What Role Can AI Really Play in Averting a Climate Catastrophe?
This century will be shaped by two major forces: Climate Change and Artificial Intelligence (AI). To avoid a climate catastrophe, we need to drastically reduce emissions and adapt to an already changed climate at scales and speeds never seen before. Within this century, we will need to transform entire industries. A lot needs to happen already within this decade. This is unprecedented. At the same time, we are witnessing the birth of a technology of unimaginable potential, AI, with its abilities growing exponentially over time. With the release of ChatGPT in November 2022, AI has dominated discussions across spheres of society - from living rooms to board rooms.
Naturally, the question arises: what role can AI play in our efforts to avert an irreversible climate catastrophe - a scenario that becomes more likely every day? Living in San Francisco, at the heart of the AI revolution, I often encounter the belief that advanced AI will resolve our climate crisis by driving scientific and technological breakthroughs. At the same time, I also hear those voices on the other side, which dismiss AI entirely as part of the solution, even claiming that it will have an entirely negative impact on our climate efforts. Both perspectives are overly simplistic and risky.
To those arguing that AI is the panacea to climate change: First, climate change is not just a technological problem but also driven by political, social, cultural, and economic factors. If we want to create a better future for humanity, we need holistic solutions that address those dimensions, tailored to local contexts and guided by a clear vision. Second, we are running out of time. We cannot afford to bet on future technology (e.g., even if we figured nuclear fusion out today with the help of AI, it would still take us decades to get it to market at scale) while we lock ourselves into irreversible worst-case climate scenarios. Every emission we avoid today makes a difference in the long term.
To those categorically dismissing AI: We must act now - cut current emissions, prevent future ones, and adapt to climate risks. However, while technology alone is not the answer to “fixing” climate change and building a prosperous, sustainable world, existing and future technologies are essential tools in helping us get there. We must leverage EVERY tool available - and AI is a mighty one that must be part of the equation.
Therefore, what we really need to ask ourselves is: Where and how can AI already support impactful efforts that are net-positive for climate and people this decade, and what does it take to unlock this potential?
We are off track and running out of time
To begin with, we need to understand the status quo, the severity of the climate crisis, and what it takes to mitigate climate risks. Scientists warn us, that we have entered unchartered territory never witnessed in human history. 2023 brought unprecedented climate extremes, from record-breaking global heatwaves and warm oceans to the lowest Antarctic sea ice levels on record. From June to August 2023, the planet experienced its warmest period on record, with July seeing the highest daily average surface temperature in over 100,000 years! Nations worldwide were affected. Here in the US, 2023 marked a historic year in terms of climate extremes, with 28 climate-related disasters costing at least $92.9 billion!
As our planet shifts towards dangerous climate instability, greenhouse gas (GHG) emissions continue to rise alarmingly. The United Nations Global Stocktake says that on our current emissions trajectory, we will overshoot emissions by ~40% (~22 gigatons CO2) by 2030, above where we need to be to limit warming to 1.5°C - critical to avoid a long-term climate disaster. To course-correct, we must expedite decarbonization across all sectors, avoid new emissions sources and carbon lock-ins (e.g., installing new GHG-emitting infrastructure and machines commits us to years of emissions), and start adapting to the climate changes already set in motion. A lot of this needs to happen within this decade!
No miracles are needed (for now) - we need to deploy already existing climate technologies rapidly
50% of the emission reductions needed by 2030 can be achieved with existing climate technologies. Half of this can be achieved with technologies that are currently available at below $20/ton CO2, such as solar and wind energy, energy efficiency measures, methane emission reduction measures, and natural ecosystem conservation. So, the technologies we already have can take us a long way. However, many of those technologies are only taking off slowly in the market.
We need to find ways to accelerate the deployment of those technologies. To put things into perspective, the 2023 Climate Conference commitment to triple renewable energy to 11,000 GW by 2030 requires a deployment speed seven times the 2023 deployment rate. A lot of this needs to happen in emerging and developing economies - which will drive future global energy demand. In South Africa, I led the development of net-zero pathways, which showed that to decarbonize the country’s power system, renewables need to be rolled out at an unprecedented rate of ~6 GW per year, ten times the current roll-out rate! Speeding up deployment at these rates requires innovative policy, financial, but also technical solutions.
In summary, we are off track, continuing to emit GHGs at record levels, are already facing severe climate impacts, need unprecedented systemic transformations that consider equity and justice - and time is running out. How can AI fit into this critical and complex situation?
AI already boosts climate action by enhancing and helping scale existing climate technologies
First, what does AI actually mean and encompass? Today, when we speak of AI, we mean dedicated systems that can independently identify patterns in vast amounts of data, make predictions, and solve complex optimization problems. This can be achieved through a range of AI approaches: for example, deep learning models classify and recognize patterns in visual, audio, and other data; generative models like ChatGPT produce new content from text, images, and audio; and reinforcement learning enables decision-making in complex environments, such as autonomous driving. Today’s AI is 'narrow,' specialized for specific tasks. However, AI capabilities are exponentially advancing, promising radically different - potentially today unimaginable - applications in the near future. There are many ways that AI can - and already does - contribute to climate action. AI can:
Enhance data analysis capabilities, e.g., allowing the processing of extensive climate-related data, such as satellite imagery, to inform adaptation measures
Improve projections, e.g., forecasts for renewable energy production, transportation demands, and weather predictions
Optimize existing infrastructure and systems, e.g., optimizing system efficiencies and facilitating predictive maintenance
Accelerate scientific and technological breakthroughs, e.g., the development of next-generation energy sources like nuclear fusion or advanced battery materials
In simple terms, AI can help us better understand climate change, improve the effectiveness and speed of the climate action we already pursue today, and open up avenues for entirely new, transformational climate solutions. Already today, “narrow” AI plays a crucial role in supporting climate efforts across various sectors and industries - leveraging technologies that are widely available and commercially proven, such as renewable energy. To make this more tangible, let's look at examples in the power sector, which accounts for ~30% of global emissions. There, AI already
Speeds up the construction of renewable energy. I have worked in many regions where the permitting of new renewable infrastructure is painfully slow. Various AI startups are trying to address this. Symbium has developed a “permitting pilot” to automate permitting for residential solar installations. This is achieved through a Logic Programming Language model that encodes the relevant legal codes and regulations, which the pilot leverages to provide immediate feedback on project plans. Previously, it took a worker one hour to review a residential solar project plan. Now, it is done instantaneously, resulting in massive time and cost savings.
Boosts the viability of renewable sources like solar and wind. For example, Google trained a machine learning algorithm using weather forecasts and historical wind turbine data to predict the output of its wind turbine fleets 36 hours in advance. Another model recommends optimal hourly delivery commitments to the power grid 24 hours ahead. By making volatile wind power predictable and schedulable - which decreases the dependence on energy storage and the necessity for standby power sources - Google improved the economic value of its wind power operations by 20%. Others, like Open Climate Fix, aim to forecast solar PV power production.
Enables the creation of smart power grids that communicate closely with demand centers. For example, WeaveGrid enables the integration of electric vehicles into the power grid at scale. It achieves this through an AI-powered platform that connects the EV to the grid autonomously, identifying optimal charging times and reducing overload risks to the grid. The platform also enables the autonomous discharge of batteries into the grid, which can turn an EV into a valuable energy storage source.
Helps utilities manage grid infrastructure, identify climate risk, and inform grid investment decisions. Rhizome’s AI-powered platform leverages historical data from utility companies on energy equipment performance integrated with global climate models to identify vulnerabilities from extreme climate events, quantify risks at high resolutions (e.g., likelihood of grid failures), and measure the socio-economic benefits of grid-enhancing investments. BuzzSolutions uses AI-powered predictive analytics tools - leveraging visual data collected from a fleet of drones - to detect faults and anomalies in vast stretches of power grid infrastructure.
Again, these are just a few examples. The field is much broader; many more applications supporting climate mitigation and adaptation exist across sectors. A recent report by my colleagues at BCG in collaboration with Google suggests AI could help reduce emissions by 5-10% by 2030. As AI computational power continues to grow exponentially over time, I expect even more opportunities for impactful climate action and substantial emission reduction to emerge. However, the key question is whether the use of AI in addressing climate change can have a net-positive impact - on both climate and people.
AI needs to have a net-positive impact on climate and people - this is not given
The widespread use of AI brings many risks with it. I want to focus on those specific to its use in the context of climate change. However, this is a living list that needs to be refined further. Hence, without being exhaustive, some of the most critical risks to consider are that
We fail to identify areas where AI can address critical bottlenecks to climate action. AI could help address roadblocks that currently hamper the scale-up of proven climate technologies/solutions. However, there is a disconnect between those who are facing roadblocks and those who can build AI-powered tools. If those two “camps” are not better connected, we risk failing to identify impactful AI use cases in climate. (Note: ClimateChange AI did an excellent job in identifying areas across domains and sectors where AI can play a role in climate action).
We do not develop solutions for where and for whom they are needed most. By 2050, global energy use and emissions hubs will likely shift from Europe, North America, and China to the Indian subcontinent, sub-Saharan Africa, and Southeast Asia. We must rapidly deploy climate technologies in these areas to enable low-carbon, climate-resilient development and prevent extensive carbon lock-ins. AI could be pivotal here, for instance, by enhancing the planning, construction, and management of power systems in under-electrified, sparsely populated areas. However, most AI research and development is happening in developed economies. Also, less developed world regions often lack accessible, high-quality data, critical for training AI. Hence, there is a risk of blindspots, bias, and failure to create AI-powered solutions that meet the needs of those growing economies and populations.
We use AI without consideration of the need for Just Transitions. We must ask ourselves for “whom” we are developing and deploying AI solutions to create “what” outcomes. Decarbonization and also the use of AI might induce significant shifts in job markets, particularly in sectors most affected by climate policies, such as energy, transport, agriculture, and manufacturing. However, it could also support a Just Transition. For example, imagine pairing local technicians with basic education with AI software to bridge a skills gap in the installation of renewable infrastructure, like residential solar installations. The challenge lies in balancing technological advancement with equitable socio-economic development.
We fail to address AI’s own potential climate and environmental footprint. AI relies on energy-intensive computing. Research showed that generating an image with a powerful AI tool requires as much energy today as fully charging a smartphone! AI’s energy (and other resources like water) demand is expected to grow. Today, AI operations make up less than 1% of global GHG emissions; however, some scenarios point towards significantly increasing energy demand (the IEA estimates that by 2026, data center’s power demand could double to 1000 TWh, same as Japan’s power consumption) and future emissions - if AI’s energy sources are fossil-based. However, what counts is whether advancements in AI algorithms, hardware, and smart decisions around where and how to operate data centers, combined with AI's contribution to driving more impactful positive climate action, can yield a net positive climate impact.
We leverage AI to support and enable actions that exacerbate climate change. This can include everything from using AI to boost oil and gas exploration and production, driving increased consumption and waste through, e.g., sophisticated marketing, to enabling AI-driven geoengineering projects in the future that could have unintended side effects that alter ecosystems or weather patterns globally—or even using AI—which consumes energy and scarce resources—for pretty non-value-adding things.
Our use of AI creates safety and security risks. If AI systems are deployed in critical infrastructure, various risks could arise, from extending the range of targets available to cyber attacks to AI malfunctions or the production of unintended effects. These risks are particularly pronounced when AI is utilized for real-time decision-making tasks in sensitive infrastructure, such as managing factory operations or controlling power grids. In addition, some use cases have “hard constraints” - think of operating a power grid. A disastrous failure could result if an AI does not produce a 100% “physically accurate” answer. So, we must be cautious about how and where we integrate AI.
The AI and Climate “camps” must connect and collaborate better
AI could offer great potential for reducing emissions and delivering socio-economic benefits already in the short-term. However, there is a disconnect: the “climate camp” often lacks a deep understanding of AI and where it could be deployed, and the “AI camp” is often unfamiliar with the complexities of climate mitigation and adaptation. To maximize AI's potential in supporting climate action within this decade, these groups must unite to uncover opportunities in deploying technologies that can rapidly scale down emissions. In my opinion, some questions these groups should jointly address:
Rapid scaling of existing climate technologies: Where and how can AI already accelerate the deployment of existing climate technologies?
Preventing carbon lock-in: How can we develop solutions for emerging and developing markets - where we have the most significant carbon lock-in and climate risks?
Overcoming barriers and risks: What are the main obstacles to deploying AI for large-scale, rapid emission reductions, and what risks should we consider?
Ensuring net-positive impact: What ethical frameworks (from social to environmental aspects) should guide AI’s role in climate action?
Convening the right experts and stakeholders: Who needs to come together (e.g., tech builders, climate experts, NGOs, etc.) to maximize AI’s positive impact on climate change?
If you have thoughts and comments, are interested in continuing this conversation, or are seeking a personal discussion on this topic, please do reach out!