Algorithms for the greener good
AI is already a huge force in business – but all those hardware advances, big data, and powerful algorithms are pointless if we don’t have clean air to breathe and clean water to drink.
Fortunately Silicon Valley has noticed, and tech giants such as Microsoft, Google and Intel are investing in projects that use deep learning, neural networks, and synthetic data to create AI-based solutions to fight climate change and preserve our planet. Here’s how AI is solving some of the world’s biggest environmental challenges.
By 2050 the demand for food is expected to outpace production by more than 70 percent.
AI can help farmers reduce their use of chemicals, minimize environmental damage, and increase food production while using less land and resources. For more efficient produce picking, Root.ai uses computer vision to “see” individual fruits and determine whether or not they’re ripe, then plucks the fruit with gentle grippers.
Instead of blanketing crops with harmful pesticides, Blue River Technology uses computer vision to scan plants for disease and only spray when necessary. Livestock farmers in the Netherlands are using Connecterra’s Intelligent Dairy Farmer’s Assistant; it collects data from sensors located on a collar worn by cattle that monitors behaviors related to their health, fertility, feeding and more.
The World Health Organization says air pollution causes 7 million deaths a year and more than 90 percent of children breathe toxic air every day. In London, the Alan Turing Institute is planting more than 1200 remote sensors across the city that monitor air quality. That data is meshed with weather, traffic flows, construction activity and known street “canyons” (streets surrounded by high buildings where toxic air can’t dissipate) to forecast the city’s most polluted areas 48 hours in advance.
In other parts of the world, IBM’s Green Horizons Initiative is working with cities with particularly poor air quality, such as Beijing and Johannesburg; its AI monitors the movement of pollutants and the chemical reactions between different pollutants, predicting highly polluted areas up to 10 days in advance.
Someday, AI will be the brains behind a proposed national smart grid that will take the input of millions of sensors and decide where to allocate energy resources in real time. Our current large regional grids would be replaced by microgrids that can fine-tune energy allocations for each community.
In the private sector, Google’s DeepMind (of Go-playing fame) is using neural networks trained on weather forecasts and historical turbine data to predict how much power wind turbines will be able to deliver a day in advance. Overseas, Spain’s Siemens Ganesa has developed drones that use AI for wind turbine safety inspections; the drones snap around 400 images of a turbine’s 22-story blades in 20 minutes and use image recognition to scan for needed repairs.
When severe weather and natural disasters can be predicted, their environmental impact (and human toll) can be minimized. Climate informatics is a relatively new scientific field uses AI to comb through years of historical data from earthquakes, storms and hurricanes to look for patterns and warnings.
Google is using AI to create better forecasting models that’ll more accurately predict when and where floods are going to happen. Inputting data such as historical events, river level readings, terrain types and elevational levels are all fed into their neural network models, which then run hundreds of simulations of potential future river flooding events in each location. Microsoft has gone one step further and developed a land cover mapping tool that you can try for yourself.
The vast plains of sub-Saharan Africa make it easy for motivated poachers to evade park rangers; sadly, poachers kill an estimated 55 elephants every day. Conservation nonprofit Resolve is combatting this with their TrailGuard AI, a new device which uses Intel computer vision chips to spot animals and humans that wander into view. The cameras will be placed on trails used by poachers and alert park rangers to any unusual activity.
Over in Africa’s Congo Basin, the Elephant Listening Project has installed audio sensors and recorders over 580 square miles to “eavesdrop” on forest elephants who are hard to spot from satellites. To analyze its data faster, ELP has partnered with Conservation Metrics, which uses neural network models to listen for elephant calls among the forest noise, as well as quickly detect threats (such as gunshots) and find the poachers before they can do harm.