The field of Artificial Intelligence has seen a phenomenal development over the past years, which led to massive investment, partly oriented towards addressing pressing societal challenges. Breakthroughs observed in the domain of machine learning during the past decade have raised hopes, and lifted expectations, about the thaumaturgical powers of this general-purpose technology. Be that the COVID-19 pandemic, the imperative of decarbonisation, or the new ambition to build resilient societies, AI is often evoked as a saviour, a magic key to a more prosperous future. And indeed, there are several emerging use cases that have the potential to make inroads in the quest for sustainability, in particular when it comes to addressing climate change. Adequate and diffuse AI deployment appears to be an essential condition to tackling key problems in sectors that have gone way outside planetary boundaries, such as agriculture and food, where the surgical application of digital technologies technology promises to enable feeding the rising population of our planet without having to increase food production (some say by as much as 70%), and without massively spraying pesticides. AI, and especially machine learning, can help optimise energy consumption, optimise complex transport systems, embed circular economy in product design, help identify environmentally friendly solutions in new buildings, and much more. Most importantly, as the world increasingly realises that climate change, the collapse of biodiversity and the increased frequency of natural and health disasters are interrelated, machine learning can help us model the interactive effects of adaptation and mitigation measures, especially in the context of large-scale models such as the “Digital Twin of the Earth”.

At the same time, it should remain clear that AI is going to be no panacea. Without the deployment of complementary technologies, a truly international collaboration, appropriate safeguards in steering the deployment of AI solutions, and adequate changes in human behaviour, government policies, corporate and consumer conduct, simply evoking AI is not going to lead us very far.

Accordingly, the time is ripe for an agenda that tackles the problem seriously and upfront. Below, I lay down some thoughts on what this agenda may include.

First, merely seeking “more AI” makes little sense. Comparing aggregate levels of investment is useful, but does not tell the whole, or even the most meaningful, story. Investing in AI solutions that help us choose better movies, or snap better selfies, does not lead us any closer to solving our grand challenges, however sizeable the impact on GDP. The right question is therefore: “AI for what?”, or even better “what AI, for what use, and what impacts”? Two years ago, in releasing its Ethics Guidelines, the EU High Level Expert Group on AI included societal and environmental well-being among the seven key requirements of Trustworthy AI. Despite its overall endorsement of the work of the HLEG, the European Commission did not replicate this requirement in the 2020 AI White Paper and in the proposed AI Act presented in April 2021. And while this may be understandable (the AI Act is more focused on risks for fundamental rights and safety, and is expected to target a rather small fraction of total AI systems), this leaves space for proactive policies and investment to ensure the deployment of AI systems with reduced environmental footprint, and oriented towards key societal and environmental challenges. Such measures include the strategic use of public procurement to create a market for trustworthy AI solutions; engaging in large-scale data sharing, creating data collaboratives and facilitating AI research in the context of the emerging data spaces (such as the one for the Green Deal); launching mission-oriented projects to achieve breakthroughs within a limited timeframe; and strengthening university curricula in the domains that can contribute to solving the climate challenge. Importantly, policymakers should consider involving the data community in reframing the policy questions, so that data science can help address them (one good example is the “100 Questions” initiative led by NYU GovLab).

Second, AI alone is not going to suffice. AI is certainly a powerful family of techniques when used as stand-alone software. But its potential is magnified when it relies on high quantity and quality of data, hardware-software combinations including the Internet of Things, suitable computing power (including quantum) and edge/cloud infrastructure. Many of these ingredients require scale and substantial international collaboration, especially on regulation, on standards for responsible AI development, and R&D projects that encompass the whole technology stack needed to address urgent environmental challenges. In the EU, initiatives on these complementary technologies are either missing, or still at the proposal stage, and the overall picture that is emerging appears to be quite fragmented. Understanding the key requirements and risks of leveraging digital technologies such as AI to fight climate change is also part of an urgent endeavour: clarifying the scope of the “twin transition” by exploring when, and under what conditions, digital technologies can be an ally of decarbonisation.

Third, the EU alone will not suffice. Given the urgency of acting on climate, and the scale of the problem, it is important to discuss whether Europe’s current quest for technological sovereignty and open strategic autonomy is well timed and conceived; or whether international AI collaboration should not be stepped up to enable researchers from the whole globe to help tackle climate change before it’s too late. The challenge is existential, and the EU has a chance to take the lead by proposing the creation of a global collaborative research and innovation platform, such as CERN or the International Space Station among others. To tackle climate change and biodiversity, the world needs an “International Earth Station” (if possible even broader than the ISS, which does not include China), and organised as a mission-oriented initiative, with a broad portfolio of projects and an independent, inclusive governance. In this context, collaborative research could also be directed at AI techniques that, by saving on data requirements and by decentralising computing power and intelligence, improve AI’s carbon footprint.

Finally, when it comes to climate change, the human component is going to be at least as important as the technological one. It is essential that both the EU and its Member States progress in terms of policy coherence for sustainable development, and in particular when it comes to aligning policies for decarbonisation. As of today, there seems to be margin for improvement in the EU industrial strategy, in the better regulation agenda, in the EU policies on sustainable finance, as well as in the circular economy action plan, in the Common Agricultural Policy and in several other areas. The EU will also have the responsibility to ensure that national recovery and resilience plans fully incorporate the logic of the Green Deal, and coherently aim for the transformation of entire economic sectors, to align them with the ambitious targets set, such as a 55% reduction in carbon emissions by 2030, and net-zero emissions by 2050. This is not going to be easy, especially since the post-pandemic recovery will have to deal with the legitimate needs of those individuals and businesses that have suffered the most from the economic depression, and this may often stand in the way of an ambitious, transformative and pan-European policy for climate change.

Only a bold, coordinated and globally oriented set of initiatives can ensure that AI’s potential to fight climate change is fully unlocked. Otherwise, we will keep missing the forest for the trees. We will continue to praise individual achievements and breakthroughs in AI, but we will fall short of achieving the systemic transformation that is needed to tackle the most existential challenge of our times.