Published:
AI poses new environmental sustainability issues
Not only blockchain, but AI too is energy intensive.
When talking about intangible tech, the sustainability aspects are often forgotten. Artificial intelligence is a classic example of this.
Training AI involves deep learning, involving vast amounts of data. So for AI to learn and be able to function as intended takes massive computer power, which means it uses a lot of electric energy.
The neural architecture search (NAS) that involves automating the design of a neural network through trial and error, took more than 270,000 hours, using 3000 times the amount of energy.
New estimates suggest that the carbon footprint of training a single AI is as much as 284 tonnes of carbon dioxide equivalent – five times the lifetime emissions of an average car.New Scientist
This is how researchers measured the energy usage of training AI. Each day, they trained one of four different AI softwares — Transformer, ELMo, BERT, and GPT-2 — and sampled the energy consumption throughout.
From an energy perspective, and from a carbon reduction perspective, we should be thinking about designing the services and making sure the algorithms are efficient as possible.Chris Priest, University of Bristol
As an energy intensive programme, AI needs sustainable energy sources to reduce its environmental impact. Since many of them are being trained by big tech, it would be ideal for them to install solar or wind power plants to combat AI's carbon footprint. This could also be one of the cornerstones of governmental regulations for AI companies.
