By Nick Dale, Senior Director, Verne Global
Society at large is becoming increasingly aware of its environmental impact, recently highlighted by the crystal-clear water in Venice’s canals and the Himalayas being visible across smog-free skies from over 120 miles away, amidst the global shutdown. This concern has also extended to the finance industry, with environmental, social, and governance (ESG) criteria rising in importance for both business and investment decisions.
At the same time, the financial sector has also been a major adopter of another significant trend – the use of AI and machine learning to improve efficiency and results. AI is particularly useful to the finance industry, from optimising asset portfolios and underwriting loans, to assessing risk and spotting fraud. In fact, the financial sector is leading the UK in the use of AI, with the great majority of banks, insurance firms and other financial institutions using such technologies.
Surprisingly, these two trends may be at cross purposes because of AI’s hefty carbon footprint. Training one deep learning model for natural language processing can emit more than 626,000 pounds of carbon dioxide equivalent, which is nearly 5 times greater than the amount generated during the entire lifetime of a car (including the manufacture of the car itself), per 2019 University of Massachusetts research.
But sustainable AI innovation is possible if financial services organisations begin to understand why AI can have a negative impact on the environment, and what they can do to minimise that impact.
AI’s appetite for power
The field of AI has been growing in leaps and bounds in the last decade, and no where more so than the finance industry. Financial institutions like Goldman Sachs, Morgan Stanley, and S&P Global are routinely using AI tools like Kensho’s for investment insight. Kensho’s algorithms can process 65 million question combinations, analysing over 90,000 world events – such as political events, economic reports, and monetary policy changes – and their impact on asset prices. Forbes reports that traders with access to Kensho’s AI-powered data were able to foresee a protracted drop in the British pound in the days after Brexit.
But as AI technology grows and develops, the computations behind it are also increasing in size and complexity. There has been a 300,000-fold increase in the computations required for deep learning research from 2012 to 2018, according to analysis conducted by the Allen Institute for AI. On top of that, these AI computing platforms can sometimes run 24-hours a day, necessitating days and even weeks of processing, plus trillions of attempts, to get the numbers lined up. As a result, these applications consume an enormous amount of energy in order to function, and require significant and constant access to power.
The carbon cost of AI becomes even greater when you factor in the energy required to keep computing equipment cool in order to prevent overheating that can impact performance and damage equipment. In a conventional data center, at least 40% of all energy consumed goes towards cooling.
Steps to minimise carbon impact of AI
Green – or greenwashing?
For businesses looking to reduce the environmental impact of their AI, the first step is to check the green credentials of the cloud providers and data centers that power these applications. Despite the “green” label, there’s no guarantee that a cloud provider or data center is powered entirely, or even partially, by green energy. Instead, these green claims can be more akin to a “carbon offset” programme, with energy providers offsetting the carbon they produce through tree planting or other similar programs.
Renewable energy sources
Instead of greenwashing, make sure that the data centers housing your AI compute are actually powered by renewable energy. In many Nordic countries, data centers are powered by renewable energy sources like hydroelectric and geothermal power. Iceland, in particular, uses 100% renewable energy with no nuclear power. These renewable energies are much less harmful to the environment because, unlike fossil fuels, they don’t cause pollution and don’t generate greenhouse gases. Not to mention, renewable energy is based on natural resources that can be replenished within an average human lifetime, as compared to fossil fuels, which can take thousands—or even millions—of years to replace.
Data center location
The next step is to look at the location of your data centers. Over 80% of compute doesn’t need to be near the end-user, and in those situations, choosing data center locations in cool climates has a significant impact on carbon emissions. In such cases, AI compute can be located in places like Iceland, which can utilise free-of-cost, natural cooling, due to its year-round cool, temperate climate.
This is in stark contrast to data centers located in hot climates, like Arizona in the US. With average high temperatures of 40° Celsius in the summer, data centers in climes like these need high-powered cooling systems in operation around the clock, often supported by up to 4 million gallons of water a day used to absorb heat through evaporation into cooling towers. As a result, when it doesn’t affect performance or accessibility, housing AI compute in data centers with natural cooling seems like an environmental no-brainer.
Better for the environment – and for business
As much as the financial sector is starting to embrace sustainability as a key ESG criteria in their corporate strategies, some may still view such efforts as an added cost to the expense side of the balance sheet. But the truth is, green AI presents financial services firms with an opportunity to align profit with purpose. By housing the servers that train AI models in data centers powered by renewable energy sources – connected to a reliable power grid –, businesses can substantially reduce energy expenses and benefit from predictable pricing.
As well, choosing locations with year-round, cool climates that allow natural cooling of powerful AI servers further minimises energy usage. When it comes to green AI, reducing environmental impact also lowers energy demands and costs – something that’s well worth the investment.
