The environmental impact of AI: challenges and solutions for a sustainable future
Similar to many digital agencies, Liquid Light has been exploring the capabilities of AI for quite some time now, integrating it into our own internal processes, and working on several AI based projects for our clients.
In a previous post I described our approach to sustainable development and how we aim to create websites and online products that keep their environmental impact to a minimum. As an agency who pride ourselves on the sustainability of our products, it's crucial for us to consider the environmental impact that AI may have as we continue to integrate it.
The hidden carbon footprint of AI
What lies behind AI’s outputs is an energy-intensive process with a staggering (and scary) carbon footprint. This issue has been in the news again this October with Google signing the world’s first ‘corporate agreement’ to buy nuclear energy produced from small modular reactors (SMRs) to power their AI technologies (Google frame this as "accelerating clean energy solutions” but whether an energy source that produces large volumes of radioactive waste can be considered clean is debatable).1
As datasets and models become more complex, the energy required to train and run AI models is also growing exponentially:
- Since 2012, the computing power needed for AI models has doubled every 3.4 months 2
- By 2040, the Information and Communications Technology (ICT) industry is projected to account for 14% of global emissions, with AI playing a significant role 3
- A University of Massachusetts study found that training a large AI model can produce approximately 626,000 pounds of carbon dioxide - equivalent to 300 round-trip flights between New York and San Francisco 4
These quite alarming statistics would deter any environmentally conscious business from using AI. Unfortunately there is more bad news:
Electronic waste: The aftermath of AI hardware
Moving on from the ridiculous amounts of energy needed to power AI, there is also the issue of electronic waste (e-waste) and its recyclability (or lack of) and disposal. The e-waste resulting from AI extends far beyond the realm of outdated consumer electronics (a problem in itself). As AI continues its rapid advancement, it's generating a new category of technological detritus.
Each iteration of more powerful AI systems potentially renders previous hardware obsolete, creating a constant stream of discarded specialised processors, memory units, and cooling systems (by 2050, the World Economic Forum projects that e-waste will exceed 120 million metric tonnes annually).5
This AI-specific e-waste presents unique challenges: it often contains rare earth elements and potentially hazardous materials that pose significant environmental risks if not properly managed.
Is there an environmentally friendly future for AI?
A common response I see from AI evangelists when confronted with AI’s carbon footprint is that AI will solve the problem itself, that the exponential rate of progress around renewables and energy efficiency will make this problem simply go away. This feels a little utopian and action concerning climate change has to happen now, not be wished for in the future.
But there are some potential positive uses of AI that could actually benefit the planet. Whether these positives are actualised and tip the balance and start to compensate for AI’s huge energy usage remains to be seen.
- Climate Modelling: AI could enhance the ability to predict and understand climate patterns, enabling scientists to forecast extreme weather events and long-term climate trends with improved accuracy.
- Smart Grids: Although not necessarily wholly driven by AI, AI could complement and improve intelligent energy distribution systems that balance energy supply and demand, with the aim of reducing waste.
- Precision Agriculture: Precision agriculture is a farming management system based on “observing, measuring and responding to temporal and spatial variability to improve agricultural production sustainability”. Similar to Smart Grids, this is not an exclusively AI solution, but one where the use of AI could complement the use of sensors, drones, and satellites to optimise water usage, reduce pesticide application, and improve crop yields, contributing to more sustainable agricultural practices.
- Waste Sorting: AI-enabled recycling systems could sort waste with high speed and accuracy, significantly improving recycling rates and reducing landfill use.
- Ocean Clean-up: AI algorithms could guide autonomous vessels in efficient collection of ocean plastic, enhancing efforts to address marine pollution.
- Building Energy Efficiency: AI-powered building management systems could optimise energy consumption in urban environments by efficiently managing heating, cooling, and lighting.
- Deforestation Monitoring: The combination of satellite imagery and AI can enable real-time detection of illegal logging activities, facilitating prompt intervention to protect endangered forests.
- Traffic Optimisation: AI can assist in reducing urban congestion and emissions by optimising traffic flow, contributing to more efficient urban mobility.
Some of the examples above are in use or being trialled and some are still at a conceptual stage, requiring significant financial investment, as well as cultural and societal changes, to come about. So although it is positive to imagine these new realities, some do seem a little far off.
The path to sustainable AI
So how can a small business like ourselves mitigate the environmental impact of AI?
- Green hosting solutions. Partner with cloud providers committed to renewable energy and check the green credentials of any AI tool provider you use
- Efficient data practices. Streamline and sanitise the datasets you work with. We work with all kinds of datasets, and putting effort into cleaning up data before any serious work starts has many positives for both ourselves and our clients. From an AI perspective, quality over quantity not only improves AI performance but also reduces energy consumption.
- Leverage transfer learning. Don't reinvent the wheel for every project. Utilise pre-trained models where possible, reducing the energy-intensive training process.
- Monitor and measure impact. If possible Implement tools to track the energy consumption of your AI projects. Many agencies like ourselves already track their own energy consumption (for our efforts in becoming a B-Corp). Extend this to your use of AI.
- Sustainable hardware choices. If and when investing in AI hardware, prioritise longevity and upgradability. It might cost more upfront (and require a bit of guesswork), but it'll pay dividends in sustainability and reduced e-waste.
- Educate your team and your clients. Advocate for sustainability practices and processes within your business. Train your team on sustainable AI practices and educate or influence your clients on the long-term benefits of eco-friendly tech solutions.
In summary
As a tech business, it's our responsibility to ensure that technological progress doesn't come at the expense of our planet. As we continue to look to integrate AI into our processes, we pledge to maintain transparency about its environmental impact while actively seeking ways to minimise our carbon footprint. We believe responsible AI adoption and sustainability can coexist – it's not about choosing one over the other, but finding innovative ways to advance both causes.
If you have an AI project in mind, or just want to talk over the possibilites AI could have for your orginsation, get in touch.
References
1. https://blog.google/outreach-initiatives/sustainability/google-kairos-power-nuclear-energy-agreement/
2. https://www.gstatic.com/gumdrop/sustainability/google-2024-environmental-report.pdf
3. https://www.sciencedirect.com/science/article/abs/pii/S095965261733233X
4. https://arxiv.org/abs/1906.02243
5. https://www3.weforum.org/docs/WEF_A_New_Circular_Vision_for_Electronics.pdf
-
Owen Priestley
Strategy Director
