If you took a peek inside the typical household, it wouldn’t be unusual to find a kid playing e-sports while their sister uploads a dance choreography video to TikTok in the next room, and Alexa guides one parent through a recipe in the kitchen as another makes a video call in a home office. Of course, much of western society doesn’t resemble this middle class, advertising favourite of the nuclear family structure. People are just as likely to be living solo, or in shared housing. However, regardless of their living situation, one thing is constant: individuals are increasingly connected to one or more personal electronic devices, whether alone in different rooms or while sitting side by side. Cultural shifts like this contribute to unprecedented spikes in data generation and consumption.
The internet is often associated with leisure time, but if we dig a little deeper, it is invisibly entangled with everyday life on a personal, public and industrial level. Electronic data capture, transfer and storage are woven into systems used by hospitals, transport, manufacturing and even energy and water supply. The Internet of Things and machine learning have the potential to make massive efficiencies to improve public and private life. Unfortunately, the continuation of trends like these, whicb involve constant connectivity and energy-intensive information and communications technology (ICT) systems, have a hefty environmental cost. Huge data centres filled with servers housing millions of gigabytes of information need to be industrially cooled, which uses vast amounts of energy. Devices themselves are powered by electricity. Networks use the most energy of all.
Internet connectivity has become like oxygen. We unconsciously bumble about our lives, constantly generating and consuming data, until we enter a 4G mobile blackspot or suffer a power outage or low battery. Unless you’re high or have insomnia, it is rare for people to question how it works. Power cuts and subsequent internet loss during disasters like the California wildfires and Hurricane Maria in Puerto Rico highlight both the dependency of the internet on electricity and our dependence on the internet. Worldwide, ICT is currently estimated to account for over 2 per cent of total energy consumption.
However, sustainable ICT specialist Andres Andrae has predicted growth of up to 21 per cent by 2030 if trends like AI and machine learning, the Internet of Things and virtual and augmented reality continue as projected. According to Andrae’s best-case scenario, ICT will take up 8 per cent of total electricity demand. Unsurprisingly, our endless appetite for digital sensing, scrolling and sharing have environmental consequences.
That said, it isn’t all bad news. At the moment, efficiency drives the likes of Facebook, Alphabet and Apple to create minimal frills, hyper-scale data centres, which has slowed the rising carbon cost of ICT, despite the growing demand for data. Yet, unfortunately, carbon isn’t the only environmental concern: a study from Imperial College London estimated that US data centres used 100 billion litres of water in 2014 to cool their servers.
Although it isn’t possible to verify predictions such as Andrae’s, the potential efficiencies promised by innovations in AI, quantum computing or other futuristic technologies aren’t necessarily enough. In 2016, Google’s Deep Mind was put to work on further improving energy efficiency in their “already highly optimised data centres.” The unspoken problem here is that efficiencies can’t continue indefinitely, but demand can continue to rise steeply.
Besides, machine learning and AI have their own sustainability issues. Recent research published by Cornell University highlighted the alarming carbon footprint of deep learning networks trained on an abundance of data. At an astronomical 626,000 pounds of CO2, the MIT Technology Review equated a large network using over five times the lifetime emissions of the average American car. Fortunately, in November, Cornell University launched a carbon calculator to help AI and machine learning researchers quantify the carbon emissions related to training a neural network. They highlighted that the carbon cost of training a network depends on the energy grid the server used for training, the energy grid that powers it, the length of the training procedure and the model of hardware.
A straightforward solution to prevent the negative impact of ICT on the environment would be to use only renewable energy to power it. Earlier this month, Yale reported that the global financial cost of switching to 100 per cent renewable energy worldwide was $73 trillion. Another option is to be more mindful about the ways we collect, transfer and store data.
In her last talk as CEO of the responsible technology think tank Dot Everyone, Rachel Coldicutt remarked, “The amount of data collected is going up and up and up, but very little of it is understood. Just as we’re burning the planet, we’re at risk of drowning ourselves in data—making new problems faster than we have time to solve them.”
Access to ‘unlimited’ data doesn’t mean we have to gorge on an endless wireless buffet. Consciously making decisions about why, when and how we connect to the internet can bring more balance to our lives and take some of the strain off the Earth’s natural resources, too. Until there is 100 per cent worldwide renewable-energy-powered internet, maybe it is worth pausing to think before binge watching Netflix for 8 hours straight or training a neural network to identify a celeb’s plastic surgery.