Will AI help the zero carbon agenda?

By Dan Llewellyn, Director of Technology at xDesign.

  • 1 month ago Posted in

There is no silver bullet to reaching the large and complex problem that is net zero. However, artificial intelligence (AI) is helping companies to advance the zero carbon agenda in new and innovative ways. These range from novel data mining techniques that generate accurate weather forecasts used for better managing the variability in energy production, to training neural networks that help scientists simulate, predict and optimise carbon storage.

But, it is becoming increasingly clear that AI is a double-edged sword in the battle to reduce carbon emissions. The energy requirements for AI are rising consistently, especially as generative technologies like ChatGPT become more popular.

As such, the very computational power we are harnessing day in day out is increasingly coming at a cost to the planet in the carbon stakes. It is crucial that we measure this impact to manage it effectively, aiming to strike a balance between its benefits and environmental costs.

The problem with AI

There is clearly a need to reduce this power consumption. Recent research found that training a new GPT-2 model generated 300,000kg of CO2 - the equivalent of 125 round-trip flights between New York and Beijing. Using GPT for very basic day-to-day requirements also threatens to make a notable impact too, with simple mathematical calculations being run through it creating much higher emissions than using the most basic desktop calculator for the task.

Whilst we can estimate the cost of training a Large Language Model (LLM), the true costs are closely guarded secrets - and with most proprietary LLMs, it is expanding all the time. In 2019, OpenAI released a paper revealing that the compute power used in various large AI training models had been doubling every 3.4 months since 2012. With a race for supremacy in AI already in full force, what will the carbon cost be in the decades to come?

A need for measurement

And therein lies one of the biggest problems we face with AI. As the famous management consultant Peter Drucker once wrote, “you can't manage what you can't measure”. To ensure we’re utilising AI to the best of its ability to fight the climate crisis, whilst limiting its impact, we must start with robust measurement. From here, we can build the frameworks, standards and internationally recognised policy required to stimulate a consolidated global response.

Opening the black box to calculate AI’s true impact on the planet will require an industry-wide collaboration effort. In light of the complex relationship between AI and carbon emissions, it is clear that expecting companies to self-report their carbon footprint is not a viable solution. Given the secretive nature surrounding the energy consumption and carbon emissions of AI technologies, a more rigorous and transparent approach is necessary.

Setting up mandatory bodies tasked with overseeing that companies carry out standardised reporting could be one way of accurately assessing and managing the environmental impact of AI. By implementing these measures, we can better balance innovation with environmental responsibility, ensuring that our technological progress does not come at a cost to the planet's health.

Providing transparency

Hugging Face is one AI startup bucking the trend by measuring its own environmental impact. After calculating the emissions of its large language model BLOOM, the company’s researchers found that the training process emitted 25,000kg of carbon, which doubled when they took the wider hardware and infrastructure costs of running the model into account.

The primary reason for this is its size - GPT-4 contains 1.76 trillion parameters compared with BLOOM’s 176 Billion. But, given a fairly staggering difference in the cost of training, would we be better off reserving use of the most expensive tools for the most complex problems?

Another reason for BLOOM’s relatively small carbon footprint is that it was trained on a supercomputer powered by nuclear energy - drastically reducing the use of fossil fuel in its production. For consumers wanting to make a positive choice, as they already do with everything from toothpaste to clothing choices, they need that information to be available. 

Preparing for the future

If we succeed in getting the measurement part right, we will better position ourselves to take advantage of AI’s future breakthroughs. From defeating grandmasters in chess to identifying antibiotics against drug-resistant bacteria, AI has demonstrated its potential to devise solutions that stretch beyond human ingenuity. Leveraging its knack for innovation could be key to reducing carbon emissions as it devises solutions beyond current human understanding.

As Drucker hinted, by accurately assessing the problem's magnitude, we can then devise an effective response when that time comes. As such, by focusing on measuring AI’s impact on the planet today, we will give ourselves the best chance of seeing tomorrow.

By Rosemary Thomas, Senior Technical Researcher, AI Labs, Version 1.
By Ian Wood, Senior SE Director at Commvault.
By Ram Chakravarti, chief technology officer, BMC Software.
By Darren Watkins, chief revenue officer at VIRTUS Data Centres.
By Steve Young, UK SVP and MD, Dell Technologies.
By Richard Chart, Chief Scientist and Co-Founder, ScienceLogic.