AIs representing different parts of the world can hammer out a hundred years鈥� worth of climate negotiations in just a few seconds. The simulation is part of a competition launched last month to help find out which policies could have the best chances of success on the international stage.
With time running short for countries to agree on how to reach net zero greenhouse gas emissions by 2050, the 鈥� a collaboration consisting of researchers from the and different universities as well as companies like Salesforce and Google 鈥� has created a digital testing ground for climate policies and ideas.
The simulation is based on the research of , who received a Nobel prize in 2018 for his work on climate economics. To evaluate the impact of policies such as a carbon tax, Nordhaus created the Regional Integrated model of Climate and the Economy (RICE) that captures how factors such as technological change, CO2 emissions, global temperatures and the resulting economic damages affect each other.
Advertisement
The collaboration has now developed a new version of that model, called RICE-N, that adds negotiations to the mix. This allows fictional regions in the simulation to propose and form international agreements through multiple rounds of negotiations.
A human kicks off the simulation by using one or more AI-controlled regions to propose a potential climate agreement, such as climate clubs where members agree to set a minimum mitigation rate for reducing carbon emissions while enforcing a tariff against regions outside the club. The AIs representing 27 different fictional regions then negotiate by accepting or rejecting such agreements. After each negotiation phase, the simulation shows how the agreements shape each region鈥檚 economic decisions and how those economic activities impact the climate. The negotiations take place every five years throughout a simulation run equivalent to 100 years.
The AIs use machine learning to figure out what regional interactions lead to the best economic and climate outcomes for their region through trial and error, learning from each simulation run how to do better the next time.
鈥淚f you actually think about global decision-making, nobody makes decisions in a vacuum,鈥� says at Salesforce in Palo Alto, California, one of the companies involved with the project. 鈥淲hat you want is to understand the relationships and the communication between countries.鈥�
The competition organisers want teams to suggest the right policy incentives 鈥� such as a mix of trade rewards and tariff punishments 鈥� for the model鈥檚 self-interested AI agents to cooperate for the good of everyone. 鈥淣ational short-term interests are the major stumbling block for achieving ambitious climate agreements,鈥� says , a political scientist at the Center for International Studies in France who wasn鈥檛 involved in the competition.
All proposals will be evaluated through RICE-N simulation scores that reflect how well the world has done from both a climate and economic standpoint. Top-scoring policies will earn additional review by a human jury that also considers whether the policies are ethical and can be implemented in the real world. By the time the competition wraps up in April 2023, the plan is to publish a paper that could help make policy suggestions to policymakers engaged in real climate negotiations.
鈥淗opefully RICE-N can bring us closer to something that allows us to have a better understanding of the behavioural [or] human dimensions of climate change cooperation,鈥� says at Georgia State University, who is not among the project organisers.
The big question is whether a winning climate policy in the simulation would be similarly successful in the real world. 鈥淭he simulation will necessarily be much more simple than reality,鈥� says Kloeck.
Reference:
