MATTER Magazine: AI & Climate Change
Independent research institution Rhodium Group released a report this year stating that carbon dioxide emissions in the U.S rose by 3.4 percent from 2017 to 2018—the largest increase in the last two decades.
For the past few years, global warming and climate change issues have been at the forefront of various discussions. The Rhodium Group’s report, along with other climate change studies, reveal the long-term impacts increased carbon emissions will cause.
An article written by Kelly Levin and Denis Tirpak (both from the World Resources Institution) reviewed the major events of climate extremes that occurred in 2018. It particularly noted the global increase of temperatures: “In June, Oman saw its highest minimum temperature at 108.7°F, a new Asian record … Meanwhile, Japan’s summer heat wave resulted in 22,000 people hospitalized with heat stroke.”
Other negative effects of climate change listed in the article include changes in precipitation patterns, greater intensity of natural disasters and storms and the decreasing volume of polar ice caps.
In order to combat global warming and prevent further environmental damage, some communities are instituting “green” initiatives by refining composting methods and changing minor aspects of daily lifestyles, such as choice of transportation. For example, Project Drawdown, a global research organization which reviews viable solutions to climate change, and Paul Hawken partnered to organize a list of eco-friendly solutions for citizens that can be used to bolster awareness for the environment.
Are there other more technological or productive ways to tackle such climate problems? In the past decade, many researchers have started looking into machine learning for further solutions. To understand machine learning, though, one must first learn about the development of artificial intelligence (AI).
AI can be categorized into four large classification groups: supervised learning, in which the program model receives a substantial dataset containing pre-classified input-output data pairs to train and notice patterns; unsupervised learning; semi-supervised learning and reinforcement learning. Coding languages like TensorFlow and PyTorch are all used to create machine learning models, and platforms such as Google Classroom are widely used for small-scale projects requiring low inference time as well. As the understanding of AI and its techniques develop rapidly, it is becoming implemented in various fields.
Scientists use AI in a variety of fields. DXplain, developed at the Laboratory of Computer Science at the Massachusetts General Hospital, is a decision support system which assesses a set of symptoms and produces treatment suggestions for the clients. AI is also used in computer science, where it constitutes the framework and inner workings of competitive gaming programs. Often times, different areas of AI intersect to create scientific breakthroughs like AlphaZero, a high-skill trained machine learning model. The program is a hybrid of reinforcement learning, unsupervised learning and decision-tree processes that can intelligently play chess, shoji and Go.
Machine learning plays a crucial role in the battle against climate change. It can make more accurate climate predictions so governments can set climate policies or goals accordingly based on relatively accurate projections. AI is the mark of a new age of human innovation, and it may be part of the key to resolving the environmental problem, or at least take a part in mitigating the effects of climate change.
Examples include:
1. SilviaTerra: This program uses AI and satellite imagery to observe and report the size, species and health of trees and has greatly increased the demand for manual searches through forests.
2. Green Horizon Project: Run by IBM, this program creates pollution forecasts. It proved helpful as it led Beijing to decrease average smog by 35 percent (2012-2017).
3. CycleGans: “GAN,” or Generative Adversarial Network, is a network that generates statistics, personas or information without a need for other additional inputs. This is advantageous for collecting greater amounts of data for the data graphs.