The Dark Side of Technology and Its Environmental Impact
Across many industries from healthcare to finance, AI is often celebrated as a transformative force although the discussions of its impact on the environment remains limited. While debates focus on AI's effects on employment and ethics, the substantial carbon footprint of AI systems deserves greater attention.
?
The researchers at the University of Massachusetts Amherst conducted a study in 2019 which highlighted increasing consumption of energy of the AI models. A single large model like BERT can release around 626,000 pounds of CO2 for training it. This is said to be almost equal to the emissions of five average cars through their whole lifetime.?
?
With 175 billion parameters, the training of OpenAI's GPT-3 consumed around 1,287 MWh and produced 552 metric tons of CO2 emissions, or about 60 typical American households' worth of emissions annually. The increase in the AI models also increases the need for substantial computing. This in turn leads to the rise of consuming energy in data centres which are powered by fossil fuels.?
?
Data centres, which are crucial for AI and digital services, have a huge environmental impact, notably in terms of water usage. On an average they are said to use between 1.1 and 1.2 million litres of water per day to cool the servers. With the rising demand for cloud storage and AI the burden on water resources also increases and becomes problematic for countries where water shortage already exists.?
?
An estimation by the International Energy Agency (2023), the energy and water consumed in the training of a single AI model is almost equal to a hundred homes in a day.?
?
This exacerbates water scarcity in underdeveloped countries, who are already facing challenges. The World Resources Institute (2021) warns that the growth of data centres may exacerbate water shortages in these areas. The water scarcity worsens due to this in developing countries.?
?
Tech companies like Microsoft are exploring sustainable energy solutions. A notable example is the restart of the company’s 20 year deal called ‘Three Mile Island’ nuclear plant expected to provide over 835 MW of carbon-free electricity by 2028.
?
Although there are several measures taken to limit AI’s environmental impact mainly focusing in terms of water and energy use. Strategies like quantization, model pruning and knowledge distillations implemented to reduce computational complexity in training and inference. These strategies can also be used to lower energy requirements while retaining performance.?
?
To summarise, the environmental implications cannot be ignored though AI offers significant benefits. Addressing these challenges is crucial for a sustainable future.
?
?
?
?
#AI #EnvironmentalImpact #Energy #CO2Emissions #Technology #Microsoft #Internet #Future #Sustainability?