If the world’s natural resources are increasingly stressed and depleted, the silver lining may be that we’re becoming better equipped at tracking that destruction and potentially doing something about it. Cheap, widespread sensor networks, the internet of things, magnitude-improvements in computing power, open source algorithms–these all allow us to manage oceans and forests more effectively, if we want the opportunity. Artificial intelligence systems that can sense, think, learn, and act on their own could allow a major upgrade in conservation efforts, in dealing with climate change, and living in a more energy-efficient manner.
A report released during the recent Davos World Economic Forum meeting laid more than 80 potential environmental applications for AI, ranging from the mundane to the futuristic. We spoke with Celine Herweijer, a partner at consultants PwC and one of the authors of the report. She argues that AI is now going mainstream: Algorithms and supercomputers that once were limited to specialist researchers at universities and government labs are now open to startups and everyday corporations. New ways of managing ecologically relevant systems are opening up as never before.
Autonomous energy and water networks
Solar, wind, and other renewables have the advantage of being carbon-free and ubiquitous. They can be situated in villages and towns and out-of-the-way places, bringing energy closer to everyone who needs it. The challenge is stitching these disparate sources together into a coherent, functional whole. That’s where autonomous systems come in. They can deal with the intermittency of renewables and react to the ebb and flow: when one source of power is coming online or going down, or when one user is ramping up demand and another is clocking off for the night. AI systems are flexible and they can do more work, and be in more places, than human grid managers.
“When you have a complex system with so many sources of renewables, you need them to talk to one another, so you can do storage and optimize the load,” Herweijer tells me. “That can’t happen without artificial intelligence enabling all these new sources to come together. They will enable these future systems where we have peer-to-peer energy trading and community exchange. They are what we need for a decentralized, autonomous grid.”
Similarly, AI will allow for a more decentralized water system, driven by sensors and new technologies like blockchain, Herweijer says. Smart contracts–legal arrangements automated with code–can enable swift trading of assets, including water rights. “Blockchain is vital for recording provenance, then you can have smart contracts and have people trading between parts of the decentralized network,” Herweijer says. “Utilities of the future, whether water or energy, will be more decentralized because that improves productivity.” The Department of Energy has some early-stage AI-based grid systems in development.
Opening up climate modeling
Modeling future weather events and climate patterns means processing complicated physical equations, like the fluid dynamics of the atmosphere and oceans. Climate scientists have relied on supercomputers, like the one at the Argonne National Laboratory, outside Chicago, to do their calculations. But there are only a few dozen true supercomputers around the world, meaning that access is limited: Many other scientific fields also require big computational capacity.
Deep-learning techniques, inspired by the way the human brain processes information, incorporate some of the complexity of the real world in climate modeling, allowing computers to run faster and do more calculations within a given period. “We’ll do simulations and modeling on home computers than we do now on supercomputers,” Herweijer says. “We can model small-scale features like wind storms that we struggled with in the past. Once you put AI in the system, you’ve got more people doing simulations and they’re doing it quickly. Forecasting of weather and climate impacts is going to get better rapidly over the next 10 years.”
Real-time data dashboards
Problems like illegal logging and illegal fishing require better monitoring systems. Data from satellites and unmanned underwater vessels can help bring greater visibility to such resources, but AI can help crunch the data to make it useful. New processing capabilities could provide close-to-real-time transparency by enabling authorities, and even the general public, to monitor fishing, shipping, ocean mining, and other activities,” the report says. “Vessel algorithmic patterns could identify illegal fishing, biological sensors could monitor the health of coral reefs, and ocean current patterns could improve weather forecasting.”
Global Forest Watch, a multi-group alliance convened by the World Resources Institute, uses satellite data to map illegal logging and offers a sort of early template for what Herweijer means. The Ocean Data Alliance is a similar public-private partnership for ocean monitoring involving groups like IBM and UC Santa Barbara’s Benioff Ocean Initiative. Its “approach could allow decision-makers to use machine learning to monitor, predict, and respond to changing conditions such as illegal fishing, a disease outbreak, or a coral-bleaching event,” the report says. Such systems need to involve industry to remain relevant, Herweijer says. For example, they can help companies prove they are abiding by commitments to avoid certain fish or trees.
Disaster resiliency and response
Decision-making in the wake of natural disasters is limited by the information available to government agencies and aid groups. It’s hampered by a lack of coordination. “Losses of life and property are multiplied when first responders can’t prioritize and target resources. Herweijer sees a role for automated systems that can analyze real-time data, like social media. “We don’t have a data-smart way of responding in real time to natural disasters,” she says. “We need public-private partnerships that bring together humanitarian agencies and big satellite companies to pinpoint where to start,” she says.
Emerging forms of AI don’t just crunch petabytes of “big data.” Techniques like “deep reinforcement” are self-learning and require little or no initial data; instead they learn, like a child, through trial and error and by being rewarded for success. “Deep reinforcement learning may one day be integrated into disaster simulations to determine optimal response strategies, similar to the way AI is currently being used to identify the best move in games like AlphaGo,” says Herweijer.
Earth Bank of Codes
The natural world contains reservoirs of innovative capacity that remain largely untapped. AI and systems analytics can help unbundle the biological and biomimetic possibilities. Scientists have begun work on the natural world equivalent of the Human Genome Project, with the aim of mapping the DNA sequences of all living things. The Amazon Third Way initiative, for instance, is developing a project called the Earth Bank of Codes, with two main intents. One, to open up potential discoveries, like blood pressure medicine derived from viper venom. And, two, to record the provenance of biological IP assets, so local people can benefit from follow-on discoveries.
“It’s not only about mapping genetic codes, but also how you change decisions around those codes. Tracking assets may be useful for a pharmaceutical company, but you are also starting to make sure that when a transaction happens, the value goes back to the community that grew the species,” says Herweijer.