Lowering AI’s environmental impact

Andy Whitehurst at Sopra Steria UK argues that quantum computing may be the silver bullet to lowering the environmental impact of artificial intelligence

Artificial intelligence (AI) is transforming the world we live in. From automating workloads to generating code, the technology is being rapidly adopted across workplaces and wider society.

Yet, while the benefits and potential challenges of AI are becoming known, it’s impact on the environment is often forgotten. This is despite the technology generating a significant carbon footprint due to the sheer amount of data it needs to collect and consume during development, training and day-to-day operation.  Those ‘operational’ impacts are then followed by the inevitable environmental impacts of decommissioning the hardware required to run AI.

As AI adoption increases, and the UK races towards net zero by 2050, we need to find a way of reducing its environmental impact. One potential solution lies in quantum computing – a type of technology that uses quantum mechanics to store and process vast quantities of data. But how exactly can quantum computing solve this challenge, and how can we balance this need without impacting future AI adoption?

The impact of AI on the planet

Both the training and operation of AI has an enormous impact on the environment. First and foremost, models such as ChatGPT are trained using vast amounts of data which consumes a significant amount of energy. For example, GPT-4 has been shown to consume between 51,773 MWh and 62,319 MWh of energy during the training phase – equivalent to the amount consumed by 1000 average homes over a 5 year-period. This high energy usage ultimately means they emit a large volume of greenhouse gas emissions, which negatively impact the environment.

This energy consumption also generates vast quantities of heat, meaning cooling is required to lower the temperature of the surrounding environment (normally a data centre) and dissipate this heat into the outdoors. This can require a large amount of water, with some generative models consuming 500ml of water – the equivalent of an entire water bottle – every time a user interacts with it. One GPT-4 model currently undergoing training in Iowa, US, recently consumed almost a tenth of the districts entire water supply – just during the initial training phase. 

Like any other technology, the underlying hardware components AI needs to operate may become damaged or fail over time and need replacing. As these components are often manufactured using heavy metals, such as lead and mercury, they produce a large amount of electronic waste if not disposed of or recycled properly, so have a negative impact on the planet as a result.

That’s not to say AI can’t also help in tackling some of those same environmental challenges facing us today. Alongside revolutionising how businesses function, the technology is already being used to monitor and measure environmental impact around the world – from tracking the carbon footprint of products and services, to measuring indoor and outdoor air quality. It’s therefore crucial we harness its value whilst also finding a way to lower its environmental impact, without affecting its adoption. But how?

The role quantum computing can play

While still in its infancy, quantum computing could be the answer to lowering the impact of AI on the planet. One way this can be done is by reducing the amount of energy AI consumes during its operation. For example, quantum computing, or quantum-inspired computing techniques, can enhance the memory capabilities of the neural networks used in AI – effectively making AI more efficient. This could significantly reduce the amount of energy consumed during the technology’s operation compared to that of a traditional computer.

Once quantum computing becomes more developed, researchers may even use qubits, a unit of computation, to replace these networks, improving the efficiency of AI even further.

Most quantum computers use cryogenic refrigerators (designed to reach temperatures of around -153°C) to operate, meaning much of the energy used is directed to the refrigerator needed to operate it. However, as the operating temperature nears absolute zero the processing becomes superconducting, allowing the computer to process information using almost no power and no heat. This requires dramatically less energy than if AI was to run on traditional computing hardware.

Quantum computers are also generally not manufactured from heavy metals such as lead and mercury, meaning they have a significantly lower manufacturing footprint in comparison to more traditional computing hardware. This means they also produce less electronic waste during their lifetime as a result, helping to reduce the environmental impact of AI if run on a quantum computer or via quantum-inspired techniques.

As AI adoption speeds up across the world we mustn’t forget the enormous impact the technology is having, and will continue to have, on the environment. By lowering the amount of energy consumed and waste produced throughout an AI model’s lifetime, quantum computing could one day make this adoption greener as we look towards net zero.

Andy Whitehurst is Chief Technology Officer at Sopra Steria UK

Main image courtesy of iStockPhoto.com and weerapatkiatdumrong