Understanding the environmental footprint of AI

Understanding the environmental footprint of AI

AI applications enable sustainability solutions. For example, Destination Earth uses AI to create a ‘digital twin’ of climate change modelling, while DeepMind’s AI reinforcement learning algorithm reduces data center energy use by up to 40%. In a scenario where computational methods are growing while sustainability concerns prevail, measuring and understanding the impact of AI computing and applications is important.

OECD organized a panel discussion that explores these issues as presented in the report “Measuring the environmental impacts of AI compute and applications: the AI ​​footprint”. Informed by experts from the OECD.AI Expert Group on AI Compute and Climate and the Global Partnership on AI (GPAI) Responsible AI Working Group (RAI), the report examines existing measurement tools and critical challenges for quantifying the positive and negative environmental impacts of training and deploying AI models and applications.

By creating and tracking AI-specific computing measures, sharing best practices, and supporting new and innovative AI applications for fighting climate change, countries can ensure that AI is trained and deployed in the most sustainable way possible while minimizing negative impacts for the good of the planet.

The panelists included Ronny Rodriguez Chaves, Ministry of Environment and Energy in Costa Rica; Keith Strier, NVIDIA, OECD.AI; Lee Tiedrich, Duke University, GPAI; George Kamiya, Expert on Climate Impacts of Digital Technologies, International Energy Agency (former); Peter Martey Addo, PhD, French Development Agency; Lee Tiedrich, Duke University School of Law, GPAI. Stephanie Ifayemi moderated the panel.

Why measurements?

AI is present from production till the end of life. The report stated that ensuring AI is part of the solution to meeting global sustainability targets. It requires measurement standards for sustainable AI, expansion of data collection on environmental impacts of AI computing and applications, separation of AI-specific measures from general-purpose computing, considering environmental impacts beyond energy use and GHG emission and efforts to improve environmental transparency and equity .

AI- involved activities have a socio-economic impact- direct and indirect. Measuring the applications will aid us in maximizing the benefits and mitigating the risks. However, the wrongful use of AI can lead to harmful effects such as increased exploitation of Fossil Fuels.

Lack of planning on investment in AI computing will reduce redness and resilience in an economy largely dependent on AI networks. The nature of computing is ever-changing. Moreover, the affordability of computing techniques has risen over the years. However, there still exists a divide between countries worldwide between the haves and have-nots.

It is also important to have AI-specific measurements. Policymakers should understand AI systems. Capacity building on tech and environmental impact could provide policymakers will be beneficial in the long run.

Equity in the global south

There are several issues in the emerging economies of the global south. However, there is a rise in Machine Learning and Deep Learning community. At the same time, they experience policy-related issues. Therefore, there is an enormous scope in the global south to invest in computing infrastructure, which will bridge the gap in the long run.

Several biodiversity hotspots are located in the global south, suffering tremendous loss, with climate change being one of the major reasons. Therefore, a more concrete understanding of the issues of the south. The nations should work together collaboratively to improve their capacity. At the same time, it is important to improve data stewardship.

Overcoming challenges

Citizens should be brought to the table during the process of policymaking. It will open the door to collective intelligence. Also, there should be an effort from companies, ML practitioners and researchers. Finally, governments should take measures to revisit policies.

A nation’s ability to plan, anticipate, respond and recover from the impact of climate change will increasingly depend on the domestic ability to develop new technologies such as AI-based prediction and simulation models.


Leave a Comment

Your email address will not be published. Required fields are marked *