The internet’s smartest machines are thirsty, electricity-hungry tools that cause massive impacts on the environment, silently depleting the Earth of resources.
While generative Artificial Intelligence (AI) tools feel environmentally immaterial, the framework behind it is not.
Every time a user prompts a generative AI model, an invisible network of machinery kicks into gear.
In the background AI consumes electricity, draws on water for cooling and releases emissions.
Training a generative AI model is not a green process. It requires immense computational power that can run for weeks on thousands of servers.
Training a model like GPT-3 consumes 1,300 MWh of electricity, the equivalent to the annual energy use of over 210 Australian households.
This energy use is largely during the training phases; however, throughout the deployment of models, power continues to be drawn.
Running 20 to 50 prompts through a model like GPT-3 uses around 500 millilitres of water each time, which is a 69 per cent increase from 2019.
The rising demand for generative AI is also affecting global emissions.
While generative AI is under scrutiny regarding its impacts on the environment, the fast fashion sector remains one of the world’s most resource-intensive industries.
To produce one cotton shirt, up to 2700 litres of water can be used. That’s enough water to hydrate one person for over two years.
The global fashion industry also contributes 8 to 10 per cent of annual global greenhouse gas (GhG) emissions.
The mining industry is another sector that is overlooked due to the shift in focus on AI.
Mining is responsible for around four to seven per cent of annual GhG emissions. As well as its requirement for over two million litres of water to produce one tonne of lithium.
While these two sectors continue to significantly contribute to global emissions and resource exhaustion, their environmental impact is often overshadowed.
Generative AI, though proven less destructive right now, has become a focal point in discussions surrounding sustainability.
UOW Ai literacy expert, Dr Juliana Peloche agrees that while generative AI has a sizeable impact on the environment, its scale in comparison to other sectors isn’t always recognised.
“The last time I looked into this, fashion’s emissions were substantially larger than what we currently assume AI causes,” Dr Peloche said.
“However, there’s a crucial issue with the data we’re using for these comparisons. We have to rely on what these industries tell us rather than what they actually show us through transparent reporting.
“Both industries highlight a critical pattern: consumer technologies that appear completely benign can have enormous environmental consequences that remain largely invisible to end users. This invisibility is perhaps the most concerning aspect of both cases.”
Dr Peloche highlights the challenge of making clear comparisons between AI and other sectors due to the lack of transparent data available to the public.
“What’s really happening behind the scenes is that large tech companies are withholding crucial data that we need to properly analyse and measure the environmental impacts of these AI tools, she said.
“Given this lack of transparency, I don’t blame the general public for their limited awareness of the issue.”
Echoing these concerns is University of Wollongong lecturer Wes Wickham, who is currently undertaking PhD research on the intersection of generative AI and its transformative role in creativity and design practice. Mr Wickham has concerns about AI but also views it as a natural progression in technological advancement.
“My own opinion is that it is just one other cog or aspect of our modern society and will become inextricably connected to most other aspects of technology, just as things like ubiquitous internet,” Mr Wickham said.
“I don’t think there’s a good outcome overall unless there is a major paradigm shift that encompasses everything.
“I do think the issue is exaggerated somewhat. Nobody considers what the environmental cost of streaming Netflix or buying a new iPhone with lithium batteries. [But] I do believe there is a need for some deeper research into the environmental footprint of AI.”
As generative AI continues to be embedded in everyday use and has a growing presence in education, many students remain unaware of the broader implications, with universities taking responsibility to increase transparency and digital literacy.
“Making the invisible visible remains one of our most powerful options, alongside bringing awareness and AI literacy to help students understand what’s actually happening behind the scenes,” Dr Peloche said.
“Students need to grasp that AI extends far beyond the simple act of generating a prompt with generative AI tools.
Mr Wickham said he struggles with the ethics surrounding generative AI tools.
“I do have a nagging feeling about the environmental costs,” he said.
“My hope is that there is a solution to it because nobody will stop using it now.”
While far from being green technology, AI’s current digital footprint is lighter in comparison to other industries, stabilising progress and its responsible use seem vital in shaping an ethical and environmentally sustainable future for generative AI.