Google’s new environmental report for 2024 is out and it makes some remarkable disclosures. Both The Register (Mann 2024) and The Guardian (Milmo 2024) have reported on the 48 percent rise in Google’s carbon output between 2019 and 2023. Other data from the report show the company’s GHG emissions “increased 13% year-over-year, partially driven by a 37% year-over-year increase in Scope 2 (market-based) emissions” (Google 2024, 31). That 37% increase happened, “despite considerable efforts and progress on carbon free energy (CFE). This was due to data center electricity consumption outpacing our ability to bring more CFE projects online” (Google 2024, 33). What Google is saying, in effect, is that when it comes to the global heating impacts of the energy consumption required by its infrastructure the company is subject to the Jevons Paradox (also called the rebound effect).Dynamic Earth – Ocean Currents by NASA Goddard Photo and Video is licensed under CC-BY 2.0

William Stanley Jevons was a 19th-century political economist who, among other things, studied the relationship between efficiency and energy consumption. Jevons investigated how coal consumption changed as industrial machines became more efficient–using less and less coal per unit of work or product produced. Expecting improvements in this kind of efficiency to lead to reduced coal use, Jevons found the opposite occurred: overall use of coal went up. Why the seemingly strange outcome? If a machine can use less coal to do the same–or even more–work as a consequence of efficiency improvements, then the supply of coal increases relative to demand for its use. That makes it cheaper to use coal. Once it’s cheaper, users can afford to use more and those potential users who couldn’t afford to use any coal before now can. So aggregate demand for coal went up even as machines got more and more efficient using it.

Jevons paradox is intimately related with conversations about decoupling. In the ICT world and beyond there’s a lot of research on decoupling–basically, the delinking of material and energy throughput from economic growth. Jevons paradox and the rebound effect both come into the decoupling debates. What Google’s new report shows very clearly is that at least in the case of this company Jevons paradox is unfolding as it would predict. Despite its aspirations, Google’s growth is not decoupling from the energy requirements of its infrastructure. Even as the company makes more and more efficiency gains, those gains are wiped out by aggregate demand for the company’s services. And, not just by little bits, but by very significant percentage points e.g. 48% rise in Google’s carbon footprint between 2019 and 2023.

Google is just one company, but of course it’s not just any old online service. Google is one of the largest companies in the world when ranked by revenue. What is true for Google in terms of increasing carbon footprint may not be true of all internet companies, but the sheer magnitude of Google’s operations make it a potentially good indicator of broader trends.

Just a few years ago a consortium of researchers at Northwestern University, UC Santa Barbara, and Lawrence Berkeley National Laboratory published a paper in the policy forum Science re-examining global data centre energy use estimates (Masanet et al. 2020). In short, the authors argue, estimates that global data centre energy use will triple or even quadruple within the next decade are based on overly simplistic assumptions. The data of these authors analyze, “suggest more modest growth in global data center energy use” (Masanet et al. 2020, 1). The authors go on to claim that, “although global data center energy has increased slightly since 2010, growth in energy use has been substantially decoupled from growth in data center compute instances over the same time period” (Masanet et al. 2020, 2). There’s a couple of things to notice here I think. One is the hedge by the authors when they use the word “substantially”. Substantial decoupling is not total decoupling or, said differently, slow growth is not no growth. A second thing to notice is the publication date of the article: 2020.

Mansanet et al.,’s paper is a policy forum piece, so while it cites references, it doesn’t have a separate methods section. The latter situation makes it a bit murky to figure out the date of data used by the authors, but it seems as though their results are based on a data set up to 2018. Just two years after this paper was published ChatGPT was publicly released.

The LLM/AI spring/summer/hype-bubble is in full swing this year. I’m not trying to attribute all of Google’s increased carbon footprint since 2019 to AI, nor am I suggesting that Masanet et al., (2020) are wrong in their claims about modest growth in data centre energy use increase. But again, modest growth is not no growth and ChatGPT was publicly released after the publication of Masanet et al.‘s paper. There’s a telling line in the paper about real energy efficiency improvements that have occurred since the earliest global estimates of energy demand for data centres. Those improvements, “can bring about a near-term plateau in energy use, which provides critical time to prepare for the possibility of future energy demand growth. But this time must be used wisely.” (Masanet et al. 2020, 3). Indeed. Perhaps the plateau has already past and we’re back to the upward trend? Predictions are always tricky.

Authors of a recent perspective for Nature argue that, “the digital carbon footprint is inherently unpredictable” (Gritsenko, Aaen, and Flyvbjerg 2024, 1). Three assumptions underlying the prediction literature do not hold up according to these authors. They argue that the digital carbon footprint cannot be meaningfully quantified. They also suggest that it is a mistake that future trends will follow a business as usual path with merely episodic followed by new plateaus of stability. Finally, the authors argue that digitization projects frequently do not deliver the benefits described in their business cases within the cost or time frames they promise.

Quantifying the carbon footprint of digital technologies is highly uncertain, even indeterminate. Quantifying this kind of carbon footprint is a fat tail problem – – the problem space occurs within a probability distribution where extreme events are more likely to occur than in a normal distribution. With the carbon footprint of digtialization ‘fat-tailedness’ happens because there are both direct and indirect effects of these technologies AND because those direct and indirect effects can vary within wide margins of uncertainty AND change depending on base-line assumptions made.

Direct carbon footprint effects of digitalization would include the material and energy consequences of mining, manufacturing, use, and discard (the latter would include anything from recycling and disposal, to incineration etc.). Indirect effects, typically cover a whole range of climate effects some of them positive and some of them negative. A positive effect might be the substitution of online shopping for travel to the mall via private automobile. A negative effect might be increased consumption arising from the convenience of online shopping.

Further uncertainty is induced by the inherently non-arbitrary decisions about where to draw system boundaries for estimating carbon impacts of digitalization. The further upstream along supply chains and toward resource extraction such an analysis goes, the more indeterminate the boundaries get – – the global electronics sector is the second largest consumer of copper after the construction industry, but attributing copper flows from any given mine to any given finished electronic product or digital system is challenging to say the least. But, as Gritsenko et al. write, “[w]hen supply chain pathways are chopped off, the climate impact becomes underestimated. Thus, estimates of digitalization carbon footprint are prone to truncation error, that is, partial exclusion of infrastructure by the traditional process of life-cycle assessment” (Gritsenko, Aaen, and Flyvbjerg 2024, 2). Ultimately, these carbon impact assessments are “inconsistent predictions” and “provide little guidance to decision-makers as to what extent—and under which conditions—digitalization can alleviate the climate crisis” (Gritsenko, Aaen, and Flyvbjerg 2024, 2).

Drawing on a broader literature of radical uncertainty, Gritsenko et al argue that one good way to get beyond the false promises of quantification in these contexts is to tell stories (or “narratives”). Doing so can mean different things, but among them are coming up with analogies from already existing systems to guide decision-making. Gritsenko et al offer the example of a tsunami wall as something that real places have actually built to mitigate (very!) negative consequences of inherently unpredictable events of magnitudes that cannot be known in advance. With respect to digitalization, the authors suggest investments in renewable energy infrastructure as an analogous protective wall against the negative climate affects of digitalization.

Gritsenko et al.’s intervention is important for drawing critical attention to the whole project of attempting to quantify the climate footprint of digitalization. The authors are doing so not just for the sake of it but because, “we can avoid wasting time and resources trying to gather and estimate quantified information that is not available (or even falling into the false security of invented numbers) and instead focus on much-needed actions to mitigate rather than predict the climate impact of digitalization.” (Gritsenko, Aaen, and Flyvbjerg 2024, 3). In other words, acknowledging indeterminacy in the climate impacts of digitalization is not just for its own sake, it is USEFUL. We can stop wasting time and resources on a search for ever great accuracy, precision, and predictability of a situation that is inherently none of these things.

The kind of storytelling by analogy advocated by Gritsenko is useful. It is a good example of the underlying principle of ‘if it exists, it is possible’(Boulding 1978). That principle helps researchers and decision makers expand their horizons of thinkable thoughts essentially by saying, “I see that X exists over there. Perhaps Xi could work over here…”.

Meanwhile though, while investing in renewable energy infrastructure is likely a good thing in some ways, it does not automatically offer a solution to Jevons paradox nor does it automatically result in decoupling. It also does not avoid very tangley issues of colonialist solutionism for big tech’s carbon problems. Renewable energy infrastructure requires land and when that land is pre-supposed to be available for settler plans and goals, that is colonialism in action (Liboiron 2021).

So in addition to stories about tsunami-wall analogues we also need stories about digitalization that don’t default to enhancing colonial dispossession. So, what are some stories that might mitigate pollution and colonialism associated with digitalization?

References

Boulding, Kenneth E. 1978. Stable Peace. University of Texas Press.

Google. 2024. “Environmental Report | 2024.” https://www.gstatic.com/gumdrop/sustainability/google-2024-environmental-report.pdf.

Gritsenko, Daria, Jon Aaen, and Bent Flyvbjerg. 2024. “Rethinking Digitalization and Climate: Don’t Predict, Mitigate | Npj Climate Action.” Npj Climate Action 3 (1): 43. https://doi.org/10.1038/s44168-024-00127-z.

Liboiron, Max. 2021. Pollution Is Colonialism. Duke University Press.

Mann, Tobias. 2024. “So Much for Green Google – Emissions Rise 48% since 2019.” July 2, 2024. https://www.theregister.com/2024/07/02/google_datacenter_emissions/.

Masanet, Eric, Arman Shehabi, Nuoa Lei, Sarah Smith, and Jonathan Koomey. 2020. “Recalibrating Global Data Center Energy-Use Estimates.” Science 367 (6481): 984–86. https://doi.org/10.1126/science.aba3758.

Milmo, Dan. 2024. “Google’s Emissions Climb Nearly 50% in Five Years Due to AI Energy Demand.” The Guardian, July 2, 2024, sec. Technology. https://www.theguardian.com/technology/article/2024/jul/02/google-ai-emissions.

https://electronicplanet.xyz/2024/07/16/rebound-bites-google/

#climateChange #energy #environment #renewableEnergy #sustainability