Artificial Intelligence and the Environment
It is incredibly important to discuss the ethics of artificial intelligence and how the growth of this technology will impact the global community. There has been much discussed about how artificial intelligence has been developing too quickly. Regulators aren’t able to properly create safeguards to protect users from a myriad of risks, and if the capabilities of artificial intelligence advance too quickly, there could be serious repercussions. There may be bias in algorithms, security concerns that can be exploited by malicious hackers, unintended consequences such as ultra-believable deepfakes, and many other negative externalities. One area I don’t think is discussed enough though, is the environmental impact of advancing artificial intelligence.\
To illustrate, as we’ve learned in class, it takes massive amounts of power to train AI models. The energy requirements and carbon emissions from training AI models contribute to greenhouse gas emissions and climate change. According to this article, GPT-4 was trained through approximately 25,000 A100 GPUs from Nvidia for almost 100 days. The amount of energy required to power that many GPUs for that length of time is anywhere from 52 million to 62 million KWh. This could equate to 12 to 15 thousand metric tons of carbon dioxide emissions if GPT-4 was trained in California. It’s important to note that this is just one of many AI models being trained in the United States. Evidently, there is a serious environmental impact from the training of AI models. As AI models become more and more complex, the energy requirement to train the models will become even more intensive.
In order to sustain AI’s longevity, it’s important to understand the environmental impacts. We’ve witnessed how environmental impact from energy consumption can hinder other developing technologies. For example, the mining of Bitcoin and other cryptocurrencies requires computers to solve increasingly complex functions, which require increasing amounts of energy. In fact, a study in the Scientific Reports concluded that from 2016 to 2021, each $1.00 (USD) worth of mined bitcoin caused $0.35 in climate damage. While there are many reasons for the decline of cryptocurrencies that I won’t discuss, there are similarities in evolution of cryptocurrency mining and artificial intelligence in that the more advances are made, the more energy is required to enable those advances. Additionally, there are rare earth metals (such as niobium and tantalum), which are highly sought after for AI hardware. Like any other mining operation, there are environmental issues such as carbon emissions, habitat destruction, and other pollution issues (e.g., water contamination). Taking into account the energy and hardware requirements for artificial intelligence, it is paramount that leaders in the AI industry create efficient algorithms and models that optimize both effectiveness and energy efficiency.
While the overall sentiment of this article seems negative on the environmental impacts from artificial intelligence, there are many ways in which artificial intelligence can benefit the environment. First, let’s focus on how artificial intelligence can help the environmental impact from natural events. In partnership with the Pentagon, Carnegie Mellon University’s Software Engineering Institute created xView2, which is a machine-learning algorithm that utilizes satellite imagery to identify and determine the severity of building and infrastructure damage from impending natural disasters. xView 2 can help government officials better prepare for natural disasters and reduce the potential environmental damages. Secondly, let’s focus on how artificial intelligence can help the environmental impact from human dependent events. According to the United Nations, “machine learning can optimize supply chains to reduce waste, monitor resource consumption and promote sustainable manufacturing processes”. One industry where this can take place is fast fashion, which has been growing at a rapid rate throughout the world. Fast fashion companies can use AI models to better predict consumer demand which can help eliminate waste from overproduction. They can also use AI to create efficient supply chains that reduce fuel emissions.
Evidently, there are many environmental considerations from the development and deployment of artificial intelligence. Training and operating AI models requires significant energy consumption, which can have negative impacts on the environment. Additionally, the procurement of materials for AI hardware such as rare earth metals for semiconductors has environmental impacts as well. On the other side, the accurate prediction ability of AI models can help government officials prepare for natural disasters, thus limiting their environmental impact. AI can also help corporations optimize their supply chains to reduce fuel consumption. As the conversations on how to regulate artificial intelligence continue, I think there should be more of an overt focus on the environmental impacts from the technology.
