Future urban planners will be able to test the effects of new traffic patterns on air quality, carbon dioxide emissions and residents' wellbeing before a single block of concrete has been lifted.
Could AI slow down climate change? To anyone who has been following recent discussions, the idea sounds strange. The water and energy consumed by generative AI would seem more likely to exacerbate than alleviate climate problems.
Yet AI may also have a part to play in building a more sustainable future. Professor of Computer Science Laura Ruotsalainen of the University of Helsinki is working with her research group to develop machine-learning methods that could help find practical solutions for building better cities.
Simulations are used to investigate how traffic can flow smoothly and people reach their destinations with ease, while air quality is preserved, emissions are kept in check, and cities remain agreeable places to live.
"Instead of having to conduct decades of observations, AI can help us experiment with solutions and obtain rapid answers," says Ruotsalainen.
"We can hardly build a new city from scratch, observe its impact on the climate and people's wellbeing, and then conclude that we got it wrong. With AI, results can be obtained in just a few weeks without any harm to the climate or people."
From seconds to decades
Urban planning involves juggling a number of competing goals and timescales. This poses a distinctive challenge for researchers, who must weigh everything from the emissions impact of an accelerating car in seconds and ten-minute differences in air quality to the habitability of cities over years.
On the one hand, people should reach their destinations with ease; on the other, their journeys should generate the fewest possible emissions.
The answer to this complex puzzle is being sought through reinforcement learning. It is a machine-learning method that enables AI to manage and weigh up several competing goals simultaneously. AI tries out a range of methods to establish which goals can be achieved and by what means. This is known as multi-objective optimisation.
An AI agent is created to perform various functions. It receives rewards or penalties depending on its performance.
"Imagine an agent driving a car through a city. When it makes a decision, such as taking a turn at an intersection, we examine how that action affects our goals. Here, the goals are to keep air quality as high as possible and to reach the destination quickly. If the agent's choices bring us closer to our goals, it is rewarded. If its choices move us further from our goals, it is penalised.
To date, reinforcement learning has been used mainly in games and simple applications, such as traffic light control. Large-scale, systemic changes have so far proved beyond its capabilities.
An urban environment involves a vast number of interacting effects. AI can help address such complex problems and the combined effects of multiple factors. This is work that humans cannot perform unaided.
Methods for tackling major challenges
Ruotsalainen argues that the real benefits of AI derive from its targeted application. When AI is developed to address a specific, well-defined problem, the breakthroughs can be significant.
"Public conversation has been dominated by ChatGPT and the generation of text and caricature images. This is far too limited a view of what AI can offer."
In medicine, AI is opening up new possibilities in treating serious diseases such as cancer. As far back as the 1990s, fusion power was seen as the answer to the world's energy needs; with AI, that prospect may finally be within reach.
As a researcher, Ruotsalainen dreams of developing machine-learning methods capable of tackling major social and societal problems. Optimising complex systems involved in areas such as urban planning allows us to identify small changes with far-reaching benefits, including in terms of urban ecological sustainability.
"I pursue this work with a clear conscience. I believe the benefits that this development will bring outweigh the disadvantages of the resources AI requires. Our research also addresses the growing energy consumption associated with computing in machine learning. AI can be a tremendous force for good, with the potential to improve lives and help address climate change."
