It might not be the silver bullet, but it could help fire the gun.
Guest Piece by Manon Theodoly
It’s time to take a new approach. Our earth is heating at an alarming rate, with no sign of slowing down. The past 22 years have been the warmest on record and current efforts to save the environment are not enough. Microsoft has taken an innovative perspective on the issue, funding a $50 million cross-company effort delivering technology-enabled solutions to environmental challenges. Microsoft’s AI for Earth, founded in 2017, is laying the foundations for the world’s first planetary computer, placing global scale data at our fingertips in an effort to solve the crisis. For the first time, we can understand how fast our resources are being depleted and what we should do to stop this. The next step is to raise awareness about the potential of AI to boost funding and support. Therefore, we must be informed about current projects, future hopes, and potential areas of concern that require further research.
What Is AI and How Does It Work?
AI, or artificial intelligence, is essentially an algorithm that solves problems. The simplest form is narrow AI, which is task-oriented and able to do some things better than humans, such as recognizing speech or images or forecasting weather. Machine learning is also rule-based but is more complex since its algorithms can learn from data. The more data the system analyses, the more accurate it becomes as the software evolves to achieve its goal. Deep learning is the newest development. It involves neural networks made up of multiple layers, similar to the structure of a brain. Its algorithms are self-learning, giving it the capacity for memory. For example, deep learning AI could figure out how to identify a cat without human input about cat features, after seeing millions of random images from the internet.
How Is It Currently Being Used?
Microsoft is the global leader in this field. While the project is only in its beginning stages, its impact is already significant. The company has partnered with a variety of NGOs, offering them research grants, with the eventual goal of working with governments to employ the technology on a global scale.
For example, Microsoft has developed software to prevent poaching and illegal fishing. Often, organisations do not have the capacity to monitor large expanses of land, so many illegal practices go undetected. Through a partnership with Microsoft, OceanMind uses AI to prevent illegal fishing. By analysing vessel movements in real-time, using algorithms to identify suspicious behaviour, it can direct patrol boats effectively. PAWS also works with Microsoft, employing machine learning to predict where poaching may occur. Microsoft’s grant will help the organisation to incorporate real-time data, analysing poacher’s behaviour from crime data to improve patrols. Machine learning can analyse data with speed and accuracy previously unimaginable, producing predictions which will significantly reduce harmful and illegal practices.
Microsoft’s technology has also been improving environmental health. In order to find solutions, one must understand the problem. When it comes to the environment, this requires data on many different variables. Due to the global nature of the problem, vast quantities of manual labour have been required to analyse these data sets. This means that not only are the results often inaccurate, but the analysis is so time-consuming that the environment has often changed before the project can be finished. AI enables earth-scaled data collection in real-time, turning information into insight to predict environmental threats and inform policy. Microsoft premonition enables biodiversity and pathogen monitoring via robotic insect capture and genomic analysis with AI. This not only offers insight into the health of the ecosystem but could also mean a reduction in future epidemics.
Responding to the need to increase the world’s food production, with limited additional arable land and receding water levels, Microsoft developed Farm Beats. This allows farmers to collect data via wifi on their smartphones, using low-cost sensors, drones, and machine learning algorithms powered by solar panels. For the Dancing Crow Farm, this technology resulted in 33% less water and 44% less lime used.
Additionally, Microsoft’s land cover mapping technology has changed the way organisations monitor, model, and manage the earth’s natural systems. Land cover mapping is the foundation of effective conservation and sustainable growth. By putting algorithms in place to keep pace with the increasing speed of data collection, more accurate maps of environmental health can be built. It took Chesapeake Bay Conservancy 1 million dollars and over a year to create a map of one of the most important watersheds in the US. Through a partnership with Microsoft AI, they were able to build a larger map for a fraction of the cost, in a fraction of the time.
Microsoft AI is also helping accelerate the protection of biodiversity. Wildlife photos and species labelling are central to conservation efforts, as policymakers can only act if robust up-to-date data is available. However, human labelling has a limited capacity. A camera trap project might deploy up to 1000 cameras for several months, leaving researchers with millions of images to sort through. Machine learning could completely automate this workflow, answering questions such as which species need the most urgent protection. Through a partnership with Microsoft, Inaturalist encourages citizen scientists to share wildlife photos in the cloud. Scientists can access this data to understand when and where organisms occur, paving the way for smarter conservation action.
How Could AI Be Used in the Future to Address the Most Polluting Sectors?
“We’re facing our last real opportunity to ask ourselves the question: how are we going to solve these planetary scale questions?” — Lucas Joppa, Microsoft’s chief environmental officer
While we understand the technology, we have yet to fully exploit its potential. Microsoft works on AI tech that is specific to environmental issues, however, AI should also be addressing the most polluting sectors. Acknowledging this problem, Microsoft released a report in 2019, summarised below, which promises breakthroughs in machine learning in the coming years.
As shown in the graph, electricity is one of the most worrying sources of CO2. Electricity grids are a delicate balancing act. Equal amounts of electricity fed into the grid must be consumed from the grid, which is particularly hard to achieve for renewable sources whose output varies based on weather. Algorithms predicting demand could automatically manage energy supply, improving efficiency.
AI could also reduce the carbon outputs from buildings, supporting building managers, and policymakers constructing buildings with low energy consumption. AI could also track sources of air pollution to produce potential strategies for dealing with it. For instance, Microsoft looks towards the eventual development of smart buildings. These are intelligent control systems that regulate the amount of energy consumed, such as predicting when lighting will be used and turning it on and off accordingly.
Globally, the transportation sector accounts for about a quarter of energy-related CO2 emissions. Machine learning could reduce transport activity by predicting the most efficient routes, improve vehicle efficiency by developing new materials, discover fuel alternatives and help the shift towards lower-carbon options by determining areas within a city where alternative options, like rail, would be utilised. AI could also promote freight consolidation, bundling shipments together, and reducing the number of trips so that trucks don’t have to make return trips empty.
Furthermore, AI has a lot of potential to reduce emissions in the industrial sector, which spends billions of dollars annually gathering data on factories and supply chains. AI could 3D print cleaner materials, direct buyers towards low-GHG options, and optimize factories for renewables. AI would have the most impact in the distribution sector. In 2011 global excess inventory amounted to about $8 trillion USD worth of goods. Optimising shipping routes, preventing overstocking, and reducing the transport of perishables would reduce the energy wasted from inefficient practices.
AI could also speed up reforestation efforts. Software such as Airlitix, currently used to measure greenhouse gas emissions, could be used to control the health of national forests. With humans needing to plant over 1.2 trillion trees to combat climate change, we should consider automating this process. Instead of taking the time to tend to national parks, the Airlitix software could be built so that drones could plant trees, release plant nutrients, or even deter forest arsonists.
What Are AI’s Shortcomings?
Due to its current successes, it’s likely that there will be greater use of AI to fight climate change in the future. As this sector grows in scale, it is important to recognise the areas requiring further research. AI is a powerful tool, so understanding its shortcomings is essential to using it safely.
Deep learning can be risky when applied to early warning systems for natural disasters where certainty is needed. Its black box conclusions mean its processes are not understandable to humans, making it impossible to determine their accuracy. From an economic perspective, companies that are slower to adopt AI may suffer economic consequences as their AI-based competitors advance. This may inadvertently leave developing economies behind. This is particularly significant in relation to farming, as farmers with no internet access would be at a disadvantage.
Scientists will have many obstacles to confront when AI is applied to real-world situations. Real-world data is often messy and/or private; a form that’s difficult for AI algorithms to use. Most worryingly, AI can itself have a large environmental footprint, as data centres require a huge amount of energy and water to run. A recent article uncovered that Google’s data centres were using billions of gallons of water. Unless alternative means of running these centres can be found, AI threatens to put even greater pressure on our global water supply.
More broadly, we are still far from constructing computers that are able to reason, abstract, understand, and communicate like the human brain. It takes 83,000 processors 40 minutes to compute what one percent of the human brain can calculate in one second. Also, a shift to AI poses security risks, as it can be hacked, interfering with transportation, energy, and other crucial sectors. AI also carries a social risk, as more automation would eliminate jobs and exacerbate existing data privacy breaches. The most important consideration will be preventing data biases. AI tools are only as good as the quality of the data they use and the socio-political systems in which they are implemented. If racially biased data is given, the algorithm will pick up on these biases, replicating and amplifying them.
When applied to real-world situations, AI is not without challenges. But given the gravity of the climate situation, these are risks we should be willing to undertake.
“Worrying about sentient AI as the ice caps melt is like standing on the tracks as the train rushes in, worrying about being hit by lightning” — Bret Victor — computer scientist and researcher at Dynamicland
The issue is that many world leaders simply don’t take the problem seriously enough. Scientists and the public have rallied around environmental policy for years, but global governments have yet to enact adequate policies. The US is the second leading emitter of greenhouse gases behind China and the leading emitter per capita. And yet, Trump has confirmed plans to pull the U.S out of the Paris Agreement which would have held them to reducing emissions by 26–28% by 2025. China continues to build coal-fired power stations, while the UK is reducing its targets for cutting emissions. To make matters worse, there is little cooperation between governments and organisations, leaving NGOs struggling to have influence. In order to find solutions, we must understand the problem. Understanding requires vast quantities of data, a daunting task for smaller organisations, preventing them from scaling solutions, and amplifying impact.
More interaction among public and private entities, technologists, and policymakers is needed to avert these potential risks. Interdisciplinary cooperation is crucial to driving the parallel development of AI and climate change-specific solutions. Society is geared towards short term results and profit, so there is a large funding gap for longer-term technologies. Investors should help with the deployment and scaling up of projects, while local and national governments should release data that might be relevant to climate change. With more researchers, technology companies, and partner companies on board, the technology could be refined in the coming years. Computer scientists know a great deal about technology but little about the environment. Environmental organisations know little about technology but a huge amount about the environment. New challenges call for new perspectives, namely the cooperation between humanity and technology.
It would be a misconception to say that AI is a silver bullet. Microsoft’s term “planetary computer” does not mean they’ve created a crystal ball that will provide us with one clear solution. Rather, it means the connection of trillions of data points to a machine capable of converting this into contextualised information. This would help us understand the functioning of our earth so we can find solutions to help it. Microsoft has begun to lead the way, targeting sectors such as farming, mapping, and photography which could help lower our global carbon footprint. The company hopes to continue scaling up this project, focusing on larger and more daunting sources of emissions. This will be impossible without the support of governments, industries, and investors. It’s important for all sectors to act quickly because when it comes to climate change, time is not on our side.