A Digital Burden: on ICT's carbon footprint and the urgency of salience
Our attitude toward digital technologies, let alone the language we use to speak about them, is both cause and consequence of a shared perception, whereby we tend to conceptualise their use as utterly disconnected from its material sources. In stark opposition with this tendency, a number of studies have recently emerged, which are aimed at highlighting the worryingly tangible aspects of these technologies. Some of those captured the attention of the general media and public opinion. Hence, we gained a basic understanding of a series of issues linked to the extraction, assembly, and recycling of the so-called hardware components of the digital realm. However, there remains a near absolute lack of awareness regarding digital technologies’ substantial share in the worldwide emission of carbon dioxide and carbon dioxide equivalents. As compared to the magnitude of such a share, and to the ubiquitous adoption of the digital paradigm, this very unawareness appears to be as much unjustified as disproportioned (at best).
Setting the focus
Surprisingly enough, the first reliable figures as to Information and Communication Technologies (ICT) sector’s impact on air pollution emerged only in 2007, as part of a now cornerstone report by the consultant agency Gartner. Back then, Gartner estimated that the ICT accounted for 2% of the anthropogenic CO2 emissions on a global scale[1]. Remarkably, that was almost identical to the amount that a notoriously polluting sector, the aviation industry, produced during the course of the same year. In the decade up to 2020, aviation’s percentage has not substantially changed, while ICT’s growing negative contribution has been increasingly debated.
However, to be sure, a fundamental difference between the two sectors needs to be promptly acknowledged: ICT’s percentage of emissions directly caused by combustion is virtually negligible, since the near totality of them is due to the electricity consumption involved in both stages of production (45%) and use (55%) of the digital apparatus[2]. These two stages, despite being almost equivalent in size, are thoroughly dissimilar in terms of composition and, most importantly, cognitive perception. Indeed, the pollutive character of the production stage appears to be more intelligible for individual users, insofar as digital devices are, de facto, consumer objects, understandably subject to the traditional manufacturing lifecycle. Of course, they often tend to ignore the existence of those employed at the infrastructural level, yet their consumption during the production stage is limited (6%), as compared to that of the so-called user end devices, instead amounting to 39% of the total consumption of the ICT[3].
Conversely, as far as the use stage is concerned, things look quite different. Not only do infrastructures such as networks and data centres come to the forefront (16% and 19%, respectively[4]), but here the source of consumption becomes even foggier. Indeed, less tech-savvy users generally assume that their computers, smartphones, or smart TVs, are easily comparable to regular home appliances. In other words, they perceive electricity consumption as a merely local process, exclusively performed via batteries, chargers, and power supplies. But it definitely isn’t. Rather, it is a thoroughly distributed one, which is triggered by every single interaction occurring between digital devices and what we commonly refer to as the internet.
In order to better grasp the complexity of these issues, we will thereby focus on this second stage, the use stage, in that it’s the one for which didactic and informational initiatives seem most urgently needed, especially in the light of a series of transformations that will drastically affect the ICT within the next few years.
A few complications
Providing a unilateral account of the polluting impact of the ICT is a complicated task. Differences in method between the existing studies are considerable, and so are fluctuations among their results. On the one hand, this is due to the very physiology of what still is a young sector, which evolves according to a set of mostly novel historical trends. On the other hand, this is to ascribe to the heterogeneous, ever-changing nature of the technologies involved, whereby every systemic model is inevitably associated to a substantial degree of approximation.
Take, for example, two of the most discordant among the existing researches, which prove useful since virtually concurrent, and mainly based on a similar, top-down approach. Ericsson researchers Jens Malmodin and Daniel Ludén paint a fundamentally positive scenario: focusing, in retrospect, on the interval between 2010 and 2015, they compute the ICT being responsible for 1.4% of the global emissions, also reporting a slow, progressive contraction[5]. By contrast, another scenario, provided by the think tank The Shift Project is decisively less optimistic: not only was ICT’s 2015 carbon footprint considerably bigger (namely around 3% globally), but it has also been rapidly growing by 9% a year ever since[6].
Of course, part of this discrepancy originates in different frames of reference—Malmodin and Lundén, for instance, don’t include smart TVs in their model. However, the main reason is that, within the considered span, these studies report decisively dissimilar figures as to the total amount of electricity consumed by the ICT.
It is, indeed, quite complicated to build a valid horizontal model for such category of consumption, since results may considerably differ depending on the selected variables, which are often too granular and geographically specific to have an accurate global relevance. In addition, this intricacy is likely to further worsen in the near future, as the ICT sector is approaching a historical turning point, where a series of unprecedented trends are about to cross its path. It is still unknown, for instance, to what extent the expected energy efficiency gains—which have long been considered inherent to digital hardware—will be affected by the end of Moore’s law, or by the slowing down of Koomey’s law, which governs the joule/bit ratio of networks and data centres[7].
Moreover, we cannot predict the size of the so-called rebound effects impacting these very gains, according to which improvements in productivity might be prone to be paralleled by increased consumption. However, the greatest degree of uncertainty is that surrounding a series of emergent technologies that are likely to be proliferating worldwide in the next few years. Internet of things (IoT), 5G, advanced AI, and blockchain, to name a few. Sure, we already know the current pollutive output of some of them. We know, for instance, that BitCoin networks (supported by the blockchain) produce as much as 22.9 Megatons of CO2 annually[8], or that training one single AI model can emit as much carbon dioxide as running five cars through their whole lifetime[9]. And yet it seems impossible to produce a clear prediction as to the aggregated future effects of these technologies, since that will strongly depend on their range of implementation, as well as on the nature of the new behavioural patterns they will inevitably spawn.
Source: Cisco VNI Global IP Traffic Forecast, 2017–2022
It is precisely with respect to the latter that education and information become critical and timely matters, which need to overlook academic discrepancies and focus on the already available tools. A set of parameters is needed, which can be transversally accepted and yet easily assimilated by consumers on an accessible, everyday scale.
What follows, identifies the so-called data traffic as the most well-suited of such parameters, in that it is accurately measurable, universally cited in studies concerning ICT, and proportionally linked to both usage and electric consumption.
Data Traffic
As previously stated, 2% was the global carbon footprint attributed to ICT in the year 2007. However, worryingly enough, this was happening when only 25% of the world population had access to the internet. Conversely, it is now estimated that internet users have nearly doubled ever since, and that, by 2025, they will make up more than 75% of the world population[10]. This doesn’t only have inevitable repercussions on the number, typology, and distribution of the connected devices. As a matter of fact, this trend is also paralleled by a more intense usage of each individual device, and, accordingly, by an exponential increase in the volume of digital data required for running them. Consider, for instance, the skyrocketing growth of what the market analytics firm IDC calls Global Datasphere—namely the measure of all new data that is captured, created, and replicated in any given year across the globe[11]. This measure, which has already risen exponentially from 0,2 zettabytes (2010) to 33 zettabytes (2019), is bewilderingly expected to be 175 zettabytes in 2025[12]. Given that 175 zettabytes, in slightly more prosaic terms, amount to approximately 1400 sextillions Bits, and provided that Bits are, essentially, electrical units of information, such a massive expansion of the Datasphere should resonate as inherently sobering.
Source: Data Age 2025, sponsored by Seagate with data from IDC Global DataSphere, Nov 2018
However, since most of the Dataspehere won’t be continuously stored locally, this gigantic growth is only one actor in the broader—and more alarming—phenomenon of the so-called data traffic explosion. Indeed, the explosion of data traffic (namely that of the digital amount of data that cross all internet networks) is an index of the ever-growing trend, whereby electric consumption is pushed away from devices, towards networks and data centres, where the cost of such consumption is concealed from consumers[13].
Thus, an ever-increasing number of bytes are flowing across the infrastructures of the digital realm, and this isn’t due only to an exponential surge in the amount of data created. But also to the unprecedented volume of data that is distributed as the by-product of devices that require a continuously interconnected, non-local exchange of information. The most typical example is that of the activities related to the online video: a great deal of the computation required—and, therefore, of energy consumption—does not happen in the very room where the video is being watched. Rather, it is remotely sustained within the data centres where the information of the selected content actually resides. Consequently, a far from negligible amount of data needs to cross the networks in order to reach the consumer devices where it is finally decoded. Hence, we could easily state that watching a movie on platforms such as Netflix or Apple TV generates an amount of data traffic that, were the same activity undertaken locally (via files or DVD), simply wouldn’t exist.
Accordingly, it is no accident that most of the studies focusing on ICT’s energy consumption take data traffic as an essential variable, thereby having to deal with a set of negative predictions. According to those of CISCO, for instance, annual data traffic is growing by 26% CAGR[14], as it measured only 1.5 Zettabytes in 2017, but will amount to 4.8 Zettabytes in 2022. That means that 2022 data traffic is set to equal that of all years since the very inception of the internet[15]. Furthermore, these scenarios don’t improve even when observed at the granular level. If, in 2017, the average individual user generated 36 Gigabytes of monthly traffic, in 2022 they will generate an average of 98. Moreover, this will be happening in the same year when also data traffic per second will be, by far, the highest on record (150,700 Gigabytes)[16].
To summarise, not only is the nature of our interconnected activities becoming significantly more distributed, but also increasingly more frequent and concentrated. And this is clearly a consequence of the systemic integration of digital technologies into our everyday life. In order to get an approximate estimation of what this means for the environment, a series of calculations have combined nation-wide figures about annual Green House Gasses (GHG) emissions and power consumption of few western countries. It was possible to find, for instance, that one standard e-mail may release 50 g of CO2 into the atmosphere[17], or that transferring 1 Gigabyte file emits as much as 3 Kg of CO2[18]. Of course, these are generalised approximations, based on figures that are too exclusive and geographically specific to be fully accurate. But they also add up to a large number of existing studies in establishing a simple, working proposition: in principle, the bigger the flow of data traffic crossing the networks, the greater the total power consumption[19]—and, accordingly, the amount of GHG emitted in the atmosphere.
Source: Cisco VNI Global IP Traffic Forecast, 2017–2022
In conclusion, given that data traffic is, essentially, the primary by-product of any digitally interconnected activity, it could be fruitfully adopted as the privileged parameter in promoting a more intuitive understanding of our own, individual degree of digital consumption. In other words, in the same fashion as we learnt to associate automotive pollution to litres of fuel consumed by petrol engines, we could gradually learn to measure the environmental impact of our online behaviour precisely by dint of data traffic. Clearly, such an operation wouldn’t claim to emancipate users from the correctness of coarse-grained scientific research, but it would commit to the role of patching the wide gap separating digital activities from the material burden of the processes they trigger.
The consumer side
In an intriguing experiment conducted by South California Edison in 2007 (coincidentally the same year as Gartner’s report), a curiously minimalistic device, the Ambient Orb, was shipped by the company to a sample of 120 customers, in the attempt to reduce their electricity waste during peak hours. The Orb would simply turn green when electricity was cheaper and underused, while turning red when electricity was overused and more expensive. To the surprise of Mark Martinez, the manager who designed the experiment and had already approached customers with a number of more direct methods, in a few weeks the Orb reduced peak consumption by 40%[20].
This isolated tryout should bear the appearance of a cautionary tale, since larger-scale scientific studies have proven that similar tools, such as smart metres and eco-visualisation devices, typically achieves up to a 5-15% reduction[21]. However, its underlying assumption might be repurposed in the context of our analysis of the behavioural patterns associated with digital media. This is due to the fact that the Orb, in its transposition of electrical data, successfully managed to visualise something which, by nature, is utterly invisible to the eye of its user. And that is a useful lesson that we need to bear in mind as we approach products, namely user end digital devices, that are perceived as increasingly unhinged from the material inputs they are built upon.
To be sure, historically, it wasn’t the so-called digital revolution to spawn this drift towards immateriality. And yet it wouldn’t be incorrect to state that digital technologies do epitomise the latest stage in such a still escalating trend. Indeed, perceived abstraction is pursued as a constitutive target by all actors in the high-tech industries, according to what technologist Adam Greenfield calls Ideology of Ease[22]—namely the insistence that all tasks be made as simple as possible, at all times[23]. As a matter of fact, in many cases, ethereal simplicity even becomes a stubborn mantra, promoted with attentive marketing, and aimed at replacing a given analogue process with its digital counterpart. Consider, for instance, some of the benefits the so-called Internet of Things is meant to introduce, such as those involving digitally managed home appliances. Greenfield:
Are the constraints presented to us by life in the non-connected world really so onerous? Is it really so difficult to wait until you get home to pre-heat the oven? And is it worth trading away so much, just to be able to do so remotely?[24]
To paraphrase this passage, are the results of purely physical actions really so different from the ones of their digitally operated equivalents?
Pondering the pros and cons of a fully interconnected environment surely becomes an individual task. Still, what if, due to a legitimate lack of knowledge, important factors such as power consumption and air pollution are altogether excluded from users’ decisional process? This is a critical question to address with precise timing, considering a context in which governments are collectively on tracks to pursue the implementation of fibre optics and 5G, and companies are preparing for the systemic spread of the digital ecosystem dubbed as IoT.
In the light of the expectable impact that such enhancements will have on consumption and data traffic, a widespread educational programme seems urgently required, before that the behavioural patterns associated with these technologies are perceived as indispensable needs. And its focus is to be sharpened on all of us, everyday digital consumers, who, according to CISCO, are responsible for roughly 83% of global data traffic[25].
Source: Cisco VNI Global IP Traffic Forecast, 2017–2022
Of course, similar bottom-up approaches have proven only mildly successful in other productive sectors, since their main result has been that of further shifting the burden of sustainability away from companies and policymakers[26]. However, this seems to be the case only if this very approach is implemented a posteriori, namely if it focuses on amending a set of habits that are already firmly crystallised and, accordingly, subject to principles that behavioural scientists call loss aversion and status quo bias. A constructive goal, instead, would be that of addressing users before that such behavioural patterns become widely established in the first place. Here, the Orb lesson might become instructive, since explicit knowledge and information aren’t enough to divert such biases; what would really affect decision making is a heightened salience of the aforementioned issues, namely an improved, intuitive understating of their implications on an everyday, functional level.
Within this backdrop, we believe that a successful initiative would be one that operates at the intersection of art and science, one where the cross-pollination of quantitative and qualitative methodologies can lead to a more conscious user experience, one whereby detailed analytic tools such as aggregate consumption and data traffic can be translated into powerful metaphors. Ideally, in a non-paternalistic fashion, users would thereby be able to produce their own attentive distinction as to which technologies are pragmatically essential and which technologies might be, conversely, trivially dispensable.
In conclusion, two factors lead to the necessity of such an initiative. Firstly, unawareness surrounding these topics is massively distributed across the public opinion. Secondly, even existing policies and frameworks, such as those provided by the EU digital literacy programme, are exclusively dedicated to teaching specific practical skills, and thereby essentially neglect both ethical and environmental concerns.
The proposed task is certainly an arduous one, and its implementation will necessarily require continuous and multilateral efforts. But consensus as to its urgency is gaining momentum also outside the academic circuits, together with a growing general interest as to the status quo of the undergoing digital transformations. And that allows us envisage new opportunities for tangible, positive outcomes.
For a full reference list please click here.