AI’s Environmental Harm: How to Reduce AI’s Energy Consumption

Canada’s Artificial Intelligence (AI) industry has witnessed remarkable growth in recent years, becoming a global leader in research and development. In fact, Stanford University’s Global AI Vibracy Tool ranked Canada fifth of 29 countries, and third among G7 nations, in AI research, development, and economy in 2021. The application potential for AI is clearly wide-reaching and growing by the day, with many examples already permeating our lives. From creating content with ChatGPT and servicing customers with chatbots to the rise in autonomous driving, AI is changing the ways in which we live, work and play.

The only limit to AI’s usage is our imagination. But is there a cost to this seemingly omnipresent technology? 

What is the Scale of AI’s Environmental Harm?

Unfortunately, there is. Many people don’t realize that every time you use an AI tool, you are contributing to the climate crisis. Traditional AI methods are an increasingly large drain on the world’s energy resources, due to the power required to train and deploy AI models. A study from the University of Massachusetts Amherst stated that training a single AI model is equal to approximately 300,000 kg of carbon dioxide emissions. That’s as much carbon as what gets created by five cars in their lifetimes or 125 round-trip flights between New York and Beijing. In other words, AI, as it currently exists, is a significant strain on the planet, and as our dependency deepens, so too will the amount of power required. 

To quantify it, data centres currently require greater than 1% of the world’s electricity — a percentage that is expected to increase as AI usage increases. This is because whenever you move data, you burn energy, and the further you move that data, the more power is required. Data centre energy consumption driven by AI is forecasted to double by 2030 – with one prediction stating that by 2025, 10% of the world’s energy consumption will be used by AI applications. 

Considering Canada is already facing an energy crisis — stemming from a growing demand for electricity, limited renewable energy sources, and aging infrastructure — this picture should come as a shock, or at the very least, a serious wake-up call. 

AI is Built on Decades-Old Technology. That Needs to Change.

Today’s AI computers are built using decades-old technology called von Neumann architecture. Developed in the 1940s, von Neumann architecture was a very powerful model that worked well for the semiconductor world up until now. It’s based on memory being on the outside and processing on the inside, facilitating a connection between the two through a long and narrow path. AI technology requires a significant amount of data that moves across great lengths — and that burns a great deal of energy. While the von Neumann model met the technology needs of the past, it falls disastrously short in the face of AI’s enormous energy consumption requirements. 90% of energy used in the von Neumann approach is expended on moving data. 

For example, the energy consumed to run ChatGPT for the month of January 2023 could power a town of 175,000 people for a month. Likewise, while autonomous vehicles are being touted as the way of the future, if the world moved to 100% autonomous vehicles tomorrow, the greenhouse impact would be larger than the current impact of all of the computers in all of the data centres worldwide, due to the highly inefficient way of AI processing.

We need to embrace a technology that’s superior to the approach that semiconductor companies have used for the past 60 years. AI is continuously evolving, and we need to evolve along with it — especially when it comes to energy usage. 

As countries worldwide work to achieve an ambitious target of net zero carbon emissions by 2050, AI is posing a significant threat to achieving the goal. However, the growth and advancement of AI should not come at the expense of the environment. Addressing these challenges is not only morally imperative, it is a strategic necessity for the future economy. 

How to Mitigate AI’s Environmental Harm

While the energy crisis presents considerable hurdles, Canada’s AI industry has the potential to play a pivotal role in addressing a myriad of challenges. Canadian universities like the University of Toronto and the University of Waterloo graduate the best of the best AI engineers and it is the technology leaders of today and tomorrow that bear a responsibility to lead the charge in minimizing the environmental harm of AI. 

However, the responsibility cannot fall on technology leaders alone. There must be collaboration among governments, researchers, industry leaders, AI experts, and environmental organizations to set regulations, standards, and best practices for AI development. Collectively, these stakeholders can make a significant difference by fostering a culture of innovation, collaboration, and sustainable practices.

Here are some recommendations for fostering a sustainable AI future:

• Deploy With Sustainability in Mind: Before deploying a product to market, it is incumbent upon companies to have a clear view of the environmental impact. Instead of simply going live,  organizations must take the time to develop a plan to ensure it is sustainable. As consumers become more aware of the environmental impact of AI, they will come to expect (and indeed, demand) this level of attention to detail. 

• Invest in Energy-Efficient AI Chip Technology: As an industry, we need to prioritize the development of AI technology that is more energy-efficient without compromising performance. At Untether AI, we are tackling the unsustainable power consumption issue head-on. Our company has created first-of-its-kind non-von Neumann technology which dramatically reduces AI’s energy use. To achieve this, we placed the processing and the memory next to each other, thereby solving the data movement bottleneck that costs energy and performance in traditional AI chips. With data movement now contained in a very short distance, the energy required to move data is reduced by up to 90%. By prioritizing performance and energy efficiency, we are addressing a critical issue facing technology companies. By working closely with leaders in the space, our energy-efficient AI technology is being deployed in industries focused on industrial automation, autonomous vehicles, smart cities, and smart retail. 

• Build Energy-Efficient Data Centres: In their current form, data centres are a tremendous drain on the environment. Each server in a data centre has a limit to how much energy it can consume. With the high power of von Neumann-based GPUs, these data centres are filled with servers that are not fully populated. Therefore, we need to build energy-efficient data centres and that can happen with the non von Neumann silicon and subsequently reduce the carbon footprint of AI technologies.

• Invest in Sustainable E-Waste Management: What happens when AI hardware, such as servers, chips, and GPUs, becomes outdated? Just as we have a system for recycling consumer materials, we need a plan for the responsible disposal and recycling of e-waste. Whether it is by partnering with third-party recycling companies or discussing large-scale recycling programs through cities and municipalities, AI companies must ensure that their e-waste is handled in a sustainable manner. Developing AI hardware with longer lifespans can also help reduce the amount of e-waste generated. 

Canada’s AI industry holds immense potential to address the ongoing energy crisis and pave the way for a sustainable future. By prioritizing energy efficiency, investing in hardware innovation, promoting responsible practices, and collaborating with an array of stakeholders, we can harness the transformative potential of AI while minimizing its environmental impact.