Can AI play the game of slowing climate change?

Facebook IconFacebook IconFacebook Icon

Society requires energy to function. We can think of energy as the calories of our economies, and we can frame climate change mitigation as a diet that we must embark on. Like any diet, there are three things we can do to lose weight. Firstly, we can eat healthier meals, or get our energy from renewable energy resources like wind or solar. Secondly, we can eat less food, or use less energy in our lives. And finally, we can keep eating the same rubbish food in similar amounts, and just exercise a few hours a day to try and burn off the calories. This is analogous to carbon capture and storage.

So on the most challenging “eating less”, how can Reinforce Learning (RL) help us? What’s the magic? Well, imagine that instead of playing Go (a complex board game), we play the game of climate change mitigation. Imagine we turn the electricity system into a game. The board is the electricity grid, our pieces are energy-intensive devices on the grid, and our goal is to minimize the emissions produced by these devices whilst maintaining a quality of service to society. 

RL has the capacity to seek new and creative ways to perform its tasks. Not only AlphaGo, the DeepMind agent that beat the world no.1 player Lee Sedol at the game of Go. But it also proved to unearth knowledge about the game that humans hadn’t discovered after thousands of years of play, for example with this now infamous move 37. Expert Go players suggest that in the board setup, human Go players have long believed that the optimal move is in the margins of the board, but AlphaGo decided to play 5 lines in from the edge; much to the surprise of the audience.

Of course, it’s not that simple. The setting of the two games is very different in terms of size and complexity, and the analogy is weak; model-free vs. model-based, continuous actions vs. discrete actions, offline learning vs. online learning, etc. But it’s definitely where we are headed.

At Sagacify, as a first step in this direction, we are thinking about solutions powered by RL to reduce energy consumption, especially for heating/cooling spaces and powering furnaces in manufacturing facilities.

In a next post, we’ll explain how to build a solution to save up to 25-40% on HVAC energy consumption with RL.

Main image source: artificial intelligence, climate change, board game, solar panels, wind turbine, utopia, photorealistic, cinematic (MidJouney)