AI can combat climate change but needs to reduce own carbon footprint

Tilda, a designer, is determined to do her bit to reduce carbon emissions on Planet Earth. She will be presenting a handmade collection of 30 garments created entirely from discarded and repurposed materials in a metaverse store on June 5, which is World Environment Day. So what, you may ask? There are countless designers around the world who are doing so.

Well! Tilda is not a human but an Artificial Intelligence (AI) artist developed by LG AI Research. She unveiled her first such collection of clothing at the New York Fashion Week this February for which she partnered on the 'Flowers on Venus' runway collection with designer YounHee Park, CEO of Greedilous--a fashion brand based in New York. Tilda's job was to create the art and prints that would appear on the clothing. Only 13 images were used in the final collection from hordes of images that had to be created. However, the images that failed to make the cut were dumped as discarded data into a virtual landfill.

For her current 'Digital Upcycling Project', Tilda has designed 13 jackets, 14 trousers, and three hats, using digital waste combined with secondhand denim and fabrics. The new designs were made from the previously discarded images she had drawn for Fashion Week in New York. The existing images were reinterpreted into various combinations (items, colors, patterns, etc.) to act as sources for the upcycled designs. Users can purchase Tilda's collection at dupbytilda.com--Tilda’s 3D metaverse store. All proceeds from the collection will be donated to supporting marginalized artists who support environmental causes, according to a press statement.

Tilda's efforts have been recognised even by the United Nations Environment Programme (UNEP) with her Digital Upcycling Project being an officially registered global event under the institution's Earth Action Numbers program. “I achieved zero-waste by upcycling my own digital waste. The least people can do is help reduce digital waste by clearing out their email inboxes, right?” Tilda exhorted in the release. The press statement pointed out that 4 grams of carbon are emitted for every outgoing email, which implies that if 2.3 billion internet users each deleted just 10 emails, this would account for 1.7 million gigabytes of energy saved on data archiving.

Tilda, according to LG AI Research, specializes in illustration, pattern design, and can create new images using its Exaone AI model that has the ability to link languages and images, having trained on "600 billion" pieces of text collections and "250 million high-resolution images integrated with text". It is also capable of tuning 300 billion different parameters (variables), the largest in South Korea. We may liken parameters to synapses in the human brain that deliver information between neurons.

Tilda is a stellar example of AI's potential in combating climate change. To be sure, as part of its 'AI for the Planet' initiative to drive AI solutions for climate change at scale, UNESCO has partnered with the United Nations Development Programme (UNDP), the UN Office of Information and Communications Technology (OICT), and the AI for Good Foundation along with Startup Inside, Boston Consulting Group (BCG) and BCG GAMMA.

In May 2019, Carbon Tracker Initiative Ltd., launched a new project that was funded by a $1.7 million grant from Google.org and uses satellite imagery to quantify carbon emissions from all large power plants worldwide and make this information available to the public. Carbon Tracker partnered with a San Francisco-based non-profit organisation WattTime and the World Resource Institute (WRI) for the same. The project works by monitoring power plants with the help of space satellites and using image processing algorithms to detect signs of power plant emissions. The project combines data from sensors operating at different wavelengths. AI algorithms will cross-validate multiple indicators of power plant emissions, from thermal infrared, indicating heat near smoke stacks and cooling water intake, to visual spectrum recognition that a power plant is emitting smoke.

However, using AI to combat climate change may seem paradoxical in the light of it being an energy guzzler with a very wide carbon footprint. According to a May 16 2018 note by OpenAI, a non-profit organization co-founded by Elon Musk, "since 2012, the amount of compute used in the largest AI training runs has been increasing exponentially with a 3.4-month doubling time (by comparison, Moore’s Law had a two-year doubling period)". This metric, according to the note, has grown by more than 300,000x (a 2-year doubling period would yield only a 7x increase).

According to a research paper published on online repository arvix by a team at UMass Amherst, training a single large NLP model is likely to devour as much energy as a car over its entire lifetime, which includes the energy needed to build it. And according to a 12 September 2018 article in Nature (How to stop data centres from gobbling up the world’s electricity (nature.com), "data centres contribute around 0.3% to overall carbon emissions".

To be sure, in many cases, AI models no longer need humungous amounts of training data as before to be efficient, which also implies a reduction in their need for computing power. But newer companies adopting AI simultaneously will continue to increase the demand for computing power. Hence, companies will continue to step up efforts to cool their data centres, recycle waste heat even as AI algorithms themselves will have to be made more energy-efficient as they tackle social issues like climate change.