The internet is the world’s largest machine, but how many of us think twice about using it? If current trends continue, the ICT sector stands to be responsible for 14% of the world’s CO2 emissions by 2040. To try and be more mindful about our AI usage, we should first try to understand the impact it has on the environment, in a positive and a negative capacity.
Finding the delineation between the impact of AI in a process and the impact of standard computer use is difficult because the definition of AI is not widely agreed upon, but for the purposes of this document and for these recommendations, it is taken as:
Artificial intelligence is at the intersection of machine learning and rules-based programming. If a process does not involve a machine learning component, it should not be considered a form of AI.
A number of different machine learning techniques exist and are being developed, but all require 3 significant, measurable inputs:
Electrical energy to power devices and processes (measured in Watts)
Large volumes of data, stored or otherwise, and its transmission (measured in Bytes)
Processing power in the form of hardware (measured in Hertz)
Electrical energy consumption is a common and easily measurable metric. Other than for generic efficiency purposes, limiting power usage may not be practicable for businesses. Choosing an energy provider that offsets their carbon footprint, or supply energy from renewable sources mitigates the company’s responsibility to measure and offset their own power usage.
Most data transmission uses the internet, this infrastructure is not controlled by an individual or organisation and is therefore out of scope although if the internet could be considered a country, it would be the 7th largest polluter. What a company can control is how much data is transmitted using the internet, and how the internet is accessed. Web pages are trending larger and are visited more frequently which requires more energy and processing power from a device, with the largest draw being video traffic. Companies can use tools to optimise their web pages for performance, CO2 footprint, and sustainability. It is worth noting that cellular networks use considerably more energy to move the equivalent amount of data as wired or wi-fi technology. Where possible, reducing the reliance on cellular internet for business operations will reduce the overall carbon footprint of the company.
Processing power is the ability of a computer to manipulate data. Generally, the faster a computer processes information, the more efficiently it operates. Most emissions from end-user hardware come from production, not use. Semiconductors are essential components of many electronics, their production is highly resource-intensive.
So how can I be more mindful about my AI usage?
For those of us with low or normal processing power requirements, where common commercial “off the shelf” computer hardware is sufficient, the environmental considerations of AI usage need not be separated from the environmental considerations of standard computer usage.
Suppliers of commercial computer equipment often display the carbon footprint or sustainability of their product for consumer consideration and carbon accounting.
Cloud providers (who provide consolidated computational data storage and processing off-site) like Microsoft Azure and Google provide some of the “greenest” cloud services available running entirely on renewable energy and some committing to net negative CO2 emissions. Choosing green cloud providers mitigates the requirement for us to individually monitor and offset our processing activities.
For those of us with high processing power requirements, where we might require a data centre or bespoke/ advanced computational hardware, we should look more closely at:
Cradle to grave hardware management - choosing modular systems for efficient upgrades, responsible recycling or disposal of electronic equipment.
Clean energy consumption - to meet all computational and ICT power demands.
Efficient infrastructure to house hardware - responsible use of air conditioning.
By making mindful decisions about purchasing and using environmentally “greener” ICT products and energy infrastructure, our resultant AI impact is minimised and offset.
There are so many positive applications of computational power, particularly operating at the intersection of AI, biodiversity, and climate change, not to mention the applications of technology that mitigate traditional emissions sources like video conferencing, forums, and online collaborations. Let’s continue to ensure technology is part of the answer, not part of the problem.
Author: Kate Turner, Carbon Offset Research Lead, AI for Climate Initiative
All views expressed on this blog post are the author's and do not necessarily reflect or represent the opinions of the AI for Climate Initiative or its founders.