COMMUNICATIONS OF THE ACM May 2000/Vol. 43, No. 5 43Human-in-the-loop computing has its limits. What must we do differently to prepare for the networking of thousands of embedded processors per person? And how do we move from human-centered to human-supervised computing?For the past 40 years, most of the IT research community hasfocused on interactive computing, J.C.R. Licklider’s powerful andhuman-centered vision of human-computer symbiosis [3]. In tan-dem with this research has come the creation of an IT industry thatis hurtling toward the human/machine/network breakpoint—thepoint at which the number of networked interactive computers willsurpass the number of people on the planet. We still have a long way to go before Licklider’s vision is attained—and aremany years from extending per-capita penetration to most parts of the world.However, “missing science” may no longer be the factor limiting progresstoward these long-cherished goals. It is reasonable, though perhaps heretical,to suggest that refinements of the existing science base will be sufficient todrive these efforts forward. It is time for a change. The computer science research community nowenjoys a rare and exciting opportunity to redefine its agenda and establish thenew goals that will propel society beyond interactive computing and thehuman/machine breakpoint. In lifting our sights toward a world in which net-worked computers outnumber human beings by a hundred or thousand toone, we should consider what these “excess” computers will be doing and crafta research agenda that can lead to increased human productivity and qualityof life. David TennenhousePROACTIVECOMPUTINGWhat Should We Do Differently? Although I lack Licklider’s clarity as to what the next40 years of computation might bring, I am con-vinced that the first steps toward a new agenda mustinclude: a fundamental reexamination of the bound-ary between the physical and virtual worlds; changesin the time constants at which computation isapplied; and movement from human-centered tohuman-supervised (or even unsupervised) comput-ing. While some work has been done in each ofthese areas, focusing significantly greater attentionon them will enable a new mode of operation,which I refer to as proactive computing. In this article, I describe three loci for newresearch activities:Getting physical. Proactive systems will be inti-mately connected to the world around them, usingsensors and actuators to both monitor and shapetheir physical surroundings. Research into “gettingphysical” explores the pervasive coupling of net-worked systems to their environments. Getting real. Proactive computers will routinelyrespond to external stimuli at faster-than-humanspeeds. Research in this area must bridge the gapbetween control theory and computer science.Getting out. Interactive computing deliberatelyplaces human beings in the loop. However, shrink-ing time constants and sheer numbers demandresearch into proactive modes of operation inwhich humans are above the loop.There are two simple reasons why we should divertsome of our intellectual resources to proactive com-puting: The vast majority of new computers will beproactive, and these nodes will be the principalsources and sinks of information.The bulk of the IT industry is presentlyfocused on office automation, e-commerce,and their associated networking. Judging byour current research profile, an independentobserver might believe that the distributionof new computers is dominated by the 150million or so new laptop, desktop, andserver nodes that will power the growth ofinteractive computation. Although the creation of an industry thatconsumes such a large number of computersper year is a tremendous achievement, thesenumbers pale by comparison to the eight-billion-or-more computational nodes thatwill be deployed worldwide this year. Asshown in Figures 1 and 2, the vast majorityof these devices will be embedded in otherobjects. Rather than being in direct contact withhuman beings, they will be in direct contact withtheir environments—able to monitor and shape thephysical phenomena that surround them. Since the rate of growth of these embeddeddevices exceeds that of their interactive cousins, thecomputer science research community has no choicebut to follow the processors and invest a larger frac-tion of its intellectual capital in this space. In doingso, we would also be following the data, movingfrom environments in which our sources of informa-tion are largely human-mediated to those in whichcomputers directly tap tremendous sources ofgrounded information concerning the world aroundthem. In fact, the essential reason for having moredevices than people, and for having them be distrib-uted, rather than placed in glass rooms, is their inti-mate connectivity with the physical world—and theincremental sources and sinks of information theyprovide.44 May 2000/Vol. 43, No. 5 COMMUNICATIONS OF THE ACMProactive■ Pervasive■ Human-supervisedInteractive■ Office/Document■ Human-centeredOffice vs. FieldManual vs. AutonomousFigure 1. The four quadrants of ubiquitous computing.Figure 2. Where will the computers be?150MperyearEmbeddedComputers80%FixedInfrastructureVehicles8Bparts/yearRobotsInteractiveComputersServersRobots 6%Vehicles 12%Interactive 2%Where has computer science focused? Where are the processors?Why Now? To date, Internet deployment hasfocused on breadth, expanding thegeographic reach of the network toinclude nodes “in every office andevery home.” Although this broad-ening of the Internet will continueat an impressive rate, the numberof nodes can be increased an addi-tional 50-fold by reaching downinto all of the embedded devices ateach geographic location. This shift toward deeply net-worked systems represents a pro-found inflection point thatdemands our attention because ofits sheer scale and systems implica-tions. Historically, the lack of network connectivityhas stranded the data obtained by embeddedprocessors and led to rigid software regimes con-strained by one-time programmability. However,various efforts are beginning to unlock the informa-tion derived by huge numbers of sensors and pro-vide remote access to their actuators (see Pottie’s andKaiser’s “Wireless Integrated Network Sensors” inthis issue). Isn’t this the same as ubiquitous computing? In his1991 article [8], Mark Weiser, who was chief tech-nologist of Xerox Palo Alto Research Center, forecastthat computation