We need global cooperation on climate change
Climate change is happening fast. The latest IPCC report warns that it is ‘now or never’ if the world is to stave off climate disaster. However, it is still a race we can win, capping the global temperature increase at 2 degrees Celsius.
To mitigate climate change, we need comprehensive long-term global cooperation. This poses a complex game-theoretic problem. There is no central entity that forces regions to adhere to climate agreements, while regions have individual policy objectives that are often misaligned.
Will your solutions lead to better climate outcomes?
Design multilateral negotiation protocols and agreements that incentivize cooperation on climate change.
Test your solutions in RICE-N: a climate-economic simulation with AI agents that has been calibrated to real-world data.
The simulation features investing, mitigation actions, international trade, tariffs, negotiation, agreements, and more!
To learn more, check out how to get started!
How can you contribute?
Building solutions for climate change is an interdisciplinary challenge: involving machine learning, economics, agent-based modeling, game theory, political science, behavioral science, mathematics, and other disciplines.
Join our working group and competition to contribute your expertise!
AI researchers, economists, climate scientists, behavioral scientists, and others
Good policy recommendations require rigorous and grounded technical work. Here are just a few examples of how you can contribute!
- Implement negotiation protocols and climate agreements.
- Extend the RICE-N climate-economic simulation to include more economic or climate features that may be necessary.
- Model real-world agents using AI and domain knowledge.
- Develop machine learning algorithms to enable rigorous experiments and analysis.
- Visualize and analyze the outcomes under your proposed solution.
- Work together with domain experts to understand what real-world requirements are important for your analysis and proposal.
Ethics, legal, and policy experts
Good science needs to be translated into good policy. Domain expertise (outside of AI, economics, or climate science) is crucial to help shape the analysis and communication of the results.
- AI-driven policy analysis is an open research challenge. For example, what does "good" mean from a non-technical perspective?
- What requirements are there for AI-driven policy analysis? For instance, regarding transparency, explainability, robustness, ethics, fairness, legality, precedence, and other dimensions?
- Communicating outcomes to governments is just as necessary as the research itself. You could contribute to clear communications as well as defining actionable insights and targeted recommendations.
Why participate?
First of all, come learn and innovate! We want to motivate you to do original research in this scientific area. Our competition is a way to test and compare your ideas. Our jury will also provide feedback on your solutions and ideas.
We will organize a workshop in December 2022. Top performing teams will be invited to present their work there. We also invite all teams to submit a workshop-style paper to document their findings, which will be peer-reviewed and published in the competition proceedings. Check out our schedule to learn more!
We plan to write a research paper based on the findings in the competition, coauthored by the working group. We will invite teams as coauthors whose findings are of sufficient scientific or policymaking novelty. This work will be submitted to a peer-reviewed journal. This work will also be reviewed for ethical use.
Given sufficient findings, we intend to write a policy brief with actionable insights for policymakers. This brief will be distributed and promoted through our partners. We also plan to organize a marketing campaign, e.g., through blogs and press releases around the findings of the competition.
Get started
- Register yourself: Sign up here (Google form, recommended) or here (Microsoft form).
- Ask any questions on Slack. Read our blog for a high-level overview.
- Join our Google Group to get email announcements! For Chinese users, there is an option to connect to our Official WeChat Account: AI4ClimateCoop.
- Find team members. To help, check out our Slack or check the list of participants here.
- Register your team: using this Google form (recommended) or alternatively using this Microsoft form. We will add you to the submission system so you can start submitting solutions. Please register a team only once, even if you plan to submit to multiple tracks.
- If your team has more than 5 team members, we will ask you to describe the expertise and background of each team member. We will refuse teams with more than 5 people that come from a single discipline only. We ask teams to consider reasonable interdisciplinary team compositions; the organization reserves the right to evaluate each case independently.
- Get the code and tutorial. Our Github repo provides simulation and reinforcement learning code, tutorials, calibration details, evaluation and submission code, and more.
- Read the technical white paper that explains the structure of the simulation, the mathematical background, and calibration details.
- Read the code documentation and submission preparation instructions.
- Submit your solutions using the instructions!
- Schedule time for Q&A at our office hours.
Leaderboard
Schedule
| Date | Time | Link | |
|---|---|---|---|
| Talks | TBD | Virtual (link TBC) | |
| Office hours | September 1 - February 15, 2022 | Mondays 9 am PST | Virtual (link TBC) |
| Last date for registration and submissions | February 15, 2023 | Anywhere-on-Earth | |
| Peer review | February 16 - March 9, 2023 | ||
| Closing event (invited talks, team presentations, workshop papers, future work) | March 2023, date TBD | Virtual (link TBC) |
Tracks
| Track 1: Score-based | Track 2: Score and real-world relevance | Track 3: Critiques and improvements | |
|---|---|---|---|
| What do you do? |
|
In this track, you will argue why your solution is practically relevant and usable in the real world. As we aim to bring the insights from the competition to policymakers, we expect the entries in this track to contain a high-level summary for policymakers. |
We strive to closely simulate real-world dynamics. But, no simulator is perfect. Thus, we invite you to point out potential improvements and loopholes. |
| What do you submit? |
Each submission should have the following:
|
In addition to the requirements in Track 1, you should also submit a written summary, justification, and explanation of your solution and insights. Your write-up should argue why your solution is feasible, technically sound, and attractive. For instance, from a game-theoretical perspective, good multilateral agreements might punish free-riders and might have to ensure that it’s difficult to “game” agreements with unrealistic behaviours. Please see the submission guidelines for a template submission with suggestions for aspects to discuss. |
This is a free-form submission: you may include a write-up, example code, or other ways to demonstrate your insights. |
| How do we evaluate? |
Each team's solution is scored by computing the hypervolume enclosed by your 10 most recent solutions. This is an lower-bound approximation of the area under the Pareto curve defined by your submitted solutions. Participants in Track 1 are not required to submit a technical report. However, we will invite the top scorers in Track 1 to write a technical report to be published in the Proceedings. |
An expert jury will review your submission and use a scoring rubric to evaluate your solution. They will also assess the real-world relevancy and impact of your proposed solution. |
An expert jury will review your submission and evaluate the significance of your suggested improvements and/or discoveries. |
| Guidelines |
Submission guidelines, evaluation rubrics, and suggested discussion topics. |
Submission guidelines and suggested topics to investigate. | |
| Submit here! |
| Track 1: Score-based | |
|---|---|
| What do you do? |
|
| What do you submit? |
Each submission should have the following:
|
| How do we evaluate? |
Each team's solution is scored by computing the hypervolume enclosed by your 10 most recent solutions. This is an lower-bound approximation of the area under the Pareto curve defined by your submitted solutions. Participants in Track 1 are not required to submit a technical report. However, we will invite the top scorers in Track 1 to write a technical report to be published in the Proceedings. |
| Guidelines | |
| Submit here! |
| Track 2: Score and real-world relevance | |
|---|---|
| What do you do? |
In this track, you will argue why your solution is practically relevant and usable in the real world. As we aim to bring the insights from the competition to policymakers, we expect the entries in this track to contain a high-level summary for policymakers. |
| What do you submit? |
In addition to the requirements in Track 1, you should also submit a written summary, justification, and explanation of your solution and insights. Your write-up should argue why your solution is feasible, technically sound, and attractive. For instance, from a game-theoretical perspective, good multilateral agreements might punish free-riders and might have to ensure that it’s difficult to “game” agreements with unrealistic behaviours. Please see the submission guidelines for a template submission with suggestions for aspects to discuss. |
| How do we evaluate? |
An expert jury will review your submission and use a scoring rubric to evaluate your solution. They will also assess the real-world relevancy and impact of your proposed solution. |
| Guidelines |
Submission guidelines, evaluation rubrics, and suggested discussion topics. |
| Submit here! |
| Track 3: Critiques and improvements | |
|---|---|
| What do you do? |
We strive to closely simulate real-world dynamics. But, no simulator is perfect. Thus, we invite you to point out potential improvements and loopholes. |
| What do you submit? |
This is a free-form submission: you may include a write-up, example code, or other ways to demonstrate your insights. |
| How do we evaluate? | An expert jury will review your submission and evaluate the significance of your suggested improvements and/or discoveries. |
| Guidelines | Submission guidelines and suggested topics to investigate. |
| Submit here! |
Office hours
Schedule a virtual meeting with us for Q&A, technical support, and more!
FAQ
How do I customize the simulation and submit solutions?
Check out the technical FAQ on Github.
Where can I ask questions?
We will announce news and updates in two ways: our Google Group and Slack (channel: #ai-for-global-climate-cooperation-competition). For Chinese users, it is optional to
connect to our Official WeChat Account and leave comments here:
How can I find other people interested in the competition?
Find team members on Slack in #ai-for-global-climate-cooperation-competition-team-search!
Ethics
While the intention of this paper and the corresponding challenge is to stimulate innovative solutions to climate change, there are some unintended consequences that we would like to acknowledge and address here. These include the carbon footprint of running the simulation itself, the economic disparities that can exist with climate negotiations, and the potential extensibility of this simulation to the real world.
First, it is important to acknowledge that running climate change simulations in RICE-N will inevitably release carbon emissions into our atmosphere. While the computational requirements of these simulations are much smaller than training large language models, they still exist. To mitigate this harm, we are encouraging participating teams to consider their energy use during experimentation, offsetting their own carbon emissions if possible.
Second, as the World Bank states "Climate change is deeply intertwined with global patterns of inequality" and yet "the most vulnerable are often also disproportionately impacted by measures to address climate change". While it is important to determine ways to mitigate climate change it is equally important to ensure that vulnerable populations are not negatively impacted by climate change measures.
Last but not least, it is important to note that the climate and economic predictions made in RICE-N may differ in a real-world setting due to externalities beyond the boundaries of the simulation. A fictional world is utilized in this competition to further illustrate the potential gap between simulation and reality, but the uncertainty of the results should be fully understood, especially before implementing any policies recommended by RICE-N.
Jury
Gillian Hadfield
University of Toronto
Professor of Law and Professor of Strategic Management
Director, Schwartz Reisman Institute for Technology and Society
David Parkes
Harvard University
Co-Director of the Harvard Data Science Initiative
Co-Chair of the FAS Master of Science in Data Science and Harvard Business Analytics Program
Lynn Kaack
Hertie School
Nicholas Muller
Carnegie Mellon University
Economics, Engineering, and Public Policy, Engineering and Public Policy, Tepper School of Business
Nicholas Kumbleben
Greenmantle
Eyck Freymann
Greenmantle, Harvard Kennedy School
Arctic Initiative, Belfer Center
Maija Halonen-Akatwijuka
University of Bristol
Andrea Tachetti
DeepMind
Romuald Elie
DeepMind
Hector Alvarado
The Decision Lab
Mike Tatigian
The Decision Lab
Advisors
Vincent Conitzer
Duke University
David Victor
UC San Diego
Richard Socher
You.com
Former adjunct professor
Stanford Computer Science
Former Chief Scientist, Salesforce
Christian Schroeder de Witt
University of Oxford
Organizers
Tianyu Zhang
Université de Montréal, MILA
Andrew Williams
Université de Montréal, MILA
Soham Phade
Salesforce
Sunil Srinivasa
Yang Zhang
Bank of Canada, MILA
Prateek Gupta
University of Oxford, The Alan Turing Institute
Yoshua Bengio
MILA
Scientific Director of Mila and IVADO
Partners
Acknowledgements
Qianyi Cheng
HfG Offenbach
