This week wintry weather battered the southern US and parts of Europe. Researchers suggest this could be a counterintuitive effect of the climate crisis where the warming of the Arctic causes the jet stream to shift. “The energy escaping from the jet stream bangs into the polar vortex so it starts to wobble and move all over the place,” said Judah Cohen, the director of seasonal forecasting at Atmospheric and Environmental Research, to the Guardian. “Where the polar vortex goes, so goes the cold air.”
While many of us do our part to avoid common climate blunders –– avoiding plastic bags, opting for walking instead of driving, using recycling bins, etc –– we don’t always see the impact of our technology use. In the field of AI, computations required for deep learning research have been doubling every few months, resulting in an estimated 300,000x increase from 2012 to 2018. Several factors impact the carbon emitted by neural network computations including the location of the server used for training, the energy grid that it plugs into, the size of the dataset, and the hardware where the training takes place. Tools are available to help individuals calculate their machine learning carbon impact.
While 1 percent of worldwide electricity consumption can be attributed to data centres, usage is forecasted to increase to 15 percent of global electricity consumption by 2025 due to the growth of inference deployments. The use of increasingly energy-efficient processing units, as well as expanding efficiencies in servers, storage, devices and hyperscale data centres, offer optimism for the future. New inference chips are coming to market that offer significant leaps in efficiency. For example, Untether AI, a Radical Ventures portfolio company, launched a new inference chip that is 8 times more energy efficient than the current market leader.
While machine learning carries an environmental footprint, it also plays a critical role in bringing efficiencies to Green House Gas-emitting industries. Some of the highest potential areas identified by researchers include forecasting supply and demand for electricity scheduling, accelerating materials science for better energy storage, preventing methane leakage from natural gas pipelines, reducing transport activity, improving urban planning, optimizing supply chains, precision agriculture, carbon dioxide removal and more accurate climate prediction.
At Radical, we will continue to look for investments that are great AI businesses, and also good for the health of our global climate.
AI News This Week
-
NASA’s Perseverance Rover Lands on Mars: Here’s How It is Using AI (Enterprise AI)
After its ambitious landing on Mars last week, NASA’s Perseverance Mars rover is already getting critical help from the multiple onboard AI systems that are designed to guide its two-year exploratory mission on the red planet. Perseverance carries more AI capabilities and technologies than any Mars rover before, including AI-enabled terrain navigation and autonomous exploration for gathering rocks.
-
A New Artificial Intelligence Makes Mistakes – on Purpose (Wired)
An AI chess program, Maia, is focusing on predicting human moves, including the mistakes we make. This is a first step toward developing AI that better understands human fallibility. The aim is to create AI systems that are better equipped to interact, assist and negotiate with humans. For example, in health care, a system that anticipates errors might be used to train doctors to read medical images or assist them in catching errors.
-
With AI, No Experience Needed for Echo? – Software proves its mettle for novices (MedPageToday)
Researchers at Northwestern Medicine in Chicago have found that individuals with no prior ultrasonographic experience were able to acquire diagnostic quality transthoracic echocardiography (TTE) images with help from artificial intelligence software. Study authors noted that the goal is not to replace sonographers who provide expert imaging. Instead, this AI technology may allow for diagnostic-quality ultrasonographic studies in settings with limited trained personnel.
-
Study shows that federated learning can lead to reduced carbon emissions (VentureBeat)
Researchers are investigating more energy-efficient approaches to training AI models. Their findings suggest that federated learning has a quantitatively greener impact despite being slower in some cases. To measure the carbon footprint of a federated learning setup, the coauthors of the new paper trained two models, recording the power consumption of the server and chipsets during training, taking into account how energy usage might vary depending on the country where the chipsets and server are located.
-
Mysteries of massive holes forming in Siberian permafrost unlocked by scientists (CNN)
Seventeen holes have appeared in the Russian Arctic since 2013 from powerful blowouts of methane gas thought to be linked to climate change. Such events pose risks to Indigenous people and to oil and gas infrastructure. Drone photography, 3D modeling and artificial intelligence are helping to predict the location of future craters. Researchers devised an algorithm to quantify changes to the height of mounds and the expansion or shrinking of lakes to predict where the next blowout crater might occur. The model correctly predicted all seven craters that had been reported by scientists since 2017 and revealed the formation of three new ones.
Radical Reads is edited by Leah Morris (Senior Director, Velocity Program, Radical Ventures).