February (First Interna- you seem to be interested in. That agent may,
vs
manipulation tional Conference on Autonomous Agents, for example, highlight particular areas in the
see interactions iii.6, Conference Preview), interface which it thinks you will be specifi-
youll see work being discussed that relates to cally interested in. That kind of interface
robots, autonomous robots. Youll see work would actually combine an agent that learns
about synthetic believable characters, and about your preferences and proactively makes
interface youll see work about software agents. So thats suggestions to you, with a nice visualization
agents one thing I wanted to respond to. Its impor- interface. The reason why you want that kind
User interfaces Trek scenario is going to take over and we will taught an agent a particular procedure, and
should be talk to our computers. I do not believe that while he was performing actions with the
predictable, so speech will be a generally usable tool. It has mouse, he would give speech inputs to tell the
that users trust important niches: opportunities for disabled agent what it had to pay attention to. For
them. User users, for certain hands-busy, eyes-busy, and example, pay attention to this corner here that
interfaces mobility-required applications. In preparing I am dragging or to this side of the rectangle.
should be thor- the third edition of my book I worked hard to So in that situation it is very useful because
oughly tested, find speech applications that do recognition your hands are already doing something else,
and users should effectively. I am quite happy with speech and you need that additional channel to con-
be thoroughly store-and-forward applications by telephone, vey some more information in parallel.
trained for all but the recognition paradigm is not being
emergencies. widely accepted, even for minor tasks such as
In emergency voice dialing. Speech output, except by tele- Question: This question is for both of you. Both
situations, phone, is also a problem because speech is of you seem to be concerned about protecting the
people cannot very slow and disruptive of cognitive process- users control of the environment, but the one
solve problems. ing. I think what annoys me the most about things studies have shown time and time again is
They can only the devotees of speech, is their failure to take that users are very good at making mistakes. So
do whats in the scientific evidence that speaking com- how do your positions relate to time-critical deci-
rehearsed and mands is cognitively more demanding than sion-support environments, such as medical sys-
predictable. pointing. Speech uses your short-term memo- tems or cockpit systems?
ry and working memory. By contrast, hand-
eye coordination can be conducted in parallel Pattie: I have actually been focusing on a
with problem solving by another part of your completely different kind of application, a
brain and therefore does not degrade your per- type of application that is not as critical. For
formance as much as speaking. example, if your World Wide Web agent gives
you a wrong Web page to look atit assumes
that you are interested in a Web page and you
Question: You can do both at the same time. are notthat is not at all critical. It is not a
big deal. I have been focusing on that kind of
Ben: Yes, you can do hand-eye tasks in paral- situation and those kinds of problems, the
lel with problem solving, more easily that you ones where if there is an error it is not very
can speak while problem solving. This fact is costly. Ive been doing that because I believe
not a barrier to use of speech, but it is a hur- that it will be very hard to make agents that
dle that designers of speech systems must rec- always come up with the right answer, always
ognize if they are to find ways to overcome it. do the right thing. I believe that there is a very
Pattie: I must admit, I actually agree with a lot large set of these kind of applications where
of what Ben says. I havent used speech at all things dont have to be completely precise or
100 percent correct in order for the agent to talking about the wonderful visual processing
be very, very useful to a user. and how important visualization isthis seems
to be assuming that everybody has perfect vision.
Ben: I like your question. I think it is an What if I am blind, if I am over sixty-five, and I have been
extremely important research area. There is a I have a very small useful field of view and I focusing on that
long history of work often called supervisory dont notice things so much on the periphery, or if kind of situation
control. Tom Sheridan is a key player in the I am, in my case, a person sitting right here who and those kinds
area for nuclear-reactor control rooms, cock- is having a problem with my contact lenses, how of problems, the
pits, and so on. I think the design of these sys- could you, Ben you were sayingwell speech is ones where if
tems is most effective when the users have a okay for disabled usershow are you going to there is an error
clear predictive model of what their actions render that diagram into speech for a blind per- it is not very
will produce. If they do not know or are son, and what is the role of that? costly. Ive been
uncertain about what the results of their doing that
actions are, they will disengage the automatic Ben: That is a legitimate concern. Direct because I
system, as is the evidence with cockpit systems manipulation does benefit from and depends believe that it
or nuclear-control rooms. So one danger is heavily on visual representations. For those will be very
that in complex control-room environments who are vision challenged or blind, alterna- hard to make
when an emergency occurs, users are uncer- tives to visual displays are important. What agents that
tain about its behavior. In these situations surprises me is there are great supporters of always come up
they are likely to disengage a potentially help- direct manipulation in the visually-challenged with the right
ful system and do what makes sense to them. community, because direct manipulation answer, always
Therefore, as Jim Foley said, keep it simple, depends on spatial relationships. Blind users do the right
very simple. often are strong at spatial processing. If you thing. I believe
User interfaces should be predictable, so that can provide movement left, down, backward, that there is a
users trust them. User interfaces should be forward, they can navigate fairly rich spaces in very large set of
thoroughly tested, and users should be thor- efficient ways. these kind of
oughly trained for all emergencies. In emer- I would say also you have been a little too applications
gency situations, people cannot solve problems. quick about criticizing menu selection. The where things
They can only do whats rehearsed and pre- question is what would the alternative be, and dont have to be
dictable. It is a good topic that I would love to how might those menus be better designed? I completely
see more attention to it by this community. do believe that fast and vast menus are a great precise or 100
benefit in many applications. percent correct
in order for the
Question: In the interest of brevity, I was going Jim: Okay. There are two more questions if agent to be
to bring up several human limitations or con- you could be very brief, please. very, very useful
straints on humans that make direct manipula- to a user.
tion a little more interesting and wanted to ask
you both for comments, not that I have all of the Question: This is directed more toward Ben
answers, but I think I would just limit it to your than toward Pattie. Will this debate between
basic kind of law that the more things you have direct manipulation and agency always exist in
to scroll through or the more things you have to interface design or will it eventually be replaced
search through, the longer it takes you to search. by some kind of fusion of the two approaches? In
The idea of how do you deal with this, particu- other words, are we going to see new Bens and
larly, for many people scrolling is not a particu- new Patties every day, or is there going to be some direct
vs
larly usable thing. I have particularly seen this kind of Shneider-Maes? manipulation
with older people, and of course, if you want a
good example just go to a fast food restaurant and Ben: I think it has been interesting to see how
watch a new person at the cash register try to find the debates evolve. I certainly will point you
how to ring up your hamburger. It is pretty ter- to the new Pattie, whom I am ready to cele-
rible. You will be there for days. brate and be partners with, as I said, but I interface
That kind of idea, and particularlywe are think the debate will move on. I think it has agents
vs
manipulation dislike? want to do is change these parameters. The
designers say, Well, you know, we didnt put
Ben: Can I go to my closing slide? I want to that in the interface. They just havent
reassert the importance of scientific evalua- thought adequately about the interface, nor
tions. We must get past the argumentation done testing early enough.
interface about my system being more friendly than I believe that this language of intelligent,
agents
yours or more natural or intuitive, and talk autonomous agents undermines human
responsibility. I can show you numerous arti- field definitely has grown a lot in the past 10
cles in the popular press which suggest the years or so. In fact, one of the ways I think in
computer is the active and responsible party. which a lot of this agent work distinguishes
We need to clarify that either programmers or itself from traditional AI is that agent research
operators are the cause of computer failures. focuses on building complete systems, systems
Agent promoters might shift some attention that are tested, systems that really have to
to showing users what is happening so that work, and those same principles and method-
they can monitor and supervise the perfor- ologies can be seen in all of the agents work,
mance of agents. I was disturbed that in the whether it be robots, synthetic characters, or
Autonomous Agents conference that Pattie is software agents. I believe that
participating in, the organizers refused to The field is maturing and paying more there are real
include the topics of supervision of agents and attention to building things that really work limits to what
user interfaces for programming agents. By and paying attention to important UI issues. we can do with
contrast, I like Patties summary slideI think As to people taking responsibility for their visualization and
her list is quite wonderful. agents, I think they indeed should. It is soft- direct manipula-
My closing comment is that I think there ware that is running on your behalf and that tion because our
are exciting opportunities in these visual inter- you have delegated certain tasks to. So, per- computer envi-
faces that give users greater control and there- sonally, I dont see why that problem is specif- ronments are
fore greater responsibility in the operation of ic to agents as opposed to software in general. becoming more
computers. Thanks. and more com-
plex. We cannot
Jim: Thanks. Okay, Pattie? Jim: Okay, I would like to thank the Shnei- just add more
dermanMaes team for coming here today and more sliders
Pattie: I want to conclude by saying that I and talking to us, and thank you all for par- and buttons.
believe that there are real limits to what we ticipating. Also, there are
can do with visualization and direct manipu- limitations
lation because our computer environments are References because the
becoming more and more complex. We can- [1] Ahlberg, C. and Shneiderman, B., Visual Informa- users are not
not just add more and more sliders and but- tion Seeking: Tight coupling of dynamic query filters computer-
tons. Also, there are limitations because the with starfield displays, Proceedings Of ACM CHI94 trained. So, I
users are not computer-trained. So, I believe Conference (April 1994), 313-317 + color plates. believe that we
that we will have to, to some extent, delegate [2] Ahlberg, C. and Shneiderman, B., AlphaSlider: A will have to, to
certain tasks or certain parts of tasks to agents compact and rapid selector, Proceedings of ACM CHI94 some extent,
that can act on our behalf or that can at least Conference, (April 1994), 365-371. delegate certain
make suggestions to us. [3] North, C., Shneiderman, B., and Plaisant, C., User tasks or certain
However, this is completely a complemen- controlled overviews of an image library: A case study parts of tasks to
tary technique to well-designed interfaces of the Visible Human, Proceedings 1st ACM Internation- agents that can
visualization, and direct manipulationnot a al Conference on Digital Libraries, (March 1996), 74-82. act on our
replacement. Users still need to be able to [4] Plaisant, C., Rose, A., Milash, B., Widoff, S., and behalf or that
bypass the agent if they want to do that. Also Shneiderman, B., LifeLines: Visualizing personal histo- can at least
I should say that what we have learned the ries, Proceedings of ACM CHI96 Conference (April make sugges-
hard way really is that we have to, when 1996), 221-227, 518. tions to us.
designing an agent, pay attention to user-
interface issues, such as understanding and
control. These are really very, very important PERMISSION TO COPY WITHOUT FEE, ALL OR PART OF THIS MATERIAL IS GRANT-
if you want to build agents that really work ED PROVIDED THAT THE COPIES ARE NOT MADE OR DISTRIBUTED FOR DIRECT
and that users can trust. The user has to be COMMERCIAL ADVANTAGE, THE ACM COPYRIGHT NOTICE AND THE TITLE OF
able to understand what the agent does. THE PUBLICATION AND ITS DATE APPEAR, AND NOTICE IS GIVEN THAT COPYING
The user has to be able to control things if IS BY PERMISSION OF THE ASSOCIATION FOR COMPUTING MACHINERY. TO COPY
they desire or to the extent that they want to OTHERWISE, OR PUBLISH, REQUIRES A FEE AND/OR SPECIFIC PERMISSION.
control things. I agree with Ben that the agent ACM 1072-5520/97/1100 $3.50