How to reconcile ecological and digital transition?
(published on the Institut Montaigne website in May 2021)
In recent months, greenhouse gas emissions from digital technologies have taken an important place in the media, in particular due to the publication of reports with explosive conclusions. Thus, the reports of the Shift Project and the High Council for the Climate both mentioned a significant increase in the environmental externalities of digital technology. While these figures have been the subject of significant controversy (see below), they have had the merit of showing how much this subject remains largely unknown.
Despite this, the legislator wanted to take up the issue without delay. Thus, the Senate adopted on January 12 a cross-partisan bill aimed at “reducing the environmental footprint of digital technology”. This revolves around four main ideas.
First, make users aware of the environmental impact of digital technology: inform, raise awareness and empower users, individuals and businesses with good practices that avoid waste or disproportionate use of energy associated with digital services. This generally introduces the principle of numerical sobriety.
Second, limit the renewal of digital terminals, the manufacture of which is the main responsible for the carbon footprint of digital in France. The bill aims in particular to penalize software obsolescence, to improve the fight against planned obsolescence and to support reconditioning and repair activities by lowering the VAT rate to 5.5%.
Third, promote ecologically virtuous digital uses, in particular by making the eco-design of websites compulsory and providing for the establishment of a general eco-design reference system.
Finally, fourthly, bring about environmental regulation to prevent the increase in consumption and emissions from networks and data centers.
These proposals follow a Senate report trying to assess the environmental impacts of the entire “digital sector” value chain, from terminals to data centers to networks.
We can obviously only praise such an approach which, in the present and future world, in which the environmental constraint must become a primary requirement, only sheds light on a sector which has not made this issue a priority. (probably because he perceived himself as sufficiently at the forefront of modernity to free himself from a critical look at its externalities).
Measuring the environmental impact of digital technology: a considerable challenge
The fact remains that measuring the environmental impact of digital technology is very complex. The rebound effect, the oft-cited theory that technological improvements increase the usability of a service, and therefore its consumption, is powerful but very difficult to measure, for several reasons.
Let us first note that, in about twenty years, the cost of administering data has been divided by a factor of around 70,000: a processing and storage system that cost one million euros in 1995 was only worth about fifteen euros in 2015. This principle applies to the performance of calculation, storage and transport (due to Moore's prediction) and is also found, with various scales, to energy efficiency. Thus, a computer of the nineties such as the Macintosh II consumed 230 watts and its screen 205 watts, making a total of 435 watts. It is incomparably less powerful (150,000 times fewer transistors) than a 2016 Samsung S8 smartphone which consumes between 8 and 12 watts when in use.
The digital universe is an area where technological breakthroughs are constant. Thus, the most "powerful" processor on the market - the Nvidia A100 tensor, comprising 52 billion transistors - sees its supremacy challenged by a new type of processor, no longer using electrons but photons, developed by company Lightmatter, which offers 1.5 to 10 times better performance for 6 times less energy consumption. This type of breakthrough innovation is more common than is generally imagined and it also applies to technological architectures. Thus, a recent data center using so-called “adiabatic” cooling technologies can consume 40% less energy than its equivalent using traditional digital technologies.
As a result, calculations that will have been correct at a given time are quickly distorted due to the accelerated obsolescence of technologies. Many analyzes made about these technologies do not sufficiently take into account this characteristic of improvement and thus project a model on more or less constant efficiency bases, even as they grow considerably, constantly reducing the energy required to perform the same task.
Moreover, the origin of the energy is an essential factor and not always well understood. Thus, the CO2 intensity of Chinese electricity is around nine to eleven times higher than ours and that of the USA (417 gr/Kwh) seven times higher than that of France. If equipment is manufactured in China, its carbon intensity will therefore be largely linked to this intensity specific to Chinese electricity. This footprint would be significantly lower if this same equipment were manufactured in France (which is only rarely the case). On a different scale, the footprint of a Netflix user will vary depending on whether he watches his film from a stream from a French or Norwegian hosting (the latter at 50 gr/KWh, against 65 gr/KWh in France) 'from an American stream (Note that this type of company uses cache technologies (CDN) which store the most requested content as close as possible to the user. Thus, in France, Netflix has directly or indirectly several datacenters to cover the demand of French users, limiting the distance traveled by the flows to a few hundred km). The fact remains that, the world being what it is today, at least 70% of technological equipment is made in China, while it is estimated that just over 50% of data is located in the USA. These concepts are likely to evolve, in particular due to the progressive consideration of environmental issues.
Finally, it should be noted that equipment depreciation strategies have a major impact on the energy consumption of the digital industry. A leasing company will seek to slightly increase its offer and, in return, to offer an accelerated renewal of the equipment it provides. Similarly, a telecom operator can decide to stop subsidizing terminals and optimize the replacement of its equipment by taking into account carbon externalities, which few are doing to date. Insofar as the carbon footprint linked to manufacturing represents between 75% and 95% (the latter figure being relative to certain passive equipment, including optical equipment for example) of the total footprint (including the manufacturing and use ), we can see how much we would gain by considering such approaches.
These points highlight the intrinsic complexity of these issues in the digital sector. They cover particularly sophisticated technological, scientific and economic dimensions, which are often sources of methodological errors. If the complexity of the subject is important, it is however not inaccessible. To give credit to recent studies, some of which announce worrying figures with regard to digital externalities, it is essential to facilitate the emergence of an activity that does not develop to the detriment of the planet. For this, a better understanding of the transformation of production models is essential.
In general, the very strong individual development of digital technologies (more and more of us are connected and increasingly multi-connected) has the structural consequence of increasing the direct environmental footprint of digital technology (purchase of terminals, development of digital infrastructures, and to a lesser extent, use). Nevertheless, the main issue, which has been fundamentally evaded by recent work, concerns the positive externalities of digital technology; a point that is also the subject of a weakness in method because, no doubt to make their work more spectacular, many of the authors of these works did not hesitate to highlight the most disastrous uses of digital technology. The example of streaming, cited above, is striking. The authors of the first Shift Project report claimed that it represented 1% of total CO2 emissions; a figure that the association will later admit to be false, with more recent work suggesting a footprint of 22 to 57 lower.
Positive externalities are numerous and often overlooked. Thus the filling rate of trucks within the European Union, which would have increased by 14% in the space of fifteen years due to the development of information systems integrated into the logistics chains according to the European Commission. If many nations are seeing a significant drop in consumption per kilometer traveled on their roads, the importance of connected GPS, which makes it possible to avoid traffic jams, probably has a lot to do with it. In the same logic, if the government has planned a device to finance connected heaters (which go on standby when no one is present),
it is because their effectiveness in relation to their cost is unmatched, and so on. We can also cite the case of teleworking in the context of the health crisis, which has considerably reduced the use of energy-intensive means of transport.
In reality, in a finite world, where the exploration of resources generates significant negative externalities, making information widely accessible and usable is one of the most effective ways of reducing the environmental footprint: this notably makes it possible to synchronize at best the needs with the offer, the flows with the infrastructures, and this at all levels of the production chains. Information technologies therefore arrive at the right time when the objective is no longer so much to produce more, but to produce better and make better use of this production.
These dimensions are not only misunderstood, but also insufficiently studied. Thus, the agricultural exploitation of legumes easily highlights more than a hundred variables, some of which are not controllable (temperature, humidity). However, optimizing these variables can have very significant environmental consequences on productivity, quantitatively and qualitatively, but also on greenhouse gas (GHG) emissions and other environmental externalities. In any event, learning machines could quickly become powerful auxiliaries for managing these multivariate environments and optimizing the environmental requirements of these activities.
The digital revolution is instilling a paradigmatic break of a magnitude at least equivalent to those that ushered us into the modern era. This revolution is both anthropological - it alters our relationship to space, the nature of our social interactions, our psyche... -, economic - observing the downgrading of the actors who dominated the 20th century by digital actors allows us to avoid a long debate - and geostrategic - the pre-eminence of the States is not so clear any more, the threats change shape, just like their initiators…. It therefore pushes us to consider the challenges facing us in a different way. The objective is not to position oneself on an ideological basis, but to grasp the nature of this revolution in order to make it serve the greatest number and, moreover, enable it to become an auxiliary of the environmental revolution which, no doubt, represents the greatest challenge of the 21st century.
It is likely that one of the objectives to be achieved is to transform the digital industry, no longer at the service of an often stultifying consumerism, but to put it at the service of this challenge of the century which is the environment. This means designing more virtuous systems, equipment and software, and creating the skills needed to cross the different economic, environmental and digital sectors. The double challenge represented by the acceleration of the digital revolution and the imminent threat linked to climate issues strongly invites us to do so.