AI is a computing tool. It can process and interrogate huge amounts of data, expand human creativity, generate new insights faster and help guide important decisions. It’s trained on human expertise, and in conservation that’s informed by interactions with local communities or governments – people whose needs must be taken into account in the solutions. How do we ensure this happens?

Last year, Reynolds joined 26 other conservation scientists and AI experts in an ‘Horizon Scan’ - an approach pioneered by Professor Bill Sutherland in the Department of Zoology - to think about the ways AI could revolutionise the success of global biodiversity conservation. The international panel agreed on a top 21 ideas, chosen from a longlist of 104, which are now published in the journal Trends in Ecology and Evolution.

Some of the ideas extrapolate from AI tools many of us are familiar with, like phone apps that identify plants from photos, or birds from sound recordings. Being able to identify all the species in an ecosystem in real time, over long timescales, would enable a huge advance in understanding ecosystems and species distributions.

Another idea proposes using fibre optic cables, which are already spread across the world’s ocean floors, to map and monitor chronically understudied marine species like whales and seals – by interpreting disturbances in cable signals caused by the sounds these species make.

But for every positive idea, there’s a negative to consider too.

The most obvious issue is that AI’s high demand for energy and physical infrastructure could directly undermine conservation goals. It works on huge amounts of data, which requires supercomputers, technical skills, and a constant supply of electricity - all more likely to be available in the Global North.

“We need to prevent the concentration of power in conservation. People in the poorer Global South, where many of the planet’s important ecosystems are located, are most likely to be affected by conservation but they’re also least likely to have access to AI infrastructure,” says Reynolds. “It isn’t equitable to take information from studies in the Global South, use it to develop and train our AI models, and then dictate how land and resources are managed there.”

The researchers are also concerned about funders prioritising conservation projects involving AI, due to their novelty. “That will be a surefire way to prioritise projects led by elite institutions in the Global North, steering resources away from fundamental work that already struggles to attract funding,” says Reynolds.

But the biggest impact of AI on nature might come from outside the field of conservation entirely.

Sandbrook's new paper, published in the journal Conservation Letters, takes a critical look ‘beyond the hype’ of AI.

Impacts could be positive, for example by transforming agriculture - currently the single largest cause of biodiversity loss. AI-enabled precision systems could help farmers grow more food in a smaller area, with fewer resources, which could massively reduce the global land area needed for agriculture.

But other advances could damage nature, like AI’s potential to generate better medicines that extend human life expectancy. Sandbrook says: “This is most likely to happen in wealthier countries like the UK and US, where each person has a high environmental footprint. The price of longer lives could be greater biodiversity loss.”

Like Reynolds, Sandbrook is concerned about how AI might cause power and wealth to become concentrated – but he’s thinking in terms of the tech companies: giants that can lobby to stop governments regulating against their environment-harming activities.

“I want to urge policy makers, and CEOs of conservation organisations, to think really carefully about how the widespread adoption of AI in society could impact the natural world - even for uses of AI that on the face of it have nothing to do with conservation,” says Sandbrook.

He adds: “Managing and regulating AI to ensure global biodiversity is protected is an ambitious goal, but it’s vital to make this part of the conversation. At the moment many people in the AI tech sector, and in government, just aren’t thinking about how their work might affect nature at all – they haven’t even noticed that it’s relevant.”

Meanwhile, the Horizon Scan ideas are catalysing discussions that are vital to ensure AI boosts, rather than damages, the impact of conservation work.

Reynolds has received invitations to speak to staff at organisations including the Wildlife Trusts, the Chartered Institute of Ecology and Environmental Management (CIEEM), and BP, all keen to know about the latest developments in AI for conservation.

He’s helping others discuss how they can embrace AI to advance their conservation work, and how it might affect conservation more broadly. For example, AI could help consultants make better environmental management plans, or help planners make better decisions about new developments to take biodiversity into account.

And in response to the discussions, he has published a Letter in the journal Trends in Ecology and Evolution, outlining ways that people involved in conservation can help support the equitable deployment of AI.

Reynolds is also helping people navigate what he calls “the Wild West of AI solutions” - the ever-growing number of AI tools being sold to organisations who don’t fully understand what they’re buying.

He says his involvement in creating an AI ‘Conservation Co-Pilot’ has given him a valuable insight into how AI models work. This tool will help guide decision-makers towards the best way to manage specific land types to conserve particular species. It’s based on a database of effective conservation actions that has taken many experts 20 years to pull together.

“We can use what we’ve learned in creating an AI system ourselves to help others be careful about which AI tools they use, and to understand their limitations. At the moment that’s a huge gap in the conversation,” he says.

“People may believe AI tools will be the answer to all their problems, but it’s important to understand that they’re only as good as the data they’ve been trained on. In most cases that comes under very little scrutiny – that’s not because people don’t care, but because they don’t know to check, and in many cases - GPT 4.0 for example - it’s impossible to find out.”

Both Sandbrook and Reynolds are clear: people should not be lulled into thinking there’s no need to worry because AI will solve everything without our help.

“AI will have enormous benefits for conservation. It’s extremely exciting and will allow us to do so many things we weren’t able to do before,” says Reynolds. “But we need to embrace it cautiously, because it also introduces a whole range of new challenges that need to be thought about carefully, and overcome.”