In today’s world, users want to be notifiedabout multiple sources of information whileengaged in other tasks. Notification systems areinterfaces specifically designed to support useraccess to additional digital information fromsources secondary to current activities. Manysuch interfaces, especially examples such as Webpage advertisements and animated softwareagents, seem to be ineffective and distracting,

and are abandoned or ignored after brief use. We believe dissatisfaction results fromincorrect estimates of the user’s task prioritization during design time. Conse-quently, information is introduced at inappropriate times and with unsuitable pre-sentation choices. Factors such as the nature of ongoing activities, perceived urgency,

COMMUNICATIONS OF THE ACM March 2003/Vol. 46, No. 3 67

and attentional focus require changes in the wayinformation should be delivered. By tracking pri-orities of user attention and inferring workloadcharacteristics through eye gaze, physical or bio-medical sensors, and input devices, attentive userinterfaces (AUI) [8] and more specifically, atten-tion-centric systems [1], can adapt informationdelivery to avoid overloading the user.This interface adaptivity suggests a keyparadigm with enormous potential fornotification systems.To best leverage the AUI paradigm for

notification design, we explore how wemight understand the associated costs and benefitsof user notification in terms of its impact on userattention. We introduce a framework that allowsthese costs and benefits to be described and designoptions to be compared. Based on this, we showhow user goal representations can be integratedwith information design guidelines from usability

studies. This demonstrates vast potential for AUIsin notifying users—compelling attentive notifica-tion systems. We also suggest some challenges forthis emerging research community.

The paramount challenge of notification is pre-venting unwanted distraction to the primary task,while still delivering information in an accurate

and timely manner. In many cases, verylittle distraction can be tolerated. Forexample, a typical in-vehicle informationsystem may notify the user about naviga-tion instructions, incoming communica-tions, and other information secondary

from the main task of the user—driving the car.Such systems should be designed to ensure notifi-cation is provided without diverting attentionfrom driving-related tasks. In other cases, a user iswilling to accept some distraction in exchange forvalued information. Desktop computer users mayperform daily word-processing tasks while casually

alert

Why is the attentive user interfaceparadigm important for human-

computer interaction? The humanattention system is so sensitive tovarious methods of notification

that traditional design involves toomuch compromise and guesswork.

� By D. Scott McCrickard and C.M. Chewar

User Goals andAttention Costs

Attuning NotificationDesign to

68 March 2003/Vol. 46, No. 3 COMMUNICATIONS OF THE ACM

maintaining awareness about unrelated information ofinterest, such as that found in a news ticker or an emailstatus indicator. They depend on alerts or alarms tostay informed about critical information and triggernecessary reaction. All too often, however, usersbecome distracted from their current activities byoveractive notifications insensitive to user priorities.

Certainly, systems that manipulate and depend onuser attention should be developed according to user-centered design and cognitive engineering [6] in orderfor the human-computer interaction component tosucceed. Users select and accomplish goals associatedwith system use based on how an interface’s physicalrepresentation of information conforms to expecta-tions. Designers must ensure benefits of presenting anotification outweigh associated costs. Costs and ben-efits of notifications to user attention either must beestablished prior to employment of a system, or canbe determined at runtime by designing notificationsystems that are attentive. Conceptual models assist inthis consideration and allow comparison of presenta-tion alternatives, helping predict what will work.

Tradeoffs of System UseAs a first step toward a conceptual model of usernotification goals, we consider general goals andtradeoffs required for their achievement. We drawfrom the convergence of ideas of researchers pursu-ing these questions. Horvitz characterizes a user’sattention system as the most constraining factor innotification systems design [1]. His paradigminvolves a Bayesian inference model that decideswhether interrupting a user will create sufficientpayoff in terms of expected information. This modelwould be ideal for filtering information to be pre-

sented to the user interested inreceiving valuable notifications,such as the receipt of urgentemail or a reminder for animportant meeting. Maglio andCampbell articulate a similartrade-off describing benefits tothe user in providing additionalinformation with escalatingcosts of obtaining information[3]. To minimize the costs asso-ciated with continuous presen-tation of secondary information

sources, they look for “peripheral” presentationoptions that preserve the focus of user attention.Combining these approaches with sensing of userattention leads to a perspective that fully accountsfor the complete range of user notification goals.

We assert that it is useful to think of “attention” asa constrained resource that can be traded for someutility. This utility is enabled by perceiving addi-tional, valued information while performing othertasks. This attention-utility trade-off can be stated asfollows:

The success of a notification system hingeson accurately supporting attention allocation

between tasks, while simultaneously enablingutility through access to additional

information.

The attention-utility theme concisely captures thesource of scarcity (the attention of the user) alongwith the user’s purpose in using the notification sys-tem (utility associated with access to an additionalsource of information). Certainly this relationship isnot smooth and differentiable, but still generallydescribes the cost of achieving user goals—a cost thatreliably yields benefits when using AUIs to infer thestate of a user’s attention, model priorities, and renderinformation appropriately.

Table 1 itemizes component cost-benefit factors ofthe attention-utility trade-off. Users ultimately use anotification system to gain benefits, which come fromspecific types of utility. We recognize four generalsources of utility that can result from associated usergoals (left side of the table). The general goals of com-prehension, reaction, and interruption can be

utility benefits

information is relatedexisting knowledge and stored for future use

attention costs

user goal general goals cost factors

identify state changesunderstand patterns and trendsassimilate complex informationmonitor resources over timegain awareness of collaborators

make decisionsmodify primary task approachprovide responseacknowledge status

pace daily activitiesprompt task transitionreceive urgent/timely informationsynchronize with colleagues

reduce stressemote humorcultivate enjoymentaugment meaning or presenceincrease feeling of security

Comprehension

Reaction

Interruption

immediate response to a notification stimulus, with or without shifting attention

information is related toexisting knowledge andstored for future use

intentional and inherentlyuseful reallocation ofattention from other tasks

overall enhancement andapproval of the generalcomputing experience

Satisfaction

Context

Usercharacteristics

Informationcharacteristics

goal relationships of taskstask perceptual-motor qualitiesdata-link dependencisrelative tasks prioritiesinterruptabilityfocus/peripheral locationplatforms and environment

skill and automaticitycognitive and perceptual abilitiescurrent overall mental workloadsender/receiver rolesdemographics

granularitydiscrete/continuousmodality (visual or auditory)complexityrepresentation richnessanticipated valuesynchronizatoncontext relevance

situation parameter

Table 1. Attention benefits and costs.Notification system users expect to gainbenefits associated with fulfillment ofuser goals (left side) by sacrificing attention from other tasks. Costs can beexacerbated by factors of the current situation (right side).

COMMUNICATIONS OF THE ACM March 2003/Vol. 46, No. 3 69

thought of as critical parameters—key measures ofsystem success that can be benchmarked to revealdesign progress. These goals are unique in that theuser is willing to sacrifice a certain amount of primarytask attention in order to achievethem. Other important system fea-tures and user needs must be typi-cally supported in user interfaces toinclude privacy, reliability, and trust.These features can negatively influ-ence the amount of required atten-tion without providing a distinctbenefit that independently motivatessystem use.

The level of cost, determined bythe amount of attention removedfrom ongoing tasks, may be elevated as a result of thefactors presented on the right side of Table 1. Forexample, above-average attention cost factors mayinclude a user’s lack of skill in perceiving unfamiliar orcomplex notification information. Unfortunately, costfactors may not carry a constant value across differentsituations or result in expected benefits. Poor designsmay result from a user accepting a certain cost in antic-ipation of a certain utility without actually receivingthat utility. Usually, the attention required for a userto perceive and process a notification is diverted fromattention focus on a primary task, but cost only resultsif primary task performance is negatively impacted.Attention supplied during natural breaks in a primarytask can minimize cost. The many cost considera-tions—and strategies to reduce them—amplify theimportance of inferring and leveraging the state of auser’s attention and semantic value of the notificationfor interface design.

Modeling Notification BenefitsThe attention-utility trade-off provides the founda-tion for a conceptual model of user notification goals

that can improve design decisions for notificationsystems. To appreciate how user notification goalscan vary, as well as how expected information pre-sentation would differ, it is helpful to consider two

intuitive scenarios. Both scenariosinvolve a desktop computer userengaged in an urgent document-processing task who is also inter-ested in stock price information. Inthe first scenario, the user wants totrack performance trends over along period of time and has nointerest in near-term trading. Inthe second, the user wants to mon-itor prices to guide transactiondecisions throughout the day.

In scenario one, the user desiresawareness of stock information, butdoes not want to disrupt the pri-mary task. This user should be ableto casually glance at a display andregister stock information almostperipherally. An interface would failif it explicitly diverts attention withobtrusive animation, colors, orother such presentations. Ambient

systems seem ideally suited for this task. Ambientnotification systems are typically calm or peripheralinterfaces used continuously for an extended dura-tion, allowing users to be aware of state changes anddetect patterns or trends in a memorable way (highcomprehension goal) without prompting significantreaction or introducing interruption (low reactionand interruption goals).

In scenario two, the user is likely to value notifica-tions that interrupt the primary task and guide timelydecision making. Here, the cost of not paying atten-tion to the interruption is higher than performancecosts associated with the primary task. When an alert

Effective notification depends on design attributes capable of preventingunwanted distraction while delivering

critical content in a timely and appropriatemanner. Specific design options affect

support of user notification goals because of the way they affect user attention.

Figure 1. Frameworkreflecting the user goals for interruption, reaction, and comprehension—critical parametersfor system success.Two types of sys-tems, ambient andalarm, are depictedaccording to thegoals they support.

grabs the user’s attention in this situation, he or sheexpects the information will be presented in a valu-able and timely fashion. However, if the notificationdoes not attract attention adequately, the user maynot receive the notification in an opportune manner,implying a failure of the system. This user desires asystem functioning as an alarm—a notification thatprovides valued transition from a primary task (high

interruption goal), prompts animmediate response (highreaction goal), but does notintroduce content worthy of

long-term assimilation (low comprehension goal). Clearly, effective notification depends on design

attributes capable of preventingunwanted distraction while deliveringcritical content in a timely and appro-priate manner. Specific designoptions—such as information layout,use of animation, and graphical encod-ing—all affect support of user notifica-tion goals because of the way they affectuser attention. A conceptual modelshould allow designers to match scenar-ios of use with appropriate informationdesign options (use of motion, displaysize, font and color attributes, amongothers). To model the sources of utility and attentioncost and associate design attributes, we introduce aframework (see Figure 1) that depicts the three criticalparameters—interruption, reaction, and comprehen-sion—as axes. Alarm and ambient systems, describedearlier, are illustrated as well.

The axis scales correspond to the level of impor-tance a user places on benefits resulting from eachparameter (three critical general goals listed in Table1). It is important to note that in plotting systemswithin this framework, there is no ideal blend of para-meters or target point. An attentive notification sys-tem would sense the desired parameter levels and

associated attention costs and convert to the appropri-ate type of display by adapting presentation options tofit the user’s priorities. This form of AUI would pro-vide the ideal rendering of information, balancingdynamic notification needs with attention constraints.Therefore, for our conceptual framework to be usefulit must be able to associate design options with usergoals.

Integrating Empirical Usability Test ResultsMany human factors affect theinformation design options fornotification systems ([10] pro-vides an overview). Challenges ofstandalone interface design arecompounded—as people splitattention across different tasks,design choices ideal for full atten-tion use can fail in notificationsystems. Information presenta-tion options have important

effects on focused attention, especially related to useof audio, colors, and animation (motion). As anexample, text-based animation is a likely choice fordisplays constrained to limited screen space, andwould support the scenarios described earlier (a userprimarily engaged in a browsing task while periodi-

cally monitoring news and stockinformation). Many animationtechniques are available, some ofwhich can be described as “in place”(for example, fading or blasting)and “scrolling” (for example, hori-zontal tickering or rolling). Varia-tions in animation speed, fontattributes, and display size produce

numerous design options that may either be detri-mental to user attention [3] or effectively allow pri-mary tasks to proceed without distraction [5].

Designers must be able to relate effects of various

70 March 2003/Vol. 46, No. 3 COMMUNICATIONS OF THE ACM

user notification goal recommendeddisplay

notrecommended

in-place animations(blast and fade),

small-sized

scrolling animations(ticker),

large-sized

scrolling animations(ticker),

fast update

in-place animations(blast and fade),

slow update

low interruption • minimal attentionreallocation from primary task

high reaction • make decisions,provide response

low comprehension • long-termknowledge gain unimportant

low interruption • minimal attentionreallocation from primary task

low reaction • no immediate responsehigh comprehension • understand

patterns and resources over time

Table 2. User performance tradeoffs for text-based animation in notification displays.

Figure 2. The Irwinnotification system.Displayed areoverviews of severalresources using(from left to right)icons, a graphicalencoding of theselected informationresource, and twotextual views.

information design options to user notification goals.For instance, we conducted two studies in a dual-taskcondition where participants performed a Web pagebrowsing task while monitoring and reacting to sec-ondary information displayed using textual anima-tions. Our studies revealeduser-performance trade-offs evidentin the use of various text-based,smooth animation methods (seeTable 2). That is, for those interestedonly in gaining an understanding ofthe information over an extendedduration, a scrolling stock tickerwould be most effective. In contrast,during periods of active stock trad-ing, when rapid reaction to changesis important, an in-place animationwould be best. Recalling the atten-tion-utility tradeoff, these two examples include differ-ent benefits: the first involves high long-termcomprehension and no immediate reaction, while thesecond requires high reaction and less long-term com-prehension. Using the appropriate type of animationprevents primary task distraction, ensuring minimalnotification cost. To this end, our conceptual frame-work associates research results with user goals, generalclasses of systems, and specific systems (as illustrated inFigure 4).

Usability results viewed through theconceptual framework can also suggestimportant design paradigms. As an exam-ple, we consider two conclusions gleanedthrough evaluation of actual notificationsystems: Irwin [4] and the Scope [7].Irwin was designed as a small, omnipresenttool that assists users in maintaining aware-ness about Internet resources such as emailfolders, Usenet newsgroups, Web pages,and weather data. Information is gatheredfrom several sources and displayed on acentral visualization; various icons, colors, and audi-tory cues keep its user updated (see Figure 2). Users ofIrwin were observed over a five-month period, leadingto identification of many usability problems. Thesecan be reduced to a single key challenge: determininghow to notify without distracting, yet providingexpected urgency according to dynamic fluctuationsin user goals. Our conceptual framework providesclarity—during an extended period of use, a pointrepresenting user notification goals may movethroughout the design space quite radically. To matchinformation design schemes with changing goals, thechanges must be anticipated (which is quite difficult)or dynamically sensed—an advantage offered by the

attentive system paradigm.As a second example, consider the Scope notifica-

tion system [7] (see Figure 3), an AUI for alerting andproviding overview about incoming email, calendartasks, and other information. Since the system learns

a user’s priorities, the interface canpresent information according toinferred expectation of urgency. LikeIrwin, the Scope presents a summaryof several resources in a glanceable,omnipresent view. However, theScope is unique in its goal to clearlyconvey notification urgency (asinferred by the system) by presentingnew items accordingly. Interfacechoices such as the circular radarmetaphor, pulsing icons, and fly-inanimation communicate urgency.These information design options(shown to have various strengths andweaknesses) can be improved withresults from basic attention research

associated with regions of the conceptual framework.The AUI characteristic of the Scope provides a dis-tinct advantage: presentation requirements are con-tinuously refined according to dynamic insight aboutuser expectations—once the best presentation options

are adopted for thisinterface design, thereal-time knowledgeabout user prioritieswill maximize theattention-utilitytradeoff.

Completing theDesign ViewBoth systems describedhere support varioususer notification goals.To more fully conveythe usefulness of ourconceptual frame-work and the poten-

tial of AUIs, Figure 4 illustrates design model plotsof Suitor (see [2] and the article by Maglio andCampbell in this section), the Scope, and Irwin. TheSuitor AUI determines a user’s notification priori-ties, shifting design models appropriately duringruntime. For example, if the user is an active investorand the system notices that he or she is browsing forinformation on IBM, Suitor will display stockquotes about the company. This supports the goal—inferred from the user’s attention—of reacting to

COMMUNICATIONS OF THE ACM March 2003/Vol. 46, No. 3 71

Figure 3. TheScope notificationsystem [7].

Figure 4. The framework for usergoals traded for attention resources.Shown here are plots for three systems and two animation options.

stock price variations for a company of interest(point a). Based on our empirical studies, Suitorshould use in-place textual animation to best sup-port user goals and maintain lowest attentional cost.By contrast, if the user shifts to a document-editingtask, Suitor displays helpful tooltips intended toincrease understanding of the document editor(point b). As a result, the recommended displaychanges to scrolling animation, supporting long-term comprehension. The second system, the Scope,receives notifications of high or low priority, causingthe design model to shift. Based on the designers’descriptions, users will not want to be interruptedby low-priority items, but desire comprehension ofthese notifications throughout the day (point c).However, immediate reaction is an important goalfor high-priority items (point d). Finally, as a non-AUI, Irwin has only one plot because the systemdoes not adapt to user attention. The Suitor and theScope examples reflect the enormous advantages ofAUIs for meeting diverse user goals with informa-tion display specifically adapted to keep attentionalcosts minimized and utility benefits maximized.

The Vast Potential of Attentive Notification SystemsIn developing our conceptual model of notificationuser goals, we recognize that AUIs have the potential tobecome the notification systems of choice. Several fac-tors summarize the importance of the AUI paradigm:

• AUIs introduce the ability to model and adapt toa user’s attentional state, bringing the right infor-mation at the right time to the user in a way thatis not achievable with a traditional notificationsystem.

• Systems trained to individual characteristics canprevent problems associated with cognitive differ-ences and interface learnability.

• AUIs can sense change in user goals and adapt adesign model and information presentationappropriately.

Responding to the challenges of notification designand to help harness the attentive paradigm for notifi-cation design, we have introduced the attention-util-ity trade-off as a foundation for conceptual modelingof user notification goals. In a review of usability test-ing, we showed how presentation options can differ insupport of goal-related utility and impact to userattention. Our framework, based on critical parame-ters, integrates user goals, system design models, andpresentation options—simplifying design choices fordevelopers and suggesting concerns for researchers.

Considering the growing demand for ubiquitousand multitasking systems, this underlying para-digm—and the AUIs it produces—will become cen-tral to computing and human-computer interaction.However, there are many challenges for this emergingresearch community. The framework introduced hereprovides a widely inclusive design space that shouldbe filled with existing systems and analyzed to iden-tify the best places for AUI augmentation. The com-munity should also endorse universally acceptedcritical parameters, which can support a reference-task research agenda [9]. We provide a suggestion, butother possibilities may lead to improved modeling ofuser goals. Standard reference tasks should beselected, publicized, and adopted as a common met-ric for system testing. Finally, interface evaluationsshould be conducted and reported for the purpose ofachieving scientific growth—allowing the commu-nity to recognize and leverage the great benefits andadvancements afforded by AUIs.

References1. Horvitz, E. Principles of mixed-initiative user interfaces. In Proceedings

of the ACM Conference on Human Factors in Computing Systems (CHI‘99). ACM Press, NY, 159–166.

2. Maglio, P.P., Barrett, R. Campbell, C.S., and Selker, T. SUITOR: Anattentive information system. In Proceedings of the Conference on Intel-ligent User Interfaces (IUI 2000), ACM Press, NY, 169–176.

3. Maglio, P.P. and Campbell, C.S. Tradeoffs in displaying peripheralinformation. In Proceedings of the ACM Conference on Human Factorsin Computing Systems (CHI ‘00). ACM Press, NY, 241–248.

4. McCrickard, D.S. Maintaining information awareness with Irwin. InProceedings of the World Conference on Educational Multimedia/Hyper-media and Educational Telecommunications (ED-MEDIA ‘99).

5. McCrickard, D.S., Catrambone, R. and Stasko, J.T. Evaluating ani-mation in the periphery as a mechanism for maintaining awareness. InProceedings of the IFIP TC.13 Conference on Human-Computer Interac-tion (INTERACT 2001), 148–156.

6. Norman, D.A. Cognitive engineering. User Centered System Design. D.A.Norman and S.W. Draper, Eds. Lawrence Erlbaum, Hillsdale, NJ,1986.

7. van Dantzich, M., Robbins, D., Horvitz, E., and Czerwinski, M.Scope: Providing awareness of multiple notifications at a glance. InProceedings of Advanced Visual Interfaces (AVI 2002).

8. Vertegaal, R., Velichkovsky, B., and Van der Veer, G. Catching theeye: Management of joint attention in cooperative work. SIGCHI Bul-letin 29, 4 (1997).

9. Whittaker, S., Terveen, L., and Nardi, B.A. Let’s stop pushing theenvelope and start addressing it: A reference task agenda for HCI.Human-Computer Interaction 15, 2-3 (2000), 75–06.

10. Wickens, C.D. and Hollands, J.G. Engineering Psychology and HumanPerformance, 3rd Edition. Prentice Hall, Upper Saddle River, NJ, 2000.

D. Scott McCrickard (mccricks@cs.vt.edu) is an assistant professor in the Department of Computer Science and a member ofthe Center for Human Computer Interaction at Virginia PolytechnicInstitute and State University, Blacksburg, VA.C.M. Chewar (cchewar@cs.vt.edu) is a graduate student in theDepartment of Computer Science at Virginia Polytechnic Institute andState University, Blacksburg, VA.

© 2003 ACM 0002-0782/03/0300 $5.00

c

72 March 2003/Vol. 46, No. 3 COMMUNICATIONS OF THE ACM