Sustainable AI and the path to responsible innovation

AI was already important to both businesses and society, but with the emergence of generative AI, its importance and impact have reached new heights. There’s no denying the huge benefits this technology has the potential to bring.  

AI innovations are already being applied across various industries, improving healthcare diagnostics, customer experience and fraud detection in banking, and even the fight against climate change. Yet, it’s no secret that AI itself is power-hungry. As demand for compute-heavy AI increases, so too will the carbon cost – particularly when it comes to more complex technologies like generative AI.  

In order to reduce AI’s environmental impact whilst still realising the benefits of this technology, organisations will need to think carefully about responsible AI innovation, and the paths to sustainability.  

Balancing AI benefits with carbon cost 

According to a recent McKinsey & Company report, business adoption of AI has more than doubled over the past five years. And with the introduction of generative AI to the public sphere with OpenAI’s AI art generator DALL-E 2, chatbot ChatGPT, and open-source development platforms like Hugging Face, it’s clear that the industry is only going to grow.  

AI provides many potential benefits to businesses – whether it’s automating routine tasks to elevate the thinking of human staff, or inspiring content prompts created by generative AI. And when it comes to larger societal concerns, like climate change, AI can help here too. Backed by machine learning, AI can help create more accurate climate predictions, simulate the effects of extreme weather, and track carbon emissions.  

And yet, despite all these possible benefits, there’s no denying that AI itself demands a huge amount of computing power – and therefore has a negative impact on the environment itself through carbon emissions. This is especially true of generative AI. The compute power used to train ChatGPT, for example, equates to enough electricity to power thousands of homes for a year, and the costs for electricity alone to train the next generation models are estimated to reach $400 million per year or more.  

Recognising this, just recently, the UK Government increased its funding for green AI initiatives to help achieve the country’s Net Zero goal by 2050.  

In order to reduce the enormous carbon cost associated with training and powering increasingly complex AI models, while keeping up with demand for AI and reaping the benefits of AI innovation, organisations will need to focus on efficiency and sustainability. 

Sustainable solutions to the AI conundrum 

To tackle AI’s carbon problem, organisations investing in this technology need to think carefully about where they house the high intensity compute required. Since this compute is often housed in data centres, an obvious solution is to prioritise those facilities that can handle the enormous compute required for AI technologies as efficiently and sustainably as possible.  

Location is key to the solution. Only about 10% of applications are truly latency-sensitive and therefore need to be located near to the end-user, often in highly-connected, but energy inefficient metro locations, like London. Crucially, this means that the vast majority – over 90% – of applications that are not latency-sensitive can be strategically located in data centres that have greater access to renewable energy sources. Training of large language models (LLMs) and other large generative AI models are perfectly aligned to the second category of latency-resilient applications. 

Iceland is a great example, as the only country in the world with 100% of its electricity powered by renewable (hydroelectric and geothermal) energy. What’s more, the temperate climate of Iceland and other Nordic countries means that data centres in these locations benefit from free air cooling -resulting in greatly improved energy efficiency considering cooling normally accounts for up to 40% of data centres’ energy consumption in locations with less favourable climates.  

Many data centres are also investing in sustainable solutions – such as alternative cooling methods like liquid cooling technologies – to boost their energy efficiency and reduce their environmental impact. If data centres and businesses continue to prioritise these solutions and invest in green data centre solutions, the environmental impact of compute-intensive technologies like AI will be greatly reduced.  

AI technologies will inspire a new generation of innovation. By making mindful decisions, we can steer these advancements to align with our collective duty to combat climate change.