We’re living through an era of radical uncertainty. Climate change isn’t simply one risk among many, it’s a risk multiplier. Its impacts don’t happen in isolation. They ripple across interconnected systems, affecting food and water, health, infrastructure, supply chains and the economy. One disruption can trigger another, creating cascading risks that are increasingly difficult to understand, predict, and manage. 

At OnePlanet, we believe technology can be an important enabler in navigating this complexity, helping governments, organisations and communities better understand these connections, anticipate risks and respond before they grow into much bigger crises. And this is where we see real potential for AI as a tool for good. Not as the answer to everything, but as a tool that can help us make sense of some of the complex challenges humanity is facing. 

We’re already seeing examples of this in practice. In England, Natural England’s AI4Peat project is using AI alongside aerial photography and LiDAR data to map damaged peatlands. The technology can identify drainage channels, erosion and areas where restoration has already taken place, giving land managers a much clearer picture of where action is needed. It’s a brilliant example of AI being used not simply to automate a task, but to better understand a complex landscape and help people make more informed decisions about where to focus nature restoration efforts. 

At OnePlanet, we’re exploring this from another angle too: how can we use AI while reducing the amount of processing it needs to do? Instead of asking an AI model to read an entire plan or strategy PDF every time someone asks a question, OnePlanet structures plans into Outcomes, Actions and Indicators first. This means the AI can find and process the pieces of information that are actually relevant to the question, rather than repeatedly processing the whole document. 

Our early testing shows just how significant that difference can be. In one test involving 25 questions, the structured OnePlanet approach used 22,105 tokens, compared with 318,166 tokens when querying the full PDF, around 14 times fewer tokens. When working across two larger plans, the difference was even more striking: 2,059 tokens compared with 152,154, around 74 times fewer.  When analysing complex sets of plans, say for a geographic area, we think we’ll use around one thousandth (0.1%) of the AI load. 

This helps to illustrate an important principle: if we can get useful answers while asking AI to process far less information, we can use the technology more efficiently. 

But perhaps the bigger question isn’t simply how we use AI. It’s what do we actually want to use it for? 

If we can better understand the connections between action plans, spot opportunities for collaboration and identify where actions might conflict, we can start making more joined-up decisions. But that still leaves us with a much bigger question: 

AI can help us see connections and possibilities, but it’s up to us to imagine the future we want and work together to make it happen.