There is a tangible sense of excitement in the growing field of artificial intelligence (AI) about the potential of generative models to transform entire sectors, improve productivity, and boost creativity. However, as AI permeates more aspects of our daily lives, we are starting to address the environmental effects of its use. Dr. Ascelin Gordon, Senior Lecturer in Sustainability and Urban Planning at RMIT University, is one voice bringing attention to this problem. His caution to the Australian Senate Select Committee on Artificial Intelligence is a welcome reminder that we need to recognize and deal with AI’s environmental effects, especially those caused by energy use.
The Carbon Cost of AI: Unpacking Dr. Gordon’s Warning
Dr. Gordon’s testimony highlights a pressing but often overlooked issue: the immense energy consumption of AI, particularly generative models like ChatGPT, and the subsequent rise in carbon emissions. He compares the current AI revolution to the industrial and information revolutions that preceded it, noting that while these previous revolutions transformed society, they also left lasting environmental scars. The AI revolution, he argues, is poised to do the same if not managed carefully.
The foundation of his concerns lies in the massive energy demand of AI models. For instance, using ChatGPT to generate text consumes 10 to 90 times more energy than a conventional Google search. Meanwhile, producing an image using generative AI is about 20 times more energy-intensive than a text query. With AI-generated videos beginning to take off, the energy cost is expected to soar even higher. At the heart of this energy consumption are data centers, which currently account for 1-3% of global electricity use. In Australia, these centers use approximately 5% of the country’s electricity supply—a number likely to rise as AI adoption continues.
This trend is alarming not only because of the sheer scale of AI’s energy demands but also due to the speed at which it is growing. The popularity of AI tools like ChatGPT and their integration into everyday software platforms means that AI models are being run millions, if not billions, of times daily. As more industries and individuals rely on these tools, the strain on global energy resources will only intensify.
The Backstory: The Energy Demands of the Information Revolution
To understand the environmental impact of AI, we must first examine the history of information and communication technology (ICT) and its energy consumption. In the 1990s, as the internet became more widely accessible, web and mail servers began to proliferate, requiring increased infrastructure and energy use. Data centers were established to house and manage these servers, leading to a steady growth in global energy demand.
As ICT advanced, cloud computing, cryptocurrency mining, and on-demand streaming services added further strain to energy resources. The rise of platforms like Netflix, YouTube, and Spotify demonstrated how digital services could dramatically increase energy consumption, particularly in the areas of data storage and transmission. With the advent of AI, particularly generative models, the situation has escalated. These models require significant computational power, both to train and to run, which in turn requires more electricity and cooling for the data centers that host them.
In this context, the emergence of generative AI represents not just the latest chapter in the digital revolution but also a significant escalation in the energy demands of the ICT sector. AI models like ChatGPT are particularly resource-intensive because they rely on large-scale machine learning, requiring vast amounts of data to be processed and stored. The environmental cost of this energy consumption has been largely under-acknowledged, overshadowed by the excitement and promise of AI’s capabilities.
The Long-Term Implications: A Looming Environmental Crisis?
If the energy demands of AI continue to rise unchecked, the long-term implications for the environment could be severe. As Dr. Gordon points out, increasing the efficiency of data centers, while important, is unlikely to be a sufficient solution on its own. The sheer scale and speed of AI adoption mean that efficiency gains will be outpaced by rising demand. Moreover, the global nature of AI, with models often run in data centers located offshore, complicates efforts to regulate or mitigate their environmental impact.
One possible consequence of this unchecked growth is a significant increase in carbon emissions. Currently, data centers already contribute to global emissions, and this figure is expected to rise as AI usage increases. Without proactive measures to address this, AI could become a major driver of climate change. In Australia, where AI use is largely offshore, the country may face additional challenges in managing its environmental footprint, as it will have limited control over the energy sources powering the AI models it uses.
Furthermore, as AI becomes more embedded in critical infrastructure and industries, its energy demands could strain electricity grids, particularly in countries with aging or insufficient infrastructure. This could lead to power shortages or increased reliance on non-renewable energy sources, further exacerbating the environmental impact of AI.
Beyond energy consumption, the environmental footprint of AI extends to the physical infrastructure required to support it. Data centers require vast amounts of land, water for cooling, and rare earth metals for the construction of servers and other hardware. The mining and extraction of these materials have their own environmental costs, contributing to deforestation, pollution, and habitat destruction.
The Road Ahead: Policy Solutions and Responsible AI Adoption
Addressing the environmental impact of AI will require a multi-faceted approach, involving policy interventions, technological innovation, and public awareness. Dr. Gordon suggests that one important step is to encourage greater care in the choice of AI models for different applications. Smaller, more specialized models are often vastly more energy-efficient than larger, general-purpose models. By promoting the use of these models where appropriate, we can reduce the energy demands of AI without sacrificing its benefits.
Another potential solution is to invest in renewable energy sources for data centers. Some companies, such as Google and Microsoft, have already committed to powering their data centers with 100% renewable energy. Expanding this commitment across the tech industry could significantly reduce the carbon footprint of AI. However, this will require substantial investment and coordination between governments, businesses, and energy providers.
Public awareness is also crucial. As Dr. Gordon notes, many people are unaware of the environmental impact of the digital services they use every day. Increasing public understanding of this issue could drive demand for more sustainable AI and ICT solutions. Consumers can play a role by choosing companies and services that prioritize sustainability, while businesses can respond by adopting greener practices and technologies.
At the policy level, governments can play a key role by setting standards for energy efficiency in data centers and AI models. They can also provide incentives for companies to invest in renewable energy and sustainable technologies. In Australia, where much of the country’s AI use is offshore, international cooperation may be necessary to ensure that AI adoption does not come at the expense of the environment.
Conclusion: Balancing Innovation and Sustainability
The rise of AI offers tremendous opportunities for innovation, creativity, and efficiency across industries. However, as Dr. Ascelin Gordon’s testimony to the Australian Senate Select Committee highlights, these benefits come with significant environmental costs. If we are to avoid the mistakes of the past, where technological revolutions have often left a legacy of environmental degradation, we must act now to manage the environmental footprint of AI.
By encouraging the use of energy-efficient models, investing in renewable energy for data centers, and raising public awareness, we can ensure that AI’s transformative potential is harnessed in a way that is sustainable for future generations. The challenge is not just to develop better AI but to do so in a way that respects the planet and its resources. In this sense, the AI revolution offers an opportunity not just to transform how we live and work but to rethink our relationship with technology and the environment.