The Green Update is a bi- monthly/monthly column focusing on recent info related to climate change and the environment. Innovations, policy decisions, green models to follow, anything that can shape our future environment can be discussed here!
Artificial intelligence (AI) is not a recent phenomenon. In 1956, the term artificial intelligence was introduced at the Dartmouth Summer Research Project on Artificial Intellignce, to describe computing systems that performed tasks which usually required human intelligence. While data centers — buildings that house servers, data storage drives, and network equipment in temperature- controlled environments — have existed since the 1940s, the rate at which these facilities are being built has dramatically increased in the past few years. With the rise of language-learning models like ChatGPT, Google’s PaLM, or OpenAI, our lives have been transformed. In considering the amount of energy required to not only fuel, but also cool down these data centers, AI’s environmental impact has risen to the forefront of environmental discussion.
Interestingly, AI has proven itself useful in various ways to combat the climate crisis. For example, some have suggested the automatic adjustment of lighting, heating or ventilation in buildings using weather data could contribute to offsetting human energy and water consumption. Energy consumption in homes and buildings is responsible for a third of the US’s greenhouse gas pollution, with experts attributing the use of AI to a 30 per cent reduction in such consumption. Similarly, AI’s huge databases offer new possibilities in research and model-building. These insights can, for example, be applied to streamlining city traffic, as tested in Québec City. AI systems have also been useful in predicting extreme weather and preparing communities to adapt to rapidly changing climate conditions.
However, how helpful has AI really been in solving the climate crisis? Do these benefits outweigh
the pollution it creates? How might we change our approach to strike a healthy balance?
Though there are some positive outcomes regarding the utilization of AI to solve environmental issues, widespread AI usage through generative models, with its heavy environmental burden, brings more harm than good. The onset of accessible AI models has caused an explosion in the number and size of data centers worldwide. This level of infrastructure comes with a large environmental toll: data centers need large amounts of electricity constantly, leading them to mostly use dirty electricity as renewable sources of energy have a fluctuating output. Furthermore, the water withdrawal and consumption in data centers needed to maintain a consistent hardware temperature, which reaches high temperatures because of the number of calculations done in mere milliseconds, is astronomical. By 2027, AI is projected to cause between 4.2 and 6.6 billion cubic meters of water withdrawal worldwide, meaning the amount of water withdrawn would be between four to six times the amount of Denmark’s annual withdrawal. Moreover, the data centers’ water consumption used during the systems’ cooling process, which is not discharged back to the environment but either evaporated or too polluted to be released, is also absurdly high.
AI models’ high demand for energy has become a significant element in global electricity consumption. As seen with data centers, AI consumes a lot of dirty electricity, emitting large amounts of greenhouse gases. Even after training AI models, energy is required each time a model is used. ChatGPT, for example, requires ten times more energy than a Google search. On average, a single medium-length response from ChatGPT’s GPT-3 series consumes 500ml of water – about the volume of one plastic water bottle. OpenAI’s CEO Sam Altman even said that saying “please” and “thank you” to the model added millions in computing costs because of additional energy usage.
Overall, the expansion of AI models and its incorporation in our daily lives makes AI highly unsustainable. AI’s small dent in fighting climate change cannot balance out the environmental impacts of the whole industry, which encompasses over 100 million users for OpenAI alone. Yet the AI industry today doesn’t appear to be moving in a more sustainable direction. Golestan (Sally) Radwan, the Chief Digital Officer of the United Nations Environment Programme says that the few regulations in place in Europe and the US have no real “environmental guardrail”. We need to re-think our use of AI as a tool to fight the climate crisis rather than a technology designed to set us back in terms of pollution and unsustainable consumption