Scientists set out roadmap for the next generation of AI in ecology and conservation

An international team of interdisciplinary researchers, brought together through a Schmidt Sciences-funded workshop at the University of Oxford, has outlined how a new generation of AI models could transform our understanding of the natural world.

An international team of scientists, including researchers from the University of Oxford, has published a new vision for how artificial intelligence (AI) could help scientists understand ecosystems and predict their responses to environmental change.

The Perspective article sets out a roadmap for developing AI systems capable of learning from the vast and varied information collected about the natural world.

The paper emerged from Shaping the Future of AI for Conservation, a four-day workshop held at the University of Oxford in September 2025. Funded by Schmidt Sciences, the workshop was co-organised by Dr Amy Hinsley and Dr Rachel Parkinson, both Schmidt AI in Science Fellows at Oxford Biology at the time. It brought together scientists and conservation practitioners from across different disciplines to explore how advances in AI could help address some of the most pressing challenges in ecology and biodiversity conservation.

The resulting collaboration includes several Schmidt AI in Science Fellows from Oxford and other international institutions, alongside researchers specialising in ecology, conservation science and AI.

Bringing together the pieces of ecological knowledge

Scientists have accumulated extraordinary amounts of information about the natural world, from genetic sequences and satellite imagery to wildlife recordings, species observations and environmental measurements. But these data come in vastly different forms, at different scales, and describe different aspects of complex ecological systems.

The researchers propose developing ecological foundation models (ecoFMs): AI systems that could learn from these different sources of information, uncover relationships between organisms and their environments, and apply that knowledge to a wide range of scientific questions.

Foundation models underpin many of today's most advanced AI technologies, including large language models such as ChatGPT. While similar approaches have made major advances possible in fields such as weather forecasting and molecular biology, their potential in ecology remains largely untapped.

Existing AI tools can already identify animals in photographs, recognise wildlife sounds and predict where species might occur. Ecological foundation models would go further, aiming to learn how species interact, how ecosystems are organised and how environmental changes affect the relationships between organisms.

Dr Rachel Parkinson, corresponding author of the paper, who co-organised the Oxford workshop while a Schmidt AI in Science Fellow and is now Lecturer in Computational Ecotoxicology and Neuroethology at Queen Mary University of London, said:

"We have an extraordinary amount of information about the natural world, but bringing these different types of data together in ways that capture meaningful ecological relationships is incredibly difficult.

Our vision is to develop AI models that can learn from these diverse sources of information, uncover connections that might otherwise be missed, and ultimately help us better understand and predict how ecological systems respond to change."

From scientific vision to practical roadmap

The proposed models could eventually help researchers identify previously undocumented relationships between species, fill gaps in biodiversity knowledge and forecast how ecosystems might respond to climate change, habitat loss and other environmental pressures.

For example, a model could combine information about plants and pollinators, including their genetic relationships, physical characteristics, geographic distributions and known interactions, to predict which flowers an understudied pollinator might visit.

The Perspective does not present a completed AI model. Instead, it identifies the scientific and technical challenges that must be overcome, including how to integrate different types of ecological data, incorporate established ecological theory, evaluate predictions and account for uncertainty.

The authors also emphasise the importance of responsible development, recognising that ecological datasets are often incomplete and biased towards well-studied species and regions. Making these systems useful for conservation will require close collaboration with the researchers, practitioners and communities that generate and use ecological knowledge.

Dr Amy Hinsley, Schmidt AI in Science Fellow at the University of Oxford and co-organiser of the workshop, said:

"One of the aims of the workshop was to bring together people from around the world working across research, policy and practice, with expertise spanning ecology, conservation and AI. We wanted to create opportunities to exchange ideas and explore some of the bigger challenges that no single discipline could address alone.

What emerged was a shared ambition to think beyond individual AI applications and consider what it would take to develop tools that could help us understand ecological systems more broadly. This paper is one outcome of those discussions, alongside work exploring the role of social science data in AI and conservation, and the ethical questions that come with using these technologies.”

The international collaboration highlights the value of bringing together expertise across disciplines and institutions to tackle scientific challenges that require new approaches.

Rachel added:

"Ecology poses some genuinely difficult challenges for AI because ecosystems are dynamic, interconnected and highly dependent on context. Developing models that can cope with this complexity will require advances in AI that are grounded in ecological knowledge and theory.

We hope this roadmap will encourage collaboration between ecologists, conservation scientists and AI researchers to develop technologies that are not only scientifically meaningful, but also transparent, reliable and developed responsibly."

To read more about this research, published in Nature Machine Intelligence, visit: https://www.nature.com/articles/s42256-026-01317-6