In the just-concluded World Economic Forum’s yearly gabfest of business, political, and other elites in the Alpine snows of Davos, Switzerland, conflict, climate change, and AI got top billing.

In this context, it might be worthwhile to assess the green dilemma of AI as well. At the COP28 summit in Dubai, the world’s first AI minister, Omar Sultan Al Olama, stated that AI is the “only solution” for meeting the 1.5 degree target because it can “crunch incredible amounts of data.”

However, he issued a warning, that AI would increase emissions because it uses a huge amount of power. That’s a classic climate-AI conundrum, and a new doomer narrative about AI is brewing.

Eco quandary

AI seems to play a dual role in the environment and climate. It emerges as a crucial tool in addressing environmental challenges, developing low-emission infrastructure, modelling climate change predictions, and guiding us towards a sustainable future, making it a key player for environmental sustainability. AI has the potential to improve energy efficiency, create smarter energy grids, minimise waste, and foster innovation.

Take a few examples. Google’s AI-driven strategy for data centre cooling has reduced energy use by approximately 40 per cent, which is equivalent to removing 64,000 cars from the road annually. Tesla’s electric cars have AI-driven autonomous driving technologies, which boost fuel economy and cut pollution.

AI is included in GE Renewable Energy’s wind turbines to improve their efficiency. Waste Robotics sorts and separates recyclable materials from waste streams using AI-driven robots, increasing recycling efficiency and decreasing landfill waste.

The Ocean Cleanup attempts to clean up marine environments by tracking and collecting plastic debris in the ocean using AI-powered technologies.

According to a McKinsey estimate, manufacturing with AI improvements might cut greenhouse gas emissions by 10-20 per cent. By analysing soil data, forecasting crop yields, and spotting pest and disease outbreaks, AI can support sustainable agricultural practices as well.

However, AI’s impacts on the environment include possible destruction of ecosystems, electronic waste, and carbon emissions.

In a much-cited 2019 study, researchers at the University of Massachusetts, Amherst, conducted a life cycle assessment for training multiple popular big AI models.

It was seen that the process of training a single machine can release about 626,000 pounds of carbon dioxide, which is roughly five times the emissions of an ordinary American car during its lifetime (including during manufacture).

This is also the equivalent of over 300 round-trip flights between New York and San Francisco. That was the GPT-2 era, then. And, in contrast, we’re now entering the GPT-5 era.

Technology has been developing far too quickly; its carbon footprint also does. The amount of energy required to train and execute improved AI models increases dramatically with the complexity of the datasets and algorithms.

Power needs

How much power would be needed for the development, training, and execution of AI? Just to give you an idea, NVIDIA’s new AI servers would use more energy than countries like Sweden and Argentina by 2027, with an annual consumption of around 85.4 terawatt-hours.

Then, there’s the concern about AI-generated e-waste. Recall the 2001 Steven Spielberg movie, A.I. Artificial Intelligence. In the 22nd century setting of the movie, David, a childlike Mecha humanoid robot, was abandoned in the woods, which were full of scrap metal and obsolete Mecha.

Well, we don’t have to wait until the 22nd century. AI-generated e-waste is already posing a significant environmental threat. Dangerous substances like lead, mercury, and cadmium are found in e-waste that can contaminate soil and water supplies and harm the environment as well as human health.

The UN Global E-waste Monitor report projects that by 2030, e-waste will reach about 75 million tonnes.

The impact of AI on natural ecosystems is also alarming. An excessive amount of pesticides and fertilisers could be used as a result of the growing usage of AI in agriculture, damaging the land and water and destroying biodiversity.

In order to “develop AI that is reliable and safe and that can... supercharge climate action,” UN Secretary-General António Guterres issued a plea. It’s definitely not going to be easy to solve. It’s also obvious that significant funding will be needed to make data and electricity more environmentally friendly.

While we continue to discuss the issue further in COP29, COP30, etc., the horizon of AI will be expended exponentially, for sure. And AI’s carbon footprint would also grow alarmingly. AI’s green dilemma would persist.

The writer is Professor of Statistics, Indian Statistical Institute, Kolkata

comment COMMENT NOW