The development of artificial intelligence (AI) is accelerating with an increasing impact on society. Although industrialization can be identified as the root cause of many environmental problems, the current “Fourth Industrial Revolution”, with rapid innovations and new connected technologies, provides opportunities for societal shifts that can help address Earth’s environmental challenges and change how we manage our environment.
What is the Fourth Industrial Revolution?
Figure 1 
The “Fourth Industrial Revolution” was the official theme of the 2016 World Economic Forum’s (WEF) annual conference. They defined this as the “fusion of technologies that is blurring the lines between physical, digital and biological spheres”. This revolution distinguishes itself from the previous ones by its velocity, scope and impact. We have seen immense progress in AI in the past few years, mostly because of exponential increase in computing power and a huge increase in data volume. Therefore, developments in AI are now truly revolutionary to society.
How can AI contribute to environmental challenges?
There is no short answer to this question, because AI shows promise in many areas. You may think that AI is just about making robots, but data is literally everywhere. There’s an overwhelming amount of climate data from satellites, environmental sensors and climate-model simulations and energy data like energy use data, GIS data, weather data, social media data, and electric vehicle data.
In a collaboration between the WEF and PwC, a report was published, where they identified AI applications in the following challenge areas: climate change, biodiversity & conversation, health oceans, water security, clean air, and weather & disaster resilience. Many large companies like Google and Microsoft are developing environmentally friendly AI systems. As you can see in figure 2, we have many developments to come from AI that will change how we deal with environmental problems.
Figure 2: timeline AI applications 
An overview of all opportunities goes way beyond the scope of this post, so I will just give two examples of fields where AI has a large potential impact: climate forecasts and energy grids. These are some of the first upcoming developments, so I will explain how these are changing or will change the current system.
Climate models and forecasts
A news article published on the Nature website talks about how machine learning could improve climate forecasts. Since AI systems improve when they get more information, climate science can benefit a lot from combining with AI, since one run of a climate model can produce 1 petabyte of data. Conventionally, climate models are based on rules entered into the model by climate scientists. Deep learning models, a specific type of AI that is based on neural networks in the human brain, “learn” the rules by analysing large amounts of data. Deep learning was first used in 2016 when researchers used it to predict tropical cyclones, atmospheric rivers and weather fronts.
According to Claire Monteleoni, climate is now a data problem. The collaboration between climate scientists and data scientists has caused the emergence of a new discipline: climate informatics. For example, the IPCC reports are based on the ensemble of many conventional climate models that have high variance in predictions. On average, combining these models results in better predictions, but it has always been a challenge to determine the most accurate way to do this. AI now aims to solve this problem by helping climate scientists rank climate models.
In practice, these developments (will) lead to better forecasting of extreme weather events like hurricanes and a better understanding of the effects of climate change, and it will be much cheaper and faster as well, since new machine learning methods are much better at analysing large amounts of data compared to current climate models which need a huge amount of computational power. The advantage lies not just in better predictions, but in the potential of revolutionary change in the system, because everything will be connected, so that risks can be monitored in real-time and dangers can be detected automatically.
Figure 3: squared error between prediction and (simulated) observations. Best expert is the best out of the 20 models (in hindsight). Learn-a is the model developed by Monteleoni that learns the best way to combine climate models 
Smart energy grids
As described in the previous blogpost, the transition to intermittent renewable energy sources poses threats to the electricity grid. Not only batteries, but also AI has the potential to contribute a lot to the stability of the grid, for example by using data to forecast energy demand and supply. Machine learning techniques are becoming better and better at forecasting. Balancing demand and supply in the grid can be done by demand-side management, which affects patterns and magnitude of consumption. Figure 4 shows how the demand curve can be altered.
Figure 4: types of demand-side management 
Demand-side management is just one example. The smart grid is considered a technological paradigm shift. Energy data from smart grids allows for automatic and real-time monitoring of consumption, early blackout warnings, automatic billing, detection of energy losses, fast disturbance detection, and intelligent and real-time energy planning and pricing. All this will come with connected home devices and smart meters as part of the 4th Industrial Revolution. Then the power grid can be controlled to deliver exactly the amount of energy needed so no energy is wasted, and renewable energy can be integrated.
Of course, there are also downsides to these technological developments. It is possible that AI developments could accelerate environmental degradation. Furthermore, methods like deep learning usually work like a “black box”: it is not intuitively clear what happens exactly inside the system, so climate scientists are still a bit reluctant to trust the results. For example, if AI is used to create an early-warning system for floods, and it’s too much of a mystery how the model gets these predictions, then the risk of false alarms may be perceived as too high.
I’m aware that this may all sound quite optimistic, but these are real developments that I’m pretty excited about. While these solutions will probably not solve the entire climate problem, at least it will make it easier to predict events and it will help limit impact. It is necessary to figure out how to deal with the risks related to AI, but we have plenty of positive consequences to look out for.