Building One AI Releases Five Times As Much Carbon Dioxide Of A Car In Its Entire Lifetime

Artificial is being heralded as the technology of the future, with the capability to make more efficient every industry from transportation, to manufacturing, to medicine, and more.

The problem is, while we put AI into everything, there's a side effect we don't see.

You see, we might be using more energy to train those AI than we put back into the grid through efficiency.

According to a study led by Emma Strubell at the University of Massachusetts Amherst, training a neural network creates a carbon footprint of 284 tonnes. That's five times as much emissions an average car would put out in its entire lifetime.

Strubell's study analysed four different AIs all focused on text comprehension; Transformer, ELMo, BERT, and GPT-2. That last one is, for example, trained from 8 million web pages to predict the next word in 40 GB of Internet text

The thing is, neural networks during training work by "brute force". Basically, most use a trial and error method to learn their way, and this can take days, or even weeks to do. And when they're tuned for specific tasks, that training time goes up

The only solution, the researchers say, is to fine-tune an AI's learning algorithm before it's trained, to optimise how energy resources are used.

Because the world is on the brink of an irreversible climate change threshold. Planting trees isn't enough anymore, we need to output less carbon too.