While the metaverse continues to generate headlines and business plans around the world, academics and industry analysts say more should be done to consider the real energy and power requirements demanded by the endeavour.
And while the world waits for companies to catch-up and take concerted action at the global scale, individual consumers should consider changing their relationship with technology in order to help build a more sustainable virtual reality for the masses.
Neural networks are going to be required at enormous scale in order to deliver on the promises being made by metaverse developers. These networks have enjoyed a period of significant advances, with progress in hardware and methodologies leading to a new generation that has been trained using huge datasets.
These networks have seen significant advances in terms of accuracy, but improvements like this depend on the availability of huge computational resources that require enormous energy consumption, explain Emma Strubell, Ananya Ganesh and Andrew McCallum, of the College of Information and Computer Sciences, University of Massachusetts Amherst, in their paper Energy and Policy Considerations for Deep Learning in NLP.
As a result, these models are costly to train and develop, both financially, due to the cost of hardware and electricity or cloud compute time, and environmentally, due to the carbon footprint required to fuel modern tensor processing hardware.
One AI model could generate five cars worth of greenhouse gases
“Training just one AI model could generate 626,000 pounds of carbon dioxide, which is more than five times the amount of greenhouse gases emitted by a car in its lifetime,” says Lizzy Rosenberg, Associate Editor of Distractify in her article published in association with the World Economic Forum. “Cloud gaming, which is necessary for VR, could also raise carbon emissions by 2030. And, it will increase the necessity for high-res images, which only increases the need for more energy.”
As the metaverse will encourage users to buy new VR technology and other hardware, this could also lead to an increase in “e-waste”, which – if not recycled properly – can pollute soil, groundwater and landfills.
In their paper, Strubell, Ganesh and McCallum recommended a concerted effort by industry and academia to promote research into more efficient algorithms and hardware. There is already a precedent for NLP software packages prioritising efficient models, they say, and software developers could provide easy-to-use APIs to reduce computational requirements.
Writing for the World Economic Forum, Rosenberg says the onus is currently on major corporations to find eco-friendly means of building virtual realities, but consumers can hold themselves accountable my making a commitment to recycle e-waste and even shop for second-hand electronics.
“Also try to stream in SD [standard definition]— not HD [high definition] — when using your phone to interact with the metaverse, as HD has a higher environmental impact and releases more carbon emissions,” says Rosenberg. “Large corporations should be held accountable for this type of impact, but playing your part is important, too.”