Backgroound Image

CLIMATE: THE GREEN VEHICLE FOR THE AI OFFENSIVE

AI understands. Why don’t we?” – The mobilization of Fridays for Future Hamburg for the climate strike on 15.09.2023 assumes that the AI-based image generator understands climate change better than humans.

Artificial intelligence is currently on everyone’s lips – including in the climate movement. Read the first part of our critical series on AI to find out why this is problematic.

When Microsoft CEO Satya Nadella presented the new AI-supported version of the Bing search engine in February, he conjured up the “three-headed monster of inflation, recession and energy crisis” as a major threat scenario. Of course, only to announce immediately afterwards that the technical solution to these problems was within reach. It is not surprising that a company like Microsoft is “striving to use technology to overcome the major challenges facing people, organizations and countries,” as Nadella formulated the company’s goals. (1) After all, one of the aims of the company’s high-profile AI offensive is to finally regain market share in the Google-dominated search engine market.

What is more astonishing than Microsoft’s PR presentation is that large sections of the Green Party and the (professionalized) climate movement have also adopted this vision. The massive expansion of the use of technology and, in particular, so-called artificial intelligence is perceived as a great opportunity, if not a necessity, in the fight against the man-made climate crisis. Not a day goes by without a new technology being presented as “the” solution to the climate problem. If the vision of the comprehensive use of AI to solve political and social problems were to become a reality, it would lead to profound social upheaval in the interests of the powerful, progressive middle-class and now often green elites. In this article, we want to take a closer look at the green faith in technology. Our considerations are guided by the question of what consequences for individual and collective self-determination can be expected from combating climate change using artificial intelligence.

 

THE HOPE OF TECHNICAL SOLUTION

The initiative to introduce information technology into the further development of capitalism goes back a long way. However, contrary to what one might think, the idea did not originate in corporate headquarters such as Microsoft, but in the US counterculture of the 1960s. The current green ideas of self-regulating climate protection and a society optimized using feedback loops of information flows tie in with the utopias of the alternative movement of that time. (2) Today, these ideas lead to the push to make climate policy and AI the core of a totalizing breakthrough into a new era of capitalism in a new kind of technopolitical complex.

At the Heinrich Böll Foundation, this approach runs under the headline emphasizing reflection and openness: “AI & climate change – hype or opportunity?” The Heinrich Böll Foundation is, more than any other political foundation of other parties, the strategy and discourse organ (for opinion-forming and mood testing) of the Greens. Climate policy appears both as a vehicle for technological breakthroughs and as a vehicle for suppressing alternative political approaches (radical climate protection and social change) and assigns IT technologies a monopoly in climate policy. Ralf Fücks, one of the founders of the green think tank Zentrum Liberale Moderne and former director of the Heinrich Böll Foundation, also raves about a bright future thanks to technological innovation:

“A new economic dynamic can emerge from the race against climate change […], a long wave of environmentally friendly growth. Its drivers are artificial intelligence and the cybernetic control of production and logistics, hydrogen and synthetic fuels, e-mobility and battery technology, renewable materials, bionics and the broad field of biotechnology with higher-yielding, more robust crops and food from cell cultures.”

The literary scholar Roberto Simanowski summarizes this attitude as follows:

The hope that technology will save us in time from the disaster we are heading for by enabling ‘green’ economic growth, coupled with ‘sustainable’ consumption, is […] just an excuse for not having to change anything significant about the status quo. AI in its weak form is a declared expression of this hope that technology, rather than a turnaround, can protect us from the consequences of the technology developed so far; through the efficient use of heat using intelligent thermostats, the optimization of traffic control in the smart city or the reabsorption of CO2 from the atmosphere into the soil. ” (3)

The hope in technology is therefore a solutionist approach, i.e. the socially generated problem of climate change is translated into technical substitute problems, e.g. the enormous consumption of fossil energy, which can then be solved or optimized away using AI. (4)

 

TECHNOCENE OR ANTHROPOCENE?

Technocrats and solutions see the (current) political inability to even initiate a climate-friendly change of direction as confirmation that ‘man’ is incapable of

a) going beyond his own needs and

b) making rational decisions in the sense of a (global) common good beyond the immediate here and now. As a quasi-natural law, this insight into ‘human inability’ should pave the way for artificial intelligence. It could solve the climate crisis far better than humans because it is much better at processing data and detecting complex, climate-relevant correlations.

People might cynically remark that an AI (equipped with far-reaching decision-making powers) could hardly make climate policy any worse than current policy. However, contrary to what James Lovelock, one of the intellectual references of parts of the ecology movement, analyzes in his book Novocene – The Coming Age of Hyperintelligence, we are no longer facing a problem of knowledge, but a problem of will. The political shift from a focus on the individual in the here and now to a society that radically places the sustainable community at the centre cannot be shortened by handing over decision-making to artificial intelligence. The reason for this lies less in the technical problem of never having a balanced database, which is used to train the AI’s self-learning algorithms and thus leads to an unusable reinforcement of these data biases by the AI. Rather, the reason lies in the conceptual inadequacy of machine learning to represent a useful notion of the common good.

A semantically clueless AI that merely performs pattern recognition and optimization of statistical weights has no idea of what a common good could be and how it can be dynamically developed in a meaningful way, no matter how impressively ‘human-like’ self-learning language models à la ChatGPT already imitate problem-solving strategies. Worse still, the concept of optimization based on existing data inevitably perpetuates the past (stabilizing power) into the future. As a result, incapacitating AI as a recommendation and decision-making assistant for solving the ecological crisis turns out to be a socio-technological dead end – it is unsuitable as a (techno-) revolutionary tool. Its attractiveness merely stems from a double abdication of responsibility, firstly for the majority of people, who no longer have to deal with the climate change they have caused, as an AI will find better solutions than they can themselves anyway, and secondly for decision-makers, who can conceal the political nature of groundbreaking social decisions and shift their own responsibility to the public onto the AI to be able to push through unpopular measures if necessary.

 

GREEN AI AS A MEANS OF GAINING ACCEPTANCE FOR SOCIAL UPHEAVAL

It should be noted that the hope for technology in the green vision is not merely to be characterized as preservative in the sense of “managing an already established order”, but that the promise of AI to those who decide on its use lies precisely in being able to enforce the applicable rules down to the last entanglements of the social fabric. Such optimization of the enforcement of social order using AI-driven automation – even if this follows the progressive goals in the spirit of climate protection – is more than just a capture of ever more areas of human life in the sense of quantitative expansion. It forms the basis for a profound qualitative transformation of social relations. The policy of a combined climate policy/technological breakthrough can therefore be described as total, or rather “totalizing” in the same sense as the combination of Taylorism/Fordism and the implementation of savings and certainly, as in the USA, ecological objectives was a hundred years ago. The green AI offensive is also totalizing because it has and is intended to have an effect on people’s mentalities in the sense of changing society as a whole.

This can be seen in detail in the Heinrich Böll Foundation’s publication “Smart Technology against Climate Change, 15 Facts about Artificial Intelligence“. (5) It represents the outline of a comprehensive socio-political project. Here, the climate policy/technological application in the areas of resource consumption, Industry 4.0, transport and mobility, agriculture, forestry and species management is played out. Clearly with a tendency to expand to other areas of society. We would like to emphasize the following: A policy of unconditional avoidance of catastrophe is not even pursued, it is about “social adaptation” (p.8). Critical aspects of capitalism also no longer appear when it says “making the energy market comprehensible” (p. 14). The propaganda of “precision agriculture” using AI is strikingly analogous to the Stalinist strategy of a totalizing approach to transforming the entire agricultural sector into a Fordist/Taylorist machine, with the well-known catastrophic results described by Josephson as “brute force technology“. The totalizing tendency towards a new climate policy-technological expertocracy is so hermetic that alternative political forms of climate policy and the relationship to the new technologies no longer appear at all. Climate and AI policies are taking on the form of a closed system that no longer allows scope for autonomous processes. Criticism? “To create trust in AI, we must also deal with its possible negative consequences” (p.32). (6)

The Greens are not alone in their emphatically reflective approach to the topic. After all, the Federal Ministry for the Environment, Nature Conservation and Nuclear Safety (BMU) had already launched a “Five-point program ‘Artificial intelligence for the environment and climate'” at the federal level in 2021, i.e. even before the Greens joined the government. This is primarily concerned with the creation of so-called AI lighthouses, “projects with an impact on environmental protection. ” (7) And the BMU also likes to be critical: “Because there are ecological downsides that we have to take into account: The billions of calculations on high-performance processors that give AI systems their impressive capabilities consume a lot of energy,” reads the accompanying factsheet, only to dream one paragraph further of turning “a strong ‘Sustainable AI made in Europe’ brand” into a competitive advantage.

Together with the Federal Ministry of Labor and Social Affairs and the Federal Ministry for Family Affairs, Women, Senior Citizens and Youth, the BMU has also funded the research project “Civic Coding – Innovation Network AI for the Common Good”. In this project, the German government is placing an internationally unique focus on “public welfare-oriented” AI development, with one of the focal points here also being environmental goals with the “AI Ideas Workshop“. (8) However, the actual projects mentioned are usually rather meagre lighthouses. Here a few beehives are equipped with sensors to get to the bottom of bee mortality, and there the irrigation needs of urban trees are predicted. The unsurprising impression is that more complex problems are a long way from being solved automatically by AI and that the ministry is more interested in actively shaping the discourse on these technologies.

 

AI AS A CLIMATE KILLER

It is not without reason that the German government is giving the AI lighthouses a critical and reflective coating. After all, it is by no means a foregone conclusion that AI is part of the solution and not part of the problem. After all, the resource and energy consumption of machine learning is enormous. Facebook, for example, estimates the energy consumed in training its own language model LlaMA at 2638 MWh. (9) To put this into perspective, a modern wind turbine, which can supply around 3500 households with energy, has to run for three months to produce this amount of energy. (10) While Facebook holds out the prospect that energy consumption will be relatively low after the one-off training process and that a single GPU may be sufficient to operate the trained AI, it must be noted here that the planned mass use of the AI is expected to result in enormous energy requirements – not to mention other hidden energy costs such as the production of the necessary hardware.

It is already estimated that around 12 per cent of the world’s electricity demand goes into digital devices. And the trend is rising. (11) A single Google request – without using AI-based chat assistants, mind you – consumes around 0.3 Wh – the amount of energy needed to power an energy-saving LED for three minutes. A chat GPT3 request that is answered using already trained AI, on the other hand, already consumes 1.3 Wh, i.e. more than four times as much – although the particularly energy-intensive training is not taken into account here. (12) These figures are of course only estimates and the real energy consumption is likely to be many times higher. (13) After all, it is not only the electricity used to operate the data centres that are consumed but also the production and transport of the hardware as well as developers and their equipment consume energy, which is often not included in the above estimates.

Given the enormous energy consumption of digital technologies, scepticism is warranted when they are touted as a means of reducing energy consumption. However, the German government seems to have set itself the goal of dispelling these doubts. Therefore, among the AI lighthouses mentioned above, there are also some projects to optimize the resource consumption of AI itself or make it more transparent. One such lighthouse is the NADIKI project at the University of Stuttgart, which aims to make the real energy and resource consumption of AI available via a software interface. The press release on the funding decision states:

For sustainable AI use, it is therefore important to make the best possible use of existing infrastructure to reduce or avoid the construction of new data centres, servers or network equipment. At the same time, AI systems should be optimally utilized and resource consumption should be recorded and disclosed. ” (14)

This sets the framework for the critical debate on the ecological consequences. An open-ended question of whether AI should be used at all – even if only from an ecological perspective – is not the focus. All that remains to be clarified is how its use can be made “sustainable” – in other words, the use of AI becomes a second-order optimization problem. One thing is certain, even without the results of the Stuttgart researchers – initially, the use of machine learning will increase energy requirements. The high fixed costs involved in training AI models mean that efficient use is only conceivable if the model is subsequently applied on a large scale. AI models are therefore hardly an option for solving specialized (climate) problems where large-scale use is not expected.

There is another reason to be sceptical about the energy-saving promises of green AI: the so-called rebound effect. This states that increases in energy consumption efficiency do not lead to an overall reduction in consumption, but merely to a reduction in costs and the surplus energy that is no longer needed is instead consumed elsewhere. (15) A simple example: the reduction in fuel consumption of modern cars has not led to less fuel being used, but firstly to more cars being driven because more people can afford them, and secondly to larger cars such as SUVs being produced, which in turn have a very high fuel consumption and would not have been conceivable without the increases in efficiency. Optimizing the existing economic sectors will not lead to a real reduction in energy consumption. Under these conditions, the efficiency gains achieved through green AI also appear unlikely to make a significant contribution to combating climate change. Nevertheless, many liberals, such as the economic historian Adam Tooze, in view of the resistance of the political and economic elites to fundamental social change, are relying – sometimes more, sometimes less grudgingly – primarily on technical solutions. (16) This raises the question: How can we succeed in having a debate about tackling the political-economic causes of climate change that is not stifled at the outset by references to imminent technical solutions?

 

PROGRESSIVE, BUT NOT EMANCIPATORY

In her taz column, Charlotte Wiedemann sharply criticizes the Greens for the “unfounded extent of European faith in violence in the Ukraine war”, which the Greens in Germany are promoting with their “feminist foreign policy” like no other political force. (17) She specifies:

“Today, however, the Greens have become a force of discipline, of containment, of anaesthetizing and stultifying thought. While others desperately cling to the pavement, the Greens are glued to the prevailing conditions.

And indeed – not only the Green position in the Ukraine war, but also the use of AI to combat climate change are symptoms of a fatal idea that is typical of post-democratic societies: the implementation of progressive policies while simultaneously abandoning emancipatory demands. This policy is socially and ecologically progressive because climate change must be stopped in order to prevent the catastrophic consequences, particularly for poor people and nature. The Greens want – in part – to reduce the fight against climate change to the implementation of algorithmically calculated measures. A social debate about the concrete goals and the resulting measures is merely an accessory. The reference to technological solutions therefore serves precisely the purpose of rejecting such a debate about far-reaching changes to the causal social power relations from the outset. Unlike Adam Tooze quoted above, most Greens therefore do not regret relying almost exclusively on technological solutions due to a lack of political majorities. On the contrary: the cybernetic society is to be sold to us as a green utopia, although it actually contains more of what already exists. After all, large sections of the climate movement recognize that this green stance represents a frontal attack on all those who want to adhere to the Zapatista slogan “Another world is possible”. This disagreement between the Green party leadership and the climate movement was most clearly demonstrated at the beginning of the year by the movement’s broad resistance to the eviction of the village of Lützerath negotiated by the Greens. (18)

Even if the Greens do not – as would be in the spirit of emancipatory politics – pursue the joint and open negotiation of measures against climate change, they remain on a liberal path and (still) differ from the authoritarian path taken in China, for example. Colin Crouch coined the term post-democracy to describe this specific attitude, which formally upholds basic liberal freedoms but at the same time shifts the focus entirely to the effectiveness of the implementation of political goals.19 The aforementioned Ralf Fücks from the Zentrum Liberale Moderne (Center for Liberal Modernity) accordingly distances himself sharply from a decidedly authoritarian project by combining typical neoliberal argumentation patterns with the idea of national competitiveness:

Anyone who wants to reconcile freedom and ecology must above all focus on innovation and promote competition for the best solutions. This requires an ecological regulatory framework that steers the dynamics of the market economy in an ecological direction. Even a market-based climate policy cannot do without rules and bans. However, they are not the ideal way to overcome the ecological crisis. Top-down control through tightly meshed state regulations can never replace the innovative power of the market economy, which pools the knowledge and initiative of millions and millions of producers and consumers. ” (20)

 

AND THE CLIMATE MOVEMENT?

Progressive, but not emancipatory – the Green Party leadership is not alone with this attitude. Rather, it reflects a wider social development. It is no coincidence that the Last Generation is considered to be the political movement in Germany in 2023 that receives the most media attention and openly expresses its disinterest in the values of enlightenment and the tradition of (left-wing) liberation struggles. Carla Rochel, a member of the so-called Last Generation strategy team, makes an unequivocal rejection of emancipatory politics in favor of the supposed real-political implementation of her own goals: “We do everything for a good feedback culture, but unfortunately we have seen with other organizations that grassroots democracy takes too much time, which we don’t have. ” (21) In practice, this means that the strategy team, i.e. a handful of people, plan and the so-called “bees” merely wait for their deployment order, which tells them when and where to stick to the streets. It is not unlikely that such an attitude will fall on the group’s feet. If it should turn out that the action know-how is not distributed widely enough to continue in the long term despite state repression against the group and to react flexibly to political changes, the advantages of decentralized forms of organization could be recalled.

However, there are also other approaches within the climate movement that offer more hope because they do not want to submit to the green confinement of thought without a fight. In addition to the aforementioned resistance to opencast coal mining (in Lützerath), there are also the protests in Sainte Soline, France, against the so-called mega basins, huge artificial lakes designed to supply industrial agriculture with water in times of increasing drought. (22) Or many smaller actions, such as those documented on the blog https://switchoff.noblogs.org/. The call for action published there explicitly states:

If we are sold the illusion that climate change can be stopped technologically, then this is based on the trust that those in power only need to take the right steps, take the right measures to save this world.
 For one thing, they have absolutely no interest in ending the expansionist capitalism that secures their position of power. And for another, technological reform, with the new dependencies it produces, is also doomed to failure.”

The question is whether such an attitude is still capable of winning a majority in the climate and left-wing movement, or whether the technological attack on the task of emancipatory politics has already progressed too far.


(1) https://news.microsoft.com/wp-content/uploads/prod/2023/02/Reinventing-search-with-a-new-AI-powered-Bing-and-Edge-1.pdf

(2) Cf. https://jacobin.de/artikel/techno-okologie-astrid-zimmermann-klimawandel-blockchain-terra0-web3-whole-earth-catalogue-buckminster-fuller-solutionismus-design-thinking/

(3) Roberto Simanowski. 2020. the death algorithm. The dilemma of artificial intelligence. S. 112.

(4) For a discussion of solutionism, see Redaktionskollektiv Capulcu. AI for programmatic inequality, In: Editorial collective Capulcu. 2020. diverge – Divergence from regressive “progress”.

(5) https://www.boell.de/sites/default/files/2022-04/BoellFakten_Smarte_Technologie_gegen_den_Klimawandel_15_Fakten_ueber_Kuenstliche_Intelligenz.pdf

(6) Cf. editorial collective Capulcu. IT – The technological attack of the 21st century. In: Editorial collective Capulcu. 2017s. Disrupt – Resistance to the technological attack.

(7) https://www.bmuv.de/fileadmin/Daten_BMU/Download_PDF/Digitalisierung/factsheet_ki_bf.pdf

(8) https://www.civic-coding.de/angebote/publikationen

(9) Hugo Touvron et al. 2023. LLaMA: Open and Efficient Foundation Language Models.

(10) https://www.ndr.de/nachrichten/info/Watt-Das-leisten-die-Anlagen-im-Vergleich,watt250.html

(11) https://www.deutschlandfunk.de/stromverbrauch-digitalisierung-internet-bitcoin-rechenzentren-abwaerme-100.html

(12) Cf. https://medium.com/@zodhyatech/how-much-energy-does-chatgpt-consume-4cba1a7aef85

(13) For an overview of the problem of correctly recording the energy consumption of AI and the current state of research, see https://www.theguardian.com/technology/2023/aug/01/techscape-environment-cost-ai-artificial-intelligence

(14) https://www.uni-stuttgart.de/universitaet/aktuelles/meldungen/Foerderbescheid-fuer-KI-Leuchtturmprojekt-NADIKI/

(15) Cf. https://www.umweltbundesamt.de/themen/abfall-ressourcen/oekonomische-rechtliche-aspekte-der/rebound-effekte

(16) Cf. https://nymag.com/intelligencer/2021/05/adam-tooze-on-climate-politics-after-covid.html and https://www.youtube.com/watch?v=w4Y9SomH9Nc

(17) https://taz.de/Die-Entwicklung-der-Gruenen/!5940274/

(18) Cf. https://www.tagesschau.de/inland/innenpolitik/gruene-luetzerath-107.html

(19) Colin Crouch. 2008. post-democracy.

(20) https://libmod.de/aufbruch-statt-abbruch-mit-gruenem-wachstum-aus-der-klimakrise/

(21) https://taz.de/Wer-ist-die-Letzte-Generation/!5898641/

(22) https://tumulte.org/2023/03/articles/berichte-aus-sant-soline/