A(I) force for good?
Scott Hosking, environmental data scientist at the Alan Turing Institute and British Antarctic Survey, tells Chris Seekings how artificial intelligence is helping governments understand and predict the impacts of climate change
From the automation of jobs and the spread of fake news, to the manipulation of elections and erosion of privacy rights, the threats posed by artificial intelligence (AI) are enough to keep anyone awake at night. The technology is also transforming our society in amazing ways, bringing an end to mundane work, giving rise to life-saving medical advances, and helping to solve the world’s most pressing challenges, including climate change.
“If we can harness AI as a force for good, it could be transformational in how we reduce our carbon emissions, mitigate and adapt to climate change, support those most in need, and protect biodiversity,” explains Scott Hosking, an environmental data scientist and AI expert at the Alan Turing Institute and the British Antarctic Survey.
“This is a pivotal moment with AI – often seen as the fourth industrial revolution – and we’ve really got to make sure it supports our civilisation as we go into future decades.”
Digital twins
Hosking’s area of expertise is monitoring and predicting environmental changes, which he does as head of the British Antarctic Survey’s AI Lab, and co-director for the Turing Research and Innovation Cluster in Digital Twins, leading a team of scientists and engineers.
‘Digital twins’ are simulations, or copies, of what the physical world looks like, which, using AI and machine learning, can be manipulated to test different future scenarios and forecast changes. “We’re building a digital twin of Antarctica to understand global sea level rise,” Hosking explains. “Antarctica has 60 metres of sea level rise locked up in ice, and although it’s not all going to melt at once, even if a small part were to melt or break away and collapse, that could have significant implications for global coastal infrastructure and communities.”
Modelling something as complex as environmental changes in Antarctica requires a huge number of data inputs and considerations. Radars, satellites, underwater vehicles, ocean floats, aircraft, drones and ground sensors all collect information to help create a digital version of the physical world.
“They all have their strengths and weaknesses, so we need to bring those different datasets together, and fill in gaps using AI and machine learning approaches where possible,” Hosking says.
“With a digital twin we can ask what would happen if a piece of ice sheet breaks off? How would ice flow faster to the ocean? What would that mean for the global sea level in a few years’ or decades’ time?”
He tells me how, in January this year, a huge part of ice broke away from Antarctica, and that it took around eight months to identify the resulting increase in ice flow. “With a digital twin, you’d get that update in a much speedier manner, which can also act as an early warning system for decision-makers and governments. It can be a hugely powerful and important tool.”
A glimpse of the future
When it comes to predicting sea level rise, the Antarctic Ice Sheet contribution remains the biggest uncertainty, and its deterioration is considered a climate system ‘tipping point’. Under a worst-case scenario, the West Antarctic Ice Sheet is lost by 2300, sea levels rise by 16 metres, over a billion people face coastal flooding and emperor penguins are long extinct.
“There are also climate model scenarios that show ice-free summers in the Arctic by the middle of this century, so understanding that process is important for supporting indigenous communities that rely on sea ice for hunting, fishing and migration, and for supporting wildlife like polar bears and caribou,” Hosking explains.
However, there are more positive scenarios where we reach net zero by 2050, the climate stabilises during this century, extreme weather events stop becoming more frequent, and summer sea ice extent remains in the Arctic. “There is still a great deal of uncertainty, which is why bringing together the different sensors and data that we have is so important for tightening those forecasts. If we can close that uncertainty gap, we can make better decisions in the future with regards to coastal infrastructure to ensure we don’t over-engineer solutions or under-engineer coastal defence systems to cope with sea level rise.”
Harnessing AI and vast quantities of data to predict climate changes and the numerous additional uses is extremely carbon intensive. Researchers at the University of Massachusetts, Amherst, have found that training a single AI model can emit more than 283,949kg of CO2, which is around the same amount of greenhouse gas that would be emitted by 62 petrol-powered passenger vehicles driven for a whole year.
“As environmental scientists, that’s something we need to keep a very close eye on. But by bringing scientific and machine learning expertise close together, you can reduce the computational and carbon burn from models,” Hosking says. “We can design models that make efficient use of data and don’t have to be so big and complex.”
Sustainable models
Machine learning is often seen as a ‘black box’ process, which produces useful information without revealing the internal workings and how that information is calculated. “We’ve done a lot of work to understand how the models actually work and to untangle what we are doing to create simpler processes – there’s no point over-engineering a model when a simpler model will do it and will also reduce carbon and improve the interpretability.”
However, a lack of access to cross-disciplinary expertise remains a significant barrier to streamlining models and harnessing the true potential of AI. “There are great AI and machine learning experts who would struggle to understand the application in certain areas, whether that’s health, the environment or astrophysics,” Hosking explains. “There is a communication barrier, and the only way to get over that is to have deep conversations, ask lots of questions and get to know each other, and part of my role is bringing these groups together.”
He says that virtual and augmented reality are also going to be an important communication tool and will help stakeholders visualise the changes in the environment. “Environmental sustainability requires us to think broadly, including social sciences, environmental sciences, infrastructure for net-zero cities etc., and we really need to find a better way to explore the breadth and complexity that lies across the environment and sustainability space.”
This raises the question of how far AI can be used as a tool for tackling some of the broader social challenges we face. Could it be used for determining government budgets for health services and house building, for example? Could it be used for identifying corruption, or potentially even replace politicians?
“I see AI and machine learning as tools that can support decision-making, but the ultimate decisions should be made by people,” Hosking reasons. “However, there are a lot of processes that go on in governments, research labs and industries that are repetitive and time-consuming, which can be automated. The more we can use AI to do that, the more we can get our politicians, business leaders and scientists to focus on what they are trained to do.”
Safeguards and ethics
Earlier this year, more than 1,000 tech leaders, researchers and others, including Elon Musk and Steve Wozniak, signed an open letter warning that AI tools present “profound risks to society and humanity”, arguing that developers are “locked in an out-of-control race to develop and deploy ever more powerful digital minds that no one – not even their creators – can understand, predict or reliably control.”
Last month, 28 governments agreed to work together to combat these risks after signing up to the Bletchley declaration at the AI Safety Summit in the UK.
Hosking says it is important to put safeguards in place and explains how specialist groups are looking at the ethics of AI.
“We are seeing a J-curve in the capabilities of AI, and huge acceleration with this technology month by month, and so we’ve got to make sure that the whole of society benefits and it doesn’t just benefit the top 1%.”
To democratise AI and digital twinning technology, his team are developing open-source toolkits to make sure they can be adopted by the wider research community and other stakeholders.
“AI can be used to generate great pictures and content, but we need to make sure we have factchecking in place, and that governments and industry step up to ensure they are transparent about where that content comes from, otherwise it becomes untrustworthy and the benefits are lost,” he explains. “AI can be a gamechanger as a force for good, but without the necessary safeguards, governments could leave it in the hands of a few tech companies, and that is just too dangerous.”