Digital Rebound – Why Digitalization Will Not Redeem Us Our Environmental Sins Vlad C. Coroamă and Friedemann Mattern Department of Computer Science ETH Zurich Zurich, Switzerland {vcoroama, mattern}@ethz.ch Abstract— Digitalization as a technological phenomenon of the The continuingly rapid digitalization of societies and 21st century has the power to redeem most environmental sins economies also has its downsides and does not go undisput- of our 20th century technology. This seems to be a popular ed. Some of the more obvious reasons for concern are possi- belief shining through many of the optimistic media reports on ble security breaches due to the increasing complexity, het- digitalization. We believe, however, that this mindset is far too simplistic and counterproductive. The many indirect economic erogeneity, and interconnectivity of systems, as well as some and social effects of digitalization, which turn efficiency gains increasingly intricate privacy issues. Such negative side ef- into increased resource consumption, are often ignored. We fects seem to be generally accepted by society as the many discuss these countereffects in general, as well as their digitali- benefits of digitalization are perceived to largely outweigh zation-specific flavor (i.e., the digital rebound). We give exam- these disadvantages (which are believed to be manageable to ples of digital rebound, and also analyze several conditions that some degree). seem to lead to its eschewal. Altogether, we try to make the Increasing evidence, however, also sheds a critical light case for a faithful consideration of the rebound effects of digi- on the attributes of digitalization usually perceived as wholly talization. positive. It turns out that the increased efficiency or the im- Index Terms—Rebound, efficiency, resources, energy, digital rebound. proved access to information afforded by digitalization can often induce indirect effects, which can reduce or even re- verse its positive impact. In economics, these unwanted I. INTRODUCTION countereffects are known under the umbrella term of re- The potential economic and societal benefits of digitaliza- bound effects. tion1 are far-reaching and are often addressed in today’s pub- In a nutshell, rebound effects occur when positive initial lic discourse. Moreover, digitalization is often envisioned as effects (e.g., increased efficiency) make a good or service a silver bullet to tackle – or at least mitigate – the world’s more attractive (through lower prices or added benefits), increasingly urgent environmental issues; in particular, it is which in turn is likely to spur demand either for the same seen as a possible key factor in reducing carbon emissions good or service (which is more attractive), and maybe for and resource consumption across various economic sectors other products, due to the increased disposable income or (e.g., [1-3]). Such assertions rely on the ability of digital sys- time. This, in turn, stimulates more energy and resource con- tems to either optimize the performance of energy- and re- sumption (and consequently more pollution), diminishing the source-intensive systems and processes in industry and initial positive effect or, in the worst case, even outweighing commerce, or to virtualize and substitute them altogether. it. Digitalization can further enable more environmentally de- While relatively well-known in economics, rebound ef- sirable solutions, which would be too complex to achieve or fects have not yet been thoroughly investigated for digital manage otherwise, such as the smart electrical grid. Finally, goods and services, and even less so for the broad digitaliza- through more detailed, real-time information at their finger- tion of whole industrial and economic sectors. This is partly tips, consumers can decide in favor of more environmentally understandable because, as will be shown below, rebound friendly alternatives, such as buying goods with a small car- effects are diverse and involve subtle yet far-reaching mech- bon footprint or avoiding products with palm oil. anisms. Although their principal workings are relatively well understood, quantifying rebound effects remains a challeng- ing task. As digitalization pervades ever growing areas of 1 Traditionally, the term “digitalization” basically meant the technical societies and economies, and given the broad dissipation of process of converting analog signals into digital form. In recent years, the effects, assessing the rebound effects of digitalization is a however, digitalization took on a much wider meaning – in business contexts, it now stands for the broad use of digital information and particularly serious challenge. communication technology (ICT) and the induced change in business We use the umbrella term digital rebound to denote any operations or whole business models (“digital transformation”), often such rebound effects induced by digitalization technologies, restructuring or disrupting economic processes and social practices. whether they stem from individual IT goods and services, the Better inclusion of the elderly or disabled means they will digitalization of entire economic sectors, or indeed the whole also be able to ride autonomous vehicles instead of public economy. Ignoring digital rebound can lead to a misunder- transport, worsening the environmental impact of their mo- standing of the environmental effects of digital technologies, bility [4]. Even children could ‘drive’ autonomous vehicles and possibly result in inappropriate policy or misallocated to school! Self-driving cars are also likely to induce a sub- monetary incentives. Despite its difficult quantification, this stantial number of empty runs [11], an impossibility today. paper thus aims to increase digital rebound awareness. Until now, one of the reasons not to take the car in urban Section II starts with a familiar example for emerging environments has been the difficulty of finding a parking digitally enhanced products, self-driving cars, discussing spot at the destination. If one can, however, drive to a meet- some of their possible rebound effects. Section III presents a ing in the city center and send the empty car back home, it is more in-depth theoretical analysis of several types of re- quite likely that such empty runs will occur, inducing addi- bound effects. Section IV then shows the relevance of re- tional mileage [12]. As Chase [13] pointedly puts it, these bound effects in the context of digitalization, discussing both induced trips could be far beyond what we might imagine apparent environmental benefits and also the counteracting today: “I schedule the FAV [fully autonomous vehicle] to digital rebound for several types of digital services. By con- return at 9:30 a.m., but I don't rush out because the car will trast, Section V examines some digital services with little or just circle the neighborhood until I tell it I'm here! As I get a no rebound. Finally, Section VI contrasts the two categories, friend a gift at a hand-made jewelry shop, my FAV circles distilling insights into the design of digital services that seem the block for 15 minutes. Rather than trip-chaining to get the to be truly environmentally beneficial even after taking pos- dry cleaning, we send the FAV out anytime to pick it up (an sible digital rebound into account. employee places the cleaned and pressed clothes in my car for me). Ditto for our take-out dinner”. II. SELF-DRIVING CARS: ENVIRONMENTAL Finally, the time spent in an autonomous vehicle is likely CURE OR MENACE? to be more enjoyable or productive than when driving one’s With recent advances in computer vision technologies based self. The time while riding an autonomous vehicle free of on pattern recognition and machine learning, paired with stress or attention can be used for socializing or work. This is progress in other digital technologies such as wireless com- likely to increase the appeal of car rides, which might lead to munication and high-precision localization, self-driving cars more frequent and longer trips [7]. Car rides would also be- – or autonomous vehicles – are now expected to become a come more attractive as compared to other modes of reality in the not too distant future. Major car manufacturers transport, leading to a partial substitution of the former for as well as IT companies are developing technologies for au- the latter. This substitution was theorized for example in tonomous vehicles. Since it receives much media coverage, [14], while a questionnaire of paired comparisons devised in the topic is one of the better-known examples for how digi- [11] hints that shared autonomous vehicles might indeed talization can permeate various sectors of the economy and displace almost exclusively public transport, not private car society, and thus serves well as an introductory example case ownership. for our statements and claims. Self-driving cars can bring about undeniable societal III. TYPES OF REBOUND EFFECT: AN OVERVIEW benefits, such as better inclusion of the elderly or people with All of the above are examples of rebound effects for autono- disabilities [4, 5]. Additionally, numerous researchers have mous driving. These noteworthy effects do, however, also also highlighted their potential benefits on traffic and the appear in other contexts. Before analyzing their relevance for environment. Some [6] alleged that autonomous taxis could the broad domain of digitalization, it is worthwhile to gain a considerably reduce vehicle emissions, while others [7] ar- deeper theoretical understanding of rebound effects in gen- gued that platooning (coordinated travel in close proximities eral. on highways) can substantially reduce the average fuel con- Several definitions of rebound effects exist, some nar- sumption by coordinating driving speed and behavior, and by rower, others wider. In its classical economic interpretation, minimizing the distance between vehicles to reduce wind the notion of rebound evolved from describing one rather resistance. It has even been argued that autonomous vehicles narrow phenomenon in the energy market to an entire class are inherently safer than traditional vehicles driven by hu- of effects. A definition of today’s broader understanding is mans, and thus require lower safety standards, which in turn given by Sorrell [15]: “The ‘rebound effect’ is an umbrella leads to lower vehicle weight and thus lower fuel consump- term for a variety of mechanisms that reduce the potential tion [8]. Finally, some argue that the emergence of autono- energy savings from improved energy efficiency.” While mous vehicles would boost the market for sharing such vehi- broad, this definition still considers only the energy domain. cles to the detriment of private car ownership, reducing the As Binswanger [16] comments, however, the concept of re- overall car fleet and thus the grey energy required for vehicle bound effect can easily be applied not only to energy, but to manufacturing [9, 10]. resource use in general. These positive direct effects, however, only tell half of the story. There are also a number of subtler mechanisms and indirect consequences that induce effects to the contrary: A. The Direct Rebound Effect gy prices, energy-efficiency gains will increase energy con- Khazzoom [17] undertook an early systematic analysis of the sumption above what it would be without these gains” [20]. rebound effect. His approach relies on a single-service mod- Saunders calls it “the Khazzoom-Brookes postulate”, after el; meaning there are no repercussions from this service to the more recent work by Brookes [21]. As both Alcott [19] the rest of the economy. The service is an energy-intensive and Sorrell [15] observe, ‘postulate’ is the correct term in one, such as mobility (measured in passenger-km) or room this context as there is not enough evidence to support that temperature. According to neoclassical economic theory, the rebound always exceeds 100%. Discussing Jevons’ work, when the price of a good decreases, the demand for it in- Alcott observes “Jevons thus makes rebound theoretically creases, all other things being equal. If, due to advances in plausible, but he has not yet proven that the amount of coal energy efficiency (e.g., more fuel-efficient vehicles or better consumed must ‘more than’ make up for engineering sav- house insulation), the passenger-km or an hour of a certain ings” [19]. Likewise, Sorrell concludes that “such evidence room temperature will cost less, and as long as their needs does not yet exist” [15]. are not saturated, users will tend to use them more: more C. Indirect Rebound: Induction Effect, Income and kilometers driven, the room temperature set higher or not Substitution Effects, Producer Rebound turned off overnight. This effect may partially or entirely The first citation from Jevons’ work above already hints to- offset the savings from the original energy efficiency meas- wards more mechanisms than the mere direct rebound. An- ure. other revealing passage can be found on page 144: “Whatev- In this narrow sense, the rebound is often referred to as er, however, conduces to increase the efficiency of coal, and direct rebound effect – direct because the rebound occurs for to diminish the cost of its use, directly tends to augment the the same service that had originally gained in efficiency, and value of the steam-engine, and to enlarge the field of its op- because the rebound is a direct consequence of the price re- erations” [18, 19]. The mechanism described here alludes to duction that follows the lower input to produce the service. the induction effect [22], which other researchers consider Although originally defined for energy markets, the effect merely a specific form of the rebound effect [23]. appears for any resource efficiency measure: if less of a re- Such mechanisms that lead to different types of rebound source (any physical resource, though, in the general sense, were more formally presented soon after Khazzoom’s work. also more labor or capital) is needed to produce a good or Both Binswanger [16] and Berkhout et al. [24] discuss the service, its price will decrease and, as a result, more of it will income effect and the substitution effect as further causes for be demanded. rebound. The effects are well-described in [16]. They are B. Jevon’s Paradox or Backfire observed by leaving the single-service model behind and More than a century before Khazzoom’s work, British econ- considering a model consisting of two services, A and B, omist and logician William S. Jevons first referred to the which can be partially substituted for each other. A lower phenomenon – without using the term ‘rebound’ – in his price for service A, as a consequence of efficiency gains for 1865 book “The Coal Question” [18]. The effect described one of its inputs, has two consequences: i) consuming the by Jevons is different from Khazzoom’s rebound in that it is same amount of A and B becomes cheaper, the consumer has more general (caused by more mechanisms) than the mere a larger budget at his disposal, leading – ceteris paribus – to direct rebound put forward by Khazzoom. This will be dis- more consumption of both A and B (income effect); and ii) as cussed below. service A becomes relatively cheaper, it will partially substi- Despite attributing it to different causes, Jevons and tute service B (substitution effect). The total effect is equal to Khazzoom agree on the rebound’s size. They both assume the sum of the two effects, as reflected by the Slutsky equa- that it is larger than 100%, i.e. it is postulated to outweigh the tion [25]. Both effects lead to more consumption of service original savings. As broadly discussed by Alcott [19], Jevons A, and thus also of the resource that had originally gained argues in his original work that the rebound effect not only efficiency, which triggered these effects in the first place. reduces the potential savings of the energy efficiency meas- Berkhout et al. [24] also define what they call the pro- ure, but that it actually outweighs the reductions, leading to ducer rebound, which is essentially a substitution effect on an overall net energy increase: “[if] the quantity of coal used the producer side: Increased energy efficiency changes the in a blast furnace, for instance, be diminished in comparison optimal balance between energy and other production factors with the yield, the profits of the trade will increase, new capi- such as labor or capital. Due to the more efficient usage of tal will be attracted, the price of pig iron will fall, but the energy, the producer will, to some extent, substitute energy demand for it increases and eventually the greater number of for capital or labor. furnaces will more than make up for the diminished con- D. Time Rebound sumption of each” ([18], page 156). Binswanger [16] introduces what he calls time rebound, This particular case, when the magnitude of the rebound which stems from time-saving technological progress. He effect is more than 100%, is known in the literature as Jev- argues that a decline in the time needed to acquire a service ons’ paradox, or under additional names such as boomerang (such as traveling a certain distance) reduces the costs asso- or, more commonly, backfire. A well-known formulation of ciated with time. This is based on the economic model that Jevons’ paradox is given by Saunders: “with fixed real ener- someone’s time can be monetarily represented by the fore- gone earnings one could have achieved during that time. nomic scale because of undefined and unclear system bound- Economists say in this context that “wages are the opportuni- aries and sector-wide spillover and feedback mechanisms. ty costs (i.e., the not taken alternative, hence ‘opportunity Reviewing a large body of rebound literature, particularly costs’) of time.” by economists Len Brookes and Sam Schurr, Sorrell [15] A time efficiency measure, thus, leads to time saving points to another source of macro-level rebound: the catalyst which can be monetarily expressed as its opportunity costs, effect of energy for productivity in general. He argues that i.e., the earnings that could theoretically be achieved in the energy efficiency technologies boost total factor productivity time that was saved. To the extent that the costs of the time- (in particular, capital and labor productivity) and thereby saving measure continue to be cheaper than the costs of save much more than energy costs alone. Moreover, he ar- saved time, the former will be substituted for the latter. gues that labor costs are much higher than energy costs (typ- Time-saving technologies, however, are often quite energy ically, 25 times larger in commercial buildings in industrial- intensive, such as the technologies enabling fast means of ized countries). But if the total cost savings are much larger travel or transportation. The energy thus spent to save time, than energy savings alone, the rebound due to the income is what Binswanger calls “time rebound.” effect may also be much larger. This observation seems to apply only to energy efficiency measures and not to resource E. General Equilibrium Effects and Other efficiency in general. Macro-Level Rebound Finally, the price changes for the firms’ output, as well as the IV. DIGITALIZATION AND ITS REBOUND income and substitution effects that follow efficiency gains, The last paragraph of Section II mentioned several examples will lead to changes in demand and further readjustments of rebound effects for self-driving vehicles: Riding autono- along the entire economy. These general equilibrium effects mous vehicles, which can be much more affordable than taxi are relatively hard to grasp and almost impossible to quanti- rides and might thus displace trips via public transport are fy. In the literature, they are also called macroeconomic re- examples of the substitution effect [24]. New categories of bound [26] or world-wide rebound [23]. users such as the elderly, disabled, or even children ‘driving’ One reason the global and long-term consequences of vehicles is a form of the induction effect revealed by Hilty products becoming cheaper (due to energy efficiency im- [22]: the ease of accessing or using a service creates new provements or technical progress in general) are difficult to demand. Induction effects are also the empty runs, which do assess (and even more so to predict), lies in the fact that con- not exist in a world without autonomous vehicles, such as the sumers (and thus markets) may react in a non-linear and al- car circling the neighborhood waiting for the owner to finish most discontinuous way to price changes and product im- a business meeting. Car rides becoming more attractive as provements. Indeed, once a certain price or usability barrier they can be used for either work or socializing illustrates is surpassed, a product may suddenly become attractive to Binswanger’s time rebound [16]. As these phenomena result buyers. Emotional or networking effects, and even trends in from digitally-enabled autonomous vehicles, they can thus fashion, are certainly also relevant for such avalanche effects all be considered examples of digital rebound. and add to the complexity of their analysis and assessment. Numerous further ICT-based products and services that For example, no one could have predicted the sudden are popular for their efficiency gains or other resource-saving boom of mobile phones. Car phones existed since the 1960s mechanisms are in fact prone to digital rebound. We present and have steadily been improved, evolving into portable two examples below. phones during the 1990s (“car phones without a car”, as an advertisement at that time nicely put it). But only when they A. Teleworking became small enough to fit into trouser pockets and could Teleworking, also called telecommuting, denotes working run without heavy batteries, mobile phones quickly became a from a remote location without physically commuting to the real market success (clandestinely paving the way for the office. Communication with colleagues and access to com- next evolutionary step, their metamorphosis into smart- pany data are ensured via digital means such as email, Skype phones). and similar services, virtual private networks, screen sharing, The basic technological driver of the digitalization phe- etc. The physical location of work is often the employee’s nomenon is the steady progress (and, in fact, the steady effi- home although telework can also be performed from a holi- ciency improvements) in microelectronics neatly revealed in day spot, the partner’s house, etc. Moore’s Law. Sustained steady progress on that level, how- Teleworking has the potential to significantly reduce ever, can eventually lead to sudden disruptions on the macro commuting and therefore energy use for personal transport. scale: We now spend much more time with our mobile This can be a significant reduction since the transport sector phones than we did previously with our landline phones. But represents around 25% of the final energy demand in devel- when doing so, do we directly or indirectly use more energy? oped economies, 1/3 of which can be attributed to work Whether an avalanche effect turns into a digital rebound ef- commute [27]. Early studies have indeed indicated important fect on the global scale is a priori unclear and certainly de- reductions of both passenger vehicle use and traffic conges- pends on the circumstances of the particular case. In general, tion due to telecommuting. In 1991, [28] concluded that tel- cause and effect relations become blurred at the macroeco- eworking in the Netherlands decreased the total number of trips taken by teleworkers by 17% and peak-hour traffic con- B. E-commerce gestion by 26%. A California pilot project [29] in the same E-commerce describes a variety of commercial practices, in year resulted in 75% less distance travelled by teleworkers which the Internet is central to ordering goods. When the on their telecommuting days. A couple of years later, a dif- goods to be delivered are digital, or can be digitalized (such ferent study yielded a similar 77% reduction in distance trav- as music, movies, or books), their delivery can also take elled for the same Californian pilot project [30]. place digitally (via Internet streaming), without a physical Later studies addressing the possible rebound effects of substrate such as a DVD, CD, or paper. teleworking, however, paint a mixed picture. For example, It has long been maintained that E-commerce is more en- [31] emphasizes that telecommuters can no longer stop for ergy efficient than traditional retail. Sivaraman et al. [34], for shopping on the way home from work, but might take an instance, compared two DVD rental networks: a traditional extra trip by car for their shopping. (Empirical work, howev- one in which the customer drives to the rental shop, on the er, has shown that such non-commute travel on telecommut- one hand, and online ordering followed by mail delivery, on ing days decreases as often as it increases [32], and [27] the other. Even though the respective online model did not speculates this might be because some non-commute trips take advantage of online streaming but was still delivering could be eliminated as, without the work commute, their physical CDs, the study found that it nevertheless consumed destinations would be too far away to be attractive.) 33% less energy and emitted 40% less CO2 than the tradi- Beyond the uncertain development of non-commute trips tional option. Similarly, [35] concluded that online grocery on teleworking days, there might be several other causes for order with subsequent home delivery can save between 18- telecommuting-induced digital rebound. A study [33] esti- 87% of the CO2 emissions of individual grocery shopping in mates that the 4 million US workers who telecommute one or Finland. more days per week reduce the country’s primary energy However, [34] already found that e-commerce consumes consumption by 0.13-0.18% and its greenhouse gas emis- more energy in urban areas where, in the traditional model, sions by 0.16-0.23%, and it lists two likely causes for re- customers usually do not drive to the shops but walk or take bound: For one, telecommuting could increase the number of public transportation, while home deliveries are done by weekend trips to compensate for the activities not performed vans. Going one step further, and analyzing book delivery in during the week, such as shopping. Moreover, as they spend Japan, [36] showed that home delivery of books does not less days commuting to work, teleworkers could live further perform better environmentally than the traditional model in away from their workplace, increasing their commute effort suburban or rural areas, either. In contrast to the other two to work on non-telecommuting days and, potentially, that of studies, [36] took the multipurpose use of car trips into con- numerous other trips. One could add easily imagined scenar- sideration. Therefore, not driving to the city’s bookstore ios wherein the family car is happily used by other family saved almost no energy in the end, as the car trip still took members for their yet unmet demands, rather than resting in place for other purposes, while the induced consumption of the garage when the main income earner does not commute delivery trucks turned the e-commerce balance into the nega- to work. tive. This effect is probably more prominent for clothes or- Widening the boundaries of its analysis, [27] accounts for dering, where customers often order more models and sever- the decreased energy consumption in commercial buildings al sizes of each, and then take advantage of return deliveries. due to teleworking and, at the same time, for the increased energy consumption in residential buildings, many of which V. DIGITALIZATION WITHOUT REBOUND would have otherwise been unoccupied during the day. For We will now discuss some digital services that, in contrast to the teleworking practices of 2005, and accounting for uncer- the examples presented above, seem to produce only a small tainties, it estimates national energy savings of only 0.01- rebound, if any. 0.4% in the US, and 0.03-0.36% in Japan. Even for an ex- treme future scenario with ubiquitous teleworking, in which A. Rebound with a Smaller Footprint: A Trip is not a Trip 50% of information workers telecommute 4 days per week, In 2009, the first World Resources Forum (WRF) was orga- the national energy savings are estimated at only about 1% in nized simultaneously in Davos, Switzerland and Nagoya, both cases because of the many countereffects. Japan. This conference format was chosen so that the confer- Finally, [31] argues that “online work can produce new ence would stay truthful to its topic of resource efficiency; contacts that might generate the need for meeting people the expectation of the organizers being that offering confer- personally”. The first author of this paper can confirm the ence venues on two different continents would reduce inter- occurrence of such induction effects from personal experi- continental travel. For the four hours of daily common ses- ence: Between February 2015 and August 2016, he was re- sions (due to the 7 hours time difference), the two venues motely employed by the KTH Stockholm while living in were connected with telepresence services (i.e., highest qual- Bucharest, Romania for family reasons. Without modern ity videoconferencing), adapted from its usage among small digital communication technologies, this collaboration would teams in meeting rooms to audiences of hundreds of at- not have been possible, nor would the induced travel (11 tendees [37]. return flights jointly responsible for around 10 t CO2e) have As travel to the conference became, on average, shorter, taken place. simpler, and cheaper, a rebound effect in the number of par- ticipants was to be expected as compared to a regular single- C. Market Saturation Reached: Gas Leakage Discovery site conference: 531 participants attended in either Davos Natural gas is a popular source for heating energy, consisting (372) or Nagoya (159). Had the conference been organized primarily of methane (CH4) together with smaller quantities in Nagoya only, approximately 238 people would have at- of other hydrocarbons. Both the US and Europe have extend- tended; the 159 who came anyway plus 79 of the 372 from ed natural gas transmission and distribution networks. The Davos. Had it been a Davos-only conference, the 372 local US transmission network, for example, consists of over attendees would have been joined by 76 from Nagoya for a 300,000 miles of interstate and intrastate transmission pipe- total of 448 [37]. lines, while the distribution network contains more than one This means that the two-venue event generated indeed a million miles of low-pressure pipes [41]. As with any other rebound in the number of participants when compared to pipes, natural gas transmission and distribution networks are either of the traditional organization modes, 531 as compared prone to leaks, through which gas can be released into the to 238 and 448, respectively. Despite this increased partici- atmosphere. pation, the distributed conference had a lower travel-related Methane, though, is a potent greenhouse gas. Overall, it impact as compared to the traditional alternatives (119 t CO2 represents the second most important source of anthropogen- as compared to 189 t and 235 t, respectively) [37]. This is due ic warming after carbon dioxide (CO2); its relative impact, to the fact that the efficiency gains induced by the distributed however, is much higher: Over a time period of 20 years, a organization method implied a substantial reduction in inter- certain amount of CH4 has a warming effect 72 times greater continental travel. The rebound travel instances, on the other than the same mass of CO2 (and, although the atmospheric hand, were almost exclusively much shorter intra-continental lifetime of CH4 is shorter, the effect is still 28 times greater trips. As trips have very different energy and carbon foot- over a period of 100 years). Anthropogenic sources are esti- prints, which are generally directly proportional to their mated to be responsible for around 60% of the total CH4 lengths, the aggregated energy and carbon effects of the re- emissions, nearly 350 megatons (Mt) CH4 yearly [42]. bound travel instances were lower than the amount of energy One of the most important shares of anthropogenic me- and carbon saved by the original efficiency gains. It should thane sources are the leaks from transmission and distribu- be noted, however, that the study did not consider subtler tion networks. Global estimates for the quantities released effects such as possible income effects or time rebounds for from these leaks are difficult to make, but estimates for indi- those conference attendees who would have travelled inter- vidual regions reveal substantial numbers: [42], for example, continentally as well, but given the opportunity to travel estimates leaks of almost 0.5 Mt CH4 yearly for California’s within the same continent saved both money and time. South Coast Air Basin alone. B. A Different Limiting Factor: When Efficiency Gains Have In a collaboration between Google, the Environmental no Market Effect Defense Fund (EDF, an environmental NGO), and research- ers from Colorado State University, a couple of Google street Vending machines are very popular in Japan. So popular, in view cars were prototypically outfitted with methane sensors fact, that in the early 1990s their energy consumption be- for the rapid identification of methane leaks from urban dis- came a political issue: At that time, the 5.4 million vending tribution networks [43]. The algorithm was tweaked using machines were together responsible for 3.7% of the electrici- controlled releases of different flows of methane on an air- ty consumed in Japan [38]. Following energy efficiency field and passes with various speeds at various distances measures, the efficiency of Japanese vending machines im- from these controlled releases so that, in the end, it considers proved by 52% from 1991 to 2007 [39]. for each discovered plume (i.e., an area of elevated CH4) its Given such high efficiency improvements, one would ex- maximum CH4 concentration, the plume extension and an pect a strong rebound effect. Yet, the number of machines index for the plume’s kurtosis. At the same time, plumes increased over this time frame only slightly from 5.4 to 5.5 longer than 160 m are ignored, as they most likely belong to million throughout Japan [40]. Why was there only such a a different methane source nearby, such as dairy farms or mild rebound effect despite the large energy efficiency im- landfills [43]. This prototypical system for leak discovery in provements? The limiting factor for the installation of vend- the urban gas distribution network was deployed in a field ing machines turns out to be space, not energy consumption. experiment in New Jersey, in collaboration with the local As [38] observes: “In a densely populated country like Japan, utility company PSE&G. It has been estimated that through it may be just impossible or unaffordable to sacrifice more the faster discovery and fixing of high-flow leaks, as com- space to install additional machines. It is today possible to pared to traditional methods, this deployment might reduce operate two or three machines with the power that has been yearly CH4 flows into the atmosphere by 2.4 kt [44]. needed for only one machine in 1990s, but it is not possible As natural gas is relatively cheap, the financial effect of to operate them without claiming additional space.” A differ- these savings is rather marginal and hence no rebound effects ent (economic or physical) limiting factor than the energy or are expected [44]. Even if there was a perceivable financial resources undergoing efficiency gains may thus be likely to effect, however, the rebound effect might have been quite lead to only modest rebound effects. low had there been no additional need for heating gas. Alt- hough rising wages and relatively cheaper energy have clear- ly induced a rebound effect in the quantity of heating energy consumed over the centuries (the average winter home tem- tional Energy Agency [52], and even environmental NGOs perature increased in Europe from 13 degrees centigrade in such as the WWF [2, 53]. the 1300s to around 21 degrees today), there is most likely an Many of these and further assessments, in particular upper threshold to the comfort temperature in homes. Gener- those with an industry background, deployed questionable ally, when a market is saturated and there is no additional methods and yielded overly optimistic results. They deliver demand for a product, naturally there will be no direct re- an almost religious promise, which is being heralded by bound effects (although indirect rebound, e.g. income effects, some prominent proponents with much fervor: that digitali- may still occur). zation can be our common savior, the messiah-like technolo- gy that redeems us our environmental sins and which prom- D. Rebound of the Right Sort: Pushing Cleantech Products ises that we can maintain our current lifestyles while digitali- and Circular Economy Processes zation will handle the consequences. One theory of how digitalization affects economic processes One of the main flaws of existing assessments is by and is that energy, time, and information are the main inputs to large their disregard of rebound effects. Digitalization, how- any economic task and can, to some extent, be substituted for ever, pervades nowadays virtually all economic sectors and each other [45]. According to this theory, the digitalization of has become an indispensable part of technological infrastruc- a process allows either time or energy to be saved. The im- ture, not unlike roads or the electrical grid. Thus, it also fos- plicit assumption of this theory is that saving energy is gen- ters efficiency gains throughout the economy. Given its im- erally environmentally beneficial, while saving time (i.e., materiality, its potential for virtualization, and the low entry doing things faster and thus being able to produce more) is barriers for its adoption, it is also a technology phenomenon environmentally harmful. Moreover, as the commercial im- that develops its effects very rapidly (and often without geo- perative is output maximization, [45] establishes that “both, graphic limits). For all these reasons, digitalization seems to IT’s potential to do things with less energy input, thus gener- be particularly prone to the various incarnations of rebound ally more sustainably, and IT’s potential to do things faster, effects. i.e., less sustainably, are enormous. Unfortunately, so far, the The efficiency gains induced by digitalization are not on- latter potential has been extensively tapped while the former ly traditional resource or energy efficiency; above all, it can remains but potential.” save us all time and allow us to connect across continents This dichotomy, however, has recently been challenged. and cultures. The induced secondary effects of the latter, and In [46], it is suggested that not only energy-saving digitaliza- the time rebound of the former, are typical (although not tion, i.e. save impacts, can be environmentally beneficial, but necessarily exclusive) to digitalization, and arguably also some types of economy-accelerating digitalization, amongst the strongest mechanisms leading to rebound ef- which are called push impacts. At the beginning of this sec- fects. Concerning time rebound, [16] writes that it “will be tion, it was argued that not all trips are equal, and that the especially strong when wages are high and, at the same time, type of rebound trips is essential for the environmental out- energy prices are low, as is currently the case in most indus- come of a dual-venue conference. More generally, [46] ar- trialized countries. High wages, which represent the oppor- gues that not all products and economic processes are equal. tunity costs of time, in combination with low energy prices In its view, push impacts operate by accelerating the output encourage the increasing use of time-saving but energy- of products and processes which are beneficial for environ- intensive devices leading to an overall increase in energy use mental sustainability. In particular, these are cleantech prod- as people constantly try to ‘save’ time”. ucts (that substitute less resource-efficient technologies) and Of course, the life cycle of digitalization technologies circular economy processes (i.e., the ones optimizing re- (their production, use, and end-of-life disposal) also encom- source sharing, circulation, and longevity). If digitalization passes an energy and a material footprint. These effects are accelerates such products or processes, they will become much better understood, however, and we refrain from dis- more attractive and will tend to substitute other, more harm- cussing them in detail here since they are already thoroughly ful activities. Acceleration is thus not harmful, per se, just the studied in the literature (e.g., [54-56]). acceleration of the wrong kind of processes and products. VII. CONCLUSION VI. DISCUSSION: DIGITALIZATION AS AN ENVIRONMENTAL SILVER BULLET? Digitalization is unlikely to be the environmental silver bul- let it is sometimes claimed to be. On the contrary, the way Ongoing rapid digitalization is often envisioned as a silver digitalization changes society, making it ever faster, more bullet to tackle – or at least mitigate – the world’s increasing- connected, and allowing us unprecedented levels of efficien- ly urgent environmental issues. In particular, it is seen as a cy might in fact lead to a backfire. As Santarius [57] puts it: possible key factor in reducing carbon emissions and re- “Humanity’s ecological footprint keeps growing although we source consumption across various economic sectors. State- have already digitalized significant parts of our economy and ments to this effect have been put forward by the information society over the past years. It seems that digitalization is not and communication technologies industry itself [1, 47, 48], relaxing but rather reshaping societal metabolism in a way as well as academia [3, 49] and international bodies such as that tends to rebound on global energy and resource demand: the European Commission [50], the OECD [51], the Interna- Gains in efficiency are more than outweighed by the increase in consumption due to new digital services or falling prices on Informatics for Environmental Protection, Berlin, caused by more efficient production processes.” Germany, 2009, pp. 353-361: Shaker Verlag. We cannot, however, agree to the conclusion of [57] that [4] C. D. Harper, C. T. Hendrickson, S. Mangones, and greater efficiency should never be the goal of digitalization, C. Samaras, "Estimating potential increases in travel but its enabling power be used for human sufficiency and with autonomous vehicles for the non-driving, economic degrowth. Above, we presented several conditions elderly and people with travel-restrictive medical that seem to lead to either no rebound or only a moderate conditions," Transportation Research Part C: rebound effect, and they are all related to efficiency, not to Emerging Technologies, vol. 72, pp. 1-9, 2016. sufficiency: i) when the rebound activities inherently have a [5] J. Dokic, B. Müller, and G. Meyer, "European smaller footprint or resource consumption than the originally Roadmap Smart Systems for Automated Driving," optimized activities (such as intra-continental flights com- European Technology Platform on Smart Systems pared to intercontinental flights), ii) when there is a different Integration. 2015, Available: limiting factor (financial or physical) than the one becoming http://www.a3ps.at/site/sites/default/files/newsletter/ more efficient, or iii) when the market is saturated. Addition- 2015/no08/EPOSS.pdf. ally, we mentioned an entire category of desirable rebound [6] J. B. Greenblatt and S. Saxena, "Autonomous taxis effects: the push effects discussed in [46], where the rebound could greatly reduce greenhouse-gas emissions of US of the right (i.e., environmentally beneficial) sort – cleantech light-duty vehicles," Nature Climate Change, vol. 5, or circular economy processes – displaces the wrong kind. pp. 860-863, 2015. For most manifestations of digitalization, however, a [7] A. Brown, J. Gonder, and B. Repac, "An Analysis of strong digital rebound seems to be the rule rather than the Possible Energy Impacts of Automated Vehicles," in exception. The sometimes spectacular per-usage efficiency Road Vehicle Automation, G. Meyer and S. Beiker, gains of digitalization, bearing the toxic gift of strong digital Eds. Cham: Springer International Publishing, 2014, rebound at their very core, hardly alleviate the global issue. pp. 137-153. As discussed in Section III, the mechanisms behind rebound [8] J. M. Anderson, N. Kalra, K. D. Stanley, P. effects in general, and thus of digital rebound as well, are Sorensen, C. Samaras, and T. A. Oluwatola, essentially non-technical in nature. Their roots reside in eco- "Autonomous Vehicle Technology: A Guide for nomics and in human behavior. It is thus highly unlikely that Policymakers," RAND Corporation, Santa Monica, digital rebound can be addressed solely through technologi- CA, US. 2016, Available: cal means. While digitalization does often wait on the side- https://www.rand.org/content/dam/rand/pubs/researc line, ready to provide efficient substitutes for existing tech- h_reports/RR400/RR443-2/RAND_RR443-2.pdf. nologies and processes, the avoidance of digital rebound [9] L. D. Burns, "Sustainable mobility: A vision of our effects needs to be enforced differently, possibly by policy transport future," Nature, vol. 497, pp. 181-182, measures. 2013. More research will hopefully further refine which parts of [10] J. Bischoff and M. Maciejewski, "Simulation of City- digitalization lead to significant rebound, and which digital wide Replacement of Private Cars with Autonomous goods and services induce either only moderate rebound or Taxis in Berlin," Procedia Computer Science, vol. foster environmentally friendly technologies and processes. 83, pp. 237-244, 2016. More research is also needed to understand which are the [11] C. Pakusch, G. Stevens, and P. Bossauer, "Shared policy measures that can foster the latter and impede the Autonomous Vehicles: Potentials for a Sustainable former. Mobility and Risks of Unintended Effects," in ICT4S2018. 5th International Conference on REFERENCES Information and Communication Technology for [1] GeSI. (2008). SMART 2020: Enabling the Low Sustainability, Toronto, Canada, 2018, pp. 258-269. Carbon Economy in the Information Age. Available: [12] D. J. Fagnant, K. M. Kockelman, and P. Bansal, www.theclimategroup.org/what-we- "Operations of Shared Autonomous Vehicle Fleet for do/publications/smart2020-enabling-the-low-carbon- Austin, Texas, Market," Transportation Research economy-in-the-information-age/ Record: Journal of the Transportation Research [2] D. Pamlin and S. Pahlman. (2008). Outline for the Board, vol. 2536, pp. 98-106, 2015. first global IT strategy for CO2 reductions. [13] R. Chase. (2014) Will a World of Driverless Cars Be Available: Heaven or Hell? CityLab. Available: http://assets.panda.org/downloads/global_strategy_fo https://www.citylab.com/transportation/2014/04/will- r_the_1st__billion_tonnes_with_ict__by_wwf.pdf world-driverless-cars-be-heaven-or-hell/8784/ [3] V. Coroama and L. M. Hilty, "Energy Consumed vs. [14] R. Krueger, T. H. Rashidi, and J. M. Rose, Energy Saved by ICT – A Closer Look," in "Preferences for shared autonomous vehicles," EnviroInfo 2009: Environmental Informatics and Transportation Research Part C: Emerging Industrial Environmental Protection: Concepts, Technologies, vol. 69, pp. 343-355, 2016. Methods and Tools, 23rd International Conference [15] S. Sorrell, "Jevons’ Paradox revisited: The evidence [31] C. Fuchs, "The implications of new information and for backfire from improved energy efficiency," communication technologies for sustainability," Energy Policy, vol. 37, no. 4, pp. 1456-1469, 2009. Environment, Development and Sustainability, vol. [16] M. Binswanger, "Technological progress and 10, no. 3, pp. 291-309, 2008. sustainable development: what about the rebound [32] P. L. Mokhtarian, "A Synthetic Approach to effect?," Ecological Economics, vol. 36, no. 1, pp. Estimating the Impacts of Telecommuting on 119-132, 2001. Travel," Urban Studies, vol. 35, no. 2, pp. 215-241, [17] J. D. Khazzoom, "Economic Implications of 1998. Mandated Efficiency in Standards for Household [33] K. W. Roth, T. Rhodes, and R. Ponoum, "The energy Appliances," The Energy Journal, vol. 1, no. 4, pp. and greenhouse gas emission impacts of 21-40, 1980. telecommuting in the U.S.," in 2008 IEEE [18] W. S. Jevons, The Coal Question. London, UK: International Symposium on Electronics and the Macmillan and Co, 1865. Environment, 2008, pp. 1-6. [19] B. Alcott, "Jevons' paradox," Ecological Economics, [34] D. Sivaraman, S. Pacca, K. Mueller, and J. Lin, vol. 54, no. 1, pp. 9-21, 2005. "Comparative Energy, Environmental, and Economic [20] H. D. Saunders, "The Khazzoom-Brookes Postulate Analysis of Traditional and E-commerce DVD and Neoclassical Growth," The Energy Journal, vol. Rental Networks," Journal of Industrial Ecology, 13, no. 4, pp. 131-148, 1992. vol. 11, no. 3, pp. 77-91, 2007. [21] L. Brookes, "Energy efficiency fallacies revisited," [35] H. Siikavirta, M. Punakivi, M. Kärkkäinen, and L. Energy Policy, vol. 28, no. 6–7, pp. 355-366, 2000. Linnanen, "Effects of E-Commerce on Greenhouse [22] L. M. Hilty, Information Technology and Gas Emissions. A Case Study of Grocery Home Sustainability - Essays on the Relationship between Delivery in Finland," Journal of Industrial Ecology, Information Technology and Sustainable vol. 6, no. 2, pp. 83-97, 2002. Development. Norderstedt: Books on Demand, 2008. [36] E. Williams and T. Tagami, "Energy Use in Sales [23] M. Börjesson Rivera, C. Håkansson, Å. Svenfelt, and and Distribution via E-Commerce and Conventional G. Finnveden, "Including second order effects in Retail: A Case Study of the Japanese Book Sector," environmental assessments of ICT," Environmental Journal of Industrial Ecology, vol. 6, no. 2, pp. 99- Modelling & Software, vol. 56, pp. 105-115, 2014. 114, 2002. [24] P. H. G. Berkhout, J. C. Muskens, and J. W. [37] V. C. Coroama, L. M. Hilty, and M. Birtel, "Effects Velthuijsen, "Defining the rebound effect," Energy of Internet-Based Multiple-Site Conferences on Policy, vol. 28, no. 6–7, pp. 425-432, 2000. Greenhouse Gas Emissions," Telematics & [25] H. R. Varian, Microeconomic Analysis, 3rd ed. Informatics, vol. 29, no. 4, pp. 362-374, 2012. Mason, Ohio: Thomson Higher Education, 1992. [38] L. M. Hilty, "Why energy efficiency is not sufficient [26] L. Erdmann and L. M. Hilty, "Scenario Analysis: – some remarks on “Green by IT”," in Proceedings Exploring the Macroeconomic Impacts of of the 26th Environmental Informatics Conference Information and Communication Technologies on (EnviroInfo), Dessau Germany, 2012, pp. 13-20. Greenhouse Gas Emissions," Journal of Industrial [39] T. Yokouchi, "Today and tomorrow of vending Ecology, vol. 14, no. 5, pp. 826-843, 2010. machine and its services in Japan," in 7th [27] H. S. Matthews and E. Williams, "Telework International Conference on Service Systems and Adoption and Energy Use in Building and Transport Service Management, 2010, pp. 1-5. Sectors in the United States and Japan," Journal of [40] M. Takashi and H. Asano, "Japanese Vending Infrastructure Systems, vol. 11, no. 1, pp. 21-30, Machine and Display Cooler Energy Use Affected by 2005. Principal-Agent Problem," in "Quantifying the [28] R. Hamer, E. Kroes, and H. Van Ooststroom, Effects of Market Failures in the End-Use of "Teleworking in the Netherlands: an evaluation of Energy," International Energy Agency 2006, changes in travel behaviour," Transportation, vol. Available: 18, no. 4, pp. 365-382, 1991. http://s3.amazonaws.com/zanran_storage/www.aceee [29] R. M. Pendyala, K. G. Goulias, and R. Kitamura, .org/ContentPages/4790329.pdf. "Impact of telecommuting on spatial and temporal [41] U. S. E. I. Administration. (2008). About U.S. patterns of household travel," Transportation, vol. Natural Gas Pipelines. Available: 18, no. 4, pp. 383-409, 1991. https://www.eia.gov/naturalgas/archive/analysis_publ [30] B. Koenig, D. Henderson, and P. Mohktarian, "The ications/ngpipeline/index.html Travel and Emissions Impacts of Telecommuting for [42] D. Wunch et al., "Quantifying the loss of processed the State of California Telecommuting Pilot Project," natural gas within California's South Coast Air Basin Transportation Research Part C: Emerging using long-term measurements of ethane and Technologies, vol. 4, no. 1, pp. 13-32, 1996. methane," Atmos. Chem. Phys., vol. 16, no. 22, pp. 14091-14105, 2016. [43] J. C. von Fischer et al., "Rapid, Vehicle-Based from the Commission to the European Parliament, Identification of Location and Magnitude of Urban the Council, the European Economic and Social Natural Gas Pipeline Leaks," Environmental Science Committee and the Committee of the Regions, & Technology, vol. 51, no. 7, pp. 4091-4099, 2017. Brussels 2008. [44] V. C. Coroama and M. Höjer, "Assessing GHG [51] OECD. (2009). Measuring the Relationship between Benefits Induced by ICT Services in Practice: A Case ICT and the Environment. Available: Study and Resulting Challenges," in 4th http://www.oecd.org/sti/43539507.pdf International Conference on ICT for Sustainability [52] IEA, "Digitalization & Energy " 2017, Available: (ICT4S 2016), Amsterdam, Nederlands, 2016, pp. http://www.iea.org/digital/. 29-35: Atlantis Press. [53] D. Pamlin and K. Szomolányi, "Saving the Climate [45] D. Spreng, "Does IT have boundless influence on @ the Speed of Light," WWF Sweden 2006, energy consumption? ," in EnviroInfo 2001: Available: Sustainability in the Information Society, Zurich, http://assets.panda.org/downloads/road_map_speed_ Switzerland, 2001, pp. 81-90: Metropolis Verlag. of_light_wwf_etno.pdf. [46] J. Townsend and V. C. Coroama, "Digital [54] W. Van Heddeghem, S. Lambert, B. Lannoo, D. Acceleration of Sustainability Transition: The Colle, M. Pickavet, and P. Demeester, "Trends in Paradox of Push Impacts," Sustainability, vol. 10, no. worldwide ICT electricity consumption from 2007 to 8, 2018. 2012," Computer Communications, vol. 50, pp. 64- [47] GeSI and Accenture Strategy. (2015). 76, 2014. #SMARTer2030 – ICT Solutions for 21st Century [55] J. Malmodin, Å. Moberg, D. Lundén, G. Finnveden, Challenges. Available: http://smarter2030.gesi.org/ and N. Lövehagen, "Greenhouse Gas Emissions and [48] British Telecom. (2014). Net Good: Pathway to Operational Electricity Use in the ICT and Carbon Net Positive. Available: Entertainment & Media Sectors," Journal of https://www.btplc.com/Sharesandperformance/Prese Industrial Ecology, vol. 14, no. 5, pp. 770-790, 2010. ntations/downloads/SRInetgoodlaunch.pdf [56] R. Hischier, V. C. Coroama, D. Schien, and M. A. [49] L. M. Hilty, P. Arnfalk, L. Erdmann, J. Goodman, M. Achachlouei, "Grey Energy and Environmental Lehmann, and P. A. Wäger, "The relevance of Impacts of ICT Hardware," in ICT Innovations for information and communication technologies for Sustainability, L. M. Hilty and B. Aebischer, Eds.: environmental sustainability - A prospective Springer International Publishing, 2015, pp. 171-189. simulation study," Environmental Modelling & [57] T. Santarius. (2017). Digitalization, Efficiency and Software, vol. 21, no. 11, pp. 1618-1629, 2006. the Rebound Effect. Available: [50] European Commission, "Addressing the Challenge of https://www.degrowth.info/en/2017/02/digitalization- Energy Efficiency through Information and efficiency-and-the-rebound-effect/ Communication Technologies," Communication