I’ve just been reading a very interesting report about the CogX 2019 event written by Charles Radclyffe. He’s a Digital Ethicist and naturally in this capacity, a conference about AI and Emerging Technologies would interest him.
His witty review highlighted an aspect of Artificial Intellignce (AI) that I hadn’t completely considered before, and I am referring to “AI for Reducing Energy Emissions.” It seems Radclyffe was so bemused by the idea that he had to go to the workshop of that title, which is how I cam to learn about the topic and share his thoughts with you.
Radclyffe starts by saying that the whole idea of AI being able to clean up energy emissions is ironic. It is after all, he said, “essentially a bunch of fancy math that is able to do certain things at human levels or above if it’s fed enough data.”
But to get really complex AI you need “fancier math and more data”, and as the writer says, that means using more electricity to fuel the computational resources that run the AI.
As he says, that’s why bitcoin miners “build their data-centres under hydro-electric powerplants. If the power is cheaper, the math is too.”
Yes, AI can do some amazing things, but as Radclyffe writes, “The dirty secret is how much electricity this consumes to train, run, and manage the model behind this.”
To give you an idea of just how ‘dirty’ AI is, here’s Radclyffe’s illustration: “While a human is content with a bowl of cereal and a latte in the morning — an AI will be busy working through the equivalent of a few thousand lbs of CO2 in order to achieve the same task.”
He isn’t making this up. Researchers at Amherst have compared the lifecycle of an AI model against the carbon footprint of an average car in the US. The result showed, “The AI would emit nearly 5 times the carbon dioxide of the car (including the manufacture of the car itself).”
Climate change is one of the biggest challenges we are facing right now, so should we be pursuing the development of AI, or concede that it just isn’t good for the environment? Or at least we need to be more transparent about AI’s carbon footprint.
Which brings us back to CogX, where a number of panellists were outspoken about the need to face the challenges of getting multi-national support for initiatives, such as new methods of energy conservation, or initiatives like the International Centre for AI, Energy & Climate.
CogX has signed up to the UN’s Sustainable Goals, but as Rasclyffe says, it celebrates AI as the solution to climate change, yet for the foreseeable future it will be a significant contributor to climate change. It’s a puzzle that needs to be solved.