Artificial intelligence (AI) is rapidly advancing and holds significant potential for climate change mitigation and adaptation. However, its impact remains uncertain, with concerns about increased energy consumption and carbon footprint. Previous research relied heavily on expert studies and literature reviews, lacking large-scale data analysis. This study addresses this gap by leveraging a comprehensive dataset of US patents to quantitatively assess the relationship between AI and climate-related inventions. The research aims to identify the sectors where AI is most prevalent in climate technologies and to evaluate its impact on subsequent innovation, taking into account knowledge spillovers to other technological areas. The findings are crucial for understanding the role of AI in climate change solutions and for informing effective climate policy.

Existing research on the connection between AI and climate change often focuses on the UN Sustainable Development Goals and employs expert-based studies. These studies have highlighted both the potential benefits and drawbacks of AI in climate action, but offer mixed conclusions regarding its net effect on the climate system. While expert reviews integrate knowledge across domains, they are limited by challenges in scaling up analysis to cover extensive literature and the difficulty in fully understanding the implicit judgements within expert assessments. This study seeks to complement existing research by utilizing a large-scale data source – patent data – to obtain a more comprehensive and quantitative perspective on the topic.

This study uses a large-scale dataset of over six million US patents granted between 1976 and 2019. Climate-related patents were identified using the Y02 Cooperative Patent Classification (CPC) system, which covers technologies for mitigating and adapting to climate change. AI patents were identified using a computationally automated method developed by the World Intellectual Property Organization (WIPO), based on patent classification codes and keyword matching. The study combined these classifications to identify patents related to both AI and climate change. The analysis focused on forward citation counts, which represent the number of subsequent patents that cite a given patent, as an indicator of technological impact. Count regression models were employed to estimate the predictive difference of AI on patent citation counts, controlling for factors such as year, technological area, and other variables relevant to patent citation patterns. The analysis distinguished between citations from within and outside the climate field to understand knowledge spillovers. Highly cited patents (breakthroughs) were also analyzed to determine whether AI was associated with a higher share of such breakthroughs.

The analysis revealed several key findings. First, AI is predominantly integrated into climate patents related to transportation, energy, and industrial production technologies. Second, highly cited climate patents exhibit a higher frequency of AI in adaptation and transport sectors compared to other mitigation areas. Third, AI-related climate patents are significantly associated with a 30-100% increase in subsequent inventions across various technology areas. However, a substantial portion of these subsequent citations originates from outside the climate field. Fourth, while AI patents are associated with more subsequent inventions on average, fewer AI patents were among the highly cited breakthroughs. Analysis of highly cited breakthroughs revealed a higher share of AI breakthroughs in climate adaptation and transportation technologies compared to other mitigation areas. The study's findings reveal a double association of AI in climate patents: while AI is related to increased subsequent invention activity, it also exhibits a greater share of knowledge spillovers to non-climate fields.

The study's findings highlight the complex impact of AI on climate innovation. The significant increase in subsequent inventions associated with AI in climate patents suggests a positive effect on technological advancement. However, the substantial knowledge spillovers to non-climate areas suggest that the benefits of AI in climate-related technologies may extend beyond the immediate climate domain. This underscores the need for a holistic perspective that considers both direct and indirect impacts. The relatively lower proportion of AI among highly cited breakthroughs warrants further investigation, potentially reflecting challenges in translating AI inventions into widely adopted, high-impact technologies within the climate field. The differences observed between adaptation and mitigation technologies could inform targeted strategies for AI deployment in specific areas.

This study offers a novel approach for tracking the role of AI in climate innovation using patent data. The findings demonstrate that while AI significantly enhances subsequent invention activity in climate-related technologies, a considerable portion of its impact extends beyond the climate sector. Future research should explore the factors that influence knowledge spillovers from AI in climate technologies, investigate the reasons behind the lower proportion of AI breakthroughs in some climate mitigation areas, and extend the analysis to other countries and regions to enhance the generalizability of the findings. Understanding the interplay between AI and climate innovation is crucial for developing effective policies that maximize the benefits of AI while mitigating potential risks.

The study's reliance on US patent data limits the generalizability of the findings to other countries. The study also acknowledges that not all inventions are patented, and the criteria for patentability may vary across countries and technology domains, potentially influencing the results. Furthermore, the focus on citation counts as a proxy for technological impact may not fully capture the complex realities of technology diffusion and societal benefit.