2021 is shaping up to be a good year for artificial intelligence. According to Deloitte, more than 70% of organisations intend to increase their spending on AI solutions this year in pursuit of improved analytics and automation capabilities.
And, as UiPath’s $34bn IPO shows, European startups that provide this tech are already in huge demand.
But while AI will be hugely beneficial for businesses — enabling them to increase efficiencies and make more insightful, data-driven decisions — it also comes with a significant cost: a large and evergrowing hunger for processing power and the corresponding carbon footprint that it brings with it.
The challenge for AI startups, and those looking to procure their services, is to strike a balance between providing business with the benefits of this tech and its long term impact on the environment. If they can’t do this, AI could become the 21st century equivalent of plastic packaging — sold to us as a way to protect, transport and distribute products at scale and with an efficiency that wasn’t possible before, but which has since become a scourge on our planet.
Blinded by the benefits
The numbers revealing the scale of AI’s environmental impact are only just becoming clear. A recent report from the UK scientific academy The Royal Society notes that “digital technologies [such as machine learning and AI] have an environmental cost that should not be neglected”.
Meanwhile, the European Union warns that greenhouse gas emissions attributed to the IT industry — currently 2% — could rise sevenfold to 14% within the next 20 years. This trend is supported by data published by OpenAI, which shows that the computing power that has been required for key AI landmarks over recent years, such as DeepMind’s Go-playing programme AlphaZero, has doubled roughly every 3.4 months — a 300k-fold increase between 2012 and 2018.
There have been some high-profile cases hitting the headlines about excessive carbon consumption during the training and deployment of AI algorithms. One model that enabled computers to make sense of text had the same carbon footprint as 125 round trips between New York and Beijing.
Startups, developers and academics are starting to find ways to reduce the carbon emissions of AI.
Researchers at UC Berkeley and Google have written about how this can be done. There are three things that can be controlled: the design of your algorithms, the hardware on which you train them, and the availability of green electricity to the data centres in which they run. Large companies such as Google may have the luxury of being able to control all three of these key features, but your average AI-adopting company probably does not. It is therefore imperative that those developing the AI make sure they consider sustainability throughout the design and training process.
The green transition is underway — meet the startups driving it.
AI can also be put to use in green projects, as many companies already do. This ranges from the use of AI in manufacturing to reduce waste and energy use, to companies like UK startup Space Intelligence, who are applying machine learning and AI to satellite data to address environmental concerns such as reforestation and biodiversity conservation so that industries can take corrective steps.
In a 2019 research paper, the Allen Institute for AI in Seattle outlined an approach aimed at making AI “both greener and more inclusive”. Achieving this aim requires humans to work together with AI technology to predict, monitor and mitigate the carbon footprint of any particular AI project.
We are at the point where business leaders must take a hands-on approach in making sure the AI they adopt has taken these considerations into account — and those developing these technologies must be able to deliver on those concerns. An understanding of the costs of building and deploying AI models both financially and in terms of environmental impact must be developed across the board.
These issues can be addressed today. If we measure the success of an AI system in terms of its overall impact, then decisions can be made to select the correct levels of complexity for a particular problem, which can lead to a significant reduction in energy consumption, cost and time.
We also need to restructure the conversation about AI performance, and what that looks like in a net zero world. Business leaders are already accustomed to making tradeoffs between more ‘traditional’ functions — such as human resources versus budget, or capital expenditure versus investment — so now they need to apply the same logic to tradeoffs inherent in data systems. Carbon implications are an important part of those considerations.
The responsible use of AI will enable us to embrace the vast potential that it offers, without the high computational cost proving counterproductive in the fight against climate change.
It’s crucial, therefore, that we all develop a ‘Green AI’ approach, and ensure that AI is developed transparently and responsibly — and, equally as important, sustainably — so we’re not chasing our tails in years to come as we are today with the plastics problem.