Hacking AI for climate policy
In November, we teamed up with OpenAI for their first hackathon dedicated to climate change.
Its aspiration was to explore how AI – specifically language models – could help drive ambitious climate action, drawing parallels with what we do at Climate Policy Radar.
We threw data into the ring by building and sharing an open repository on GitHub of all the data we have in our searchable open database: over 3,000 documents covering climate laws and policies from every country.
Several teams designed AI solutions for climate policy, some using our database, ranging from analysis tools to enhancing document accessibility. We’ve rounded up the submissions that got our tech and data teams most excited about their potential applications. Read on to find out more about their ideas, and the avenues for exploration our team believes these could open up in the climate AI space.
Climate Policy Copilot
In a bid to enhance understanding and development of climate policies, Climate Policy Copilot built a variety of features to assist with analysing and writing climate policy and guidelines. They blended several approaches including fine-tuned models to write policy, and OpenAI’s GPT-3 capabilities such as automated summarisation of information and Q&A generation. It was designed to serve as a tool for both climate policymakers and the general public.
Climate Policy Copilot was created by Giulio Brugnaro, Patryk Neubauer, Nicolò Trentacoste, Jacob Nicotra and Tommaso Tassi.
Our take
Using large language models to help explore the policy space is a priority roadmap item for us – that the team was able to pull together such an interesting tool in such a short time really demonstrates their massive potential. This is a great illustration of how they could be used to combine and link previously siloed datasets from different sources (public opinion, policy, scientific research), which adds a powerful multiplier effect to insights generation.
Applying large language models to tailor language complexity is also a really neat idea. Opening up the accessibility of policy language, which might be complex and hard for computers to process, is a key value for us and this shows the power of large language models to help with that.
Climate Shell
Climate Shell is designed to help non-technical audiences quickly access open-source satellite and climate data, to support environmental decision-making “and pave the way for greener communities”. Their product takes a user’s query (e.g. how much tree cover has been lost in the Congo Republic?) and uses GPT-3 to automatically generate code to interrogate Google Earth Engine data, in addition to suggesting further research topics.
Climate Shell was created by Christina Last, Camilla Cardenas, Jannik Wiedenhaupt and Franklin Willemen.
Our take
How to connect policy commitments to environmental effects through satellite observations is very much on our team’s radar. Linking policy language with empirical data is fundamental to ensuring policy interventions are effective, and satellite images are one of the best routes we have to comprehensive globally consistent datasets. In the future, we’d also love to team up with Climate Trace and pair their emissions data with our database.
This submission is a great proof of concept showing how the very large and highly complex data source provided by Google Earth Engine could be made accessible to policymakers and other interested parties through a simple natural language interface. Search is also an imperfect way to explore data - guiding the user through their suggested query feature could therefore help them get the most out of their research.
CLAWD - the cloud for climate law
Team CLAWD believes “that looking through climate policies and getting an overview of a topic is a cumbersome task” – something that resonates deeply with us and our raison d’etre. Their solution allows users to search for policies based on keywords. They calculated similarities between documents, enabling comparisons. To make information more accessible, they augmented their app with several different functions using GPT-3, helping people better understand the complex legal and policy jargon often found in documents.
CLAWD was created by Aviral Chharia and Sabrina Herbst.
Our take
CLAWD’s ‘similar policies’ and policy comparison functions could be incredibly useful for policymakers and open up intuitive ways to explore policy evolution and policy diffusion, and eventually even allow understanding of why some policy combinations work and others less so. Adding the similarity graph provides a great way to guide the user’s journey to finding policy results they’re interested in.
Q&A and translation features all speak to use cases we’re thinking about at Climate Policy Radar and are really powerful tools to incorporate in the policy toolbox. And a hat tip to the team for using Climate Policy Radar’s open data.
Climate Policy Tracker
The team behind Climate Policy Tracker (distinct from another existing company with the same name) believes “that the most impactful changes are those made by entities like governments and organisations”. To improve transparency in the design of regulations, they created Climate Policy Tracker – a pipeline with an app for automatic processing of documents and analysis of policies within and between countries. Using OpenAI’s API, they developed features including automatic summarisation of documents, local explanations of topics, and interactive maps. Their aim is to empower citizen governance and engagement in climate policies.
Find out more on the Climate Policy Tracker website.
Climate Policy Tracker was created by Artur Żółkowski, Emilia Wiśnios, Mateusz Krzyziński, Piotr Wilczyński and Stanisław Giziński.
Our take
Empowering citizen governance is a really interesting user base. Making long and difficult-to-analyse documents more accessible to all is a key pillar in empowering society to play a more active role in the transition – supporting democratic agency – including voting, challenging policymakers, and making more informed choices about behaviour, consumption and investment.
Their radar chart showing values of indicators signalling policy commitments in different domains for each country was really powerful, and comparison within and between countries is a brilliant way to contextualise insight with data.
What’s next?
The hackathon has proved a real sandpit for innovation in climate policy AI, and we’re itching to play. The level of engagement and quality of submissions reinforces the power of open science, a commitment that runs through everything we do. As we look to build an open-source community that shares data and practices, we hope to work with many teams from the hackathon and beyond to better ideate, design and iterate tools that can support better climate decision-making.
If you have an idea for a project you’d like to share with us – whether that’s research, data, outreach or anything else that’s sparked your synapses – please get in touch, we’re open to collaboration.
Sign up for our newsletter to hear about what we’re working on and to find out about future opportunities for collaborations.
By Kalyan Dutia, Henry Franks, and Justine Alford